Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Understanding geographic variation in health care costs in British Columbia Lavergne, Miriam Ruth 2015

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2015_may_lavergne_ruth.pdf [ 4.41MB ]
Metadata
JSON: 24-1.0167685.json
JSON-LD: 24-1.0167685-ld.json
RDF/XML (Pretty): 24-1.0167685-rdf.xml
RDF/JSON: 24-1.0167685-rdf.json
Turtle: 24-1.0167685-turtle.txt
N-Triples: 24-1.0167685-rdf-ntriples.txt
Original Record: 24-1.0167685-source.json
Full Text
24-1.0167685-fulltext.txt
Citation
24-1.0167685.ris

Full Text

   UNDERSTANDING GEOGRAPHIC VARIATION IN  HEALTH CARE COSTS IN BRITISH COLUMBIA by  MIRIAM RUTH LAVERGNE  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF  THE REQUIREMENTS FOR THE DEGREE OF  DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Population and Public Health)   THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)   February 2015  © Miriam Ruth Lavergne, 2015 ii Abstract  Amidst concerns about escalating health spending, examining variation in health care delivery may reveal opportunities for improved efficiency. Influential research in the United States (US) has concluded that health care spending and service use vary substantially from place to place, and this cannot be explained by differences in the health status of populations or by better outcomes of care in higher-spending regions. Whether similar patterns exist in Canada is not clear.  This thesis uses administrative health data to examine how and why health care use and costs vary within the Canadian province of British Columbia (BC). We developed networks of patients, physicians, and hospitals that correspond to actual service use, in order to ensure that observation of variation was not obscured by unit of analysis. We also identified areas of the province representing distinct health service environments, as an improvement over existing urban/rural classifications in understanding the role of geographic context. Access to individual-level data allowed more complete adjustment for population characteristics than is typically possible.  In contrast to the US, this thesis suggests that variation in costs of physician and hospital services in BC is largely explained by population health status. The very different environments for health services that exist among metropolitan, non-metropolitan, and remote regions of the province also explained some area-level variation. Despite modest variation in total costs, there are clear differences in patterns of service use across the province due to substitution between categories of care (such as inpatient and outpatient, or generalist and specialist services). Though differences in costs are modest, marked differences in health outcomes are evident, and require further scrutiny.  Results show there are no areas with systematically higher volume or more intensive service provision for populations with similar health needs. However, this does not mean that important variation does not exist and cannot be uncovered. It may be that examining variation at the level of provider, among population groups, or in treatment for specific diseases or conditions will yield more actionable results. Ultimately policy reforms aimed at system-wide quality and efficiency, rather than targeted at high-spending regions, will likely prove most promising.  iii Preface  This dissertation is my original, unpublished work. The analysis reported is covered by UBC Behavioural Research Ethics Board Certificate number H11-02945. I was the lead investigator for all analysis, and responsible for design of the research program, data analysis, and manuscript composition. McGrail K was the supervisory author on all research and was involved throughout study design and manuscript edits. Barer M, Law M, and Wong S provided input on analysis and contributed to manuscript revisions. The involvement of my committee members is reflected in the transition between active singular and active plural voice between chapters. While I led all of the research, the use of active plural voice in Chapter 4-7 reflects the contributions of my committee members to the analyses described.    iv Table of Contents Abstract .......................................................................................................................... ii Preface .......................................................................................................................... iii Table of Contents ......................................................................................................... iv List of Tables .............................................................................................................. viii List of Figures ............................................................................................................... x List of Abbreviations .................................................................................................... xi Acknowledgements .................................................................................................... xiii Chapter 1 - Introduction ................................................................................................ 1 Background .......................................................................................................................... 1 Thesis overview .................................................................................................................... 2 Rationale ............................................................................................................................... 3 History of “variations” research and relevance for contemporary policy issues .................... 3 Unresolved challenges ........................................................................................................ 4 How to interpret observed variation: which differences are warranted? ........................................... 4 Describing variation accurately ......................................................................................................... 5 Measuring outcomes ......................................................................................................................... 5 Contributions of this work ................................................................................................... 5 Chapter 2 - What causes geographic variation in health care use, costs, and outcomes? ..................................................................................................................... 9 Background .......................................................................................................................... 9 Effective, supply-sensitive, and preference-sensitive care .................................................. 9 Existing conceptual models of health care use ...................................................................10 Conceptual framework ........................................................................................................12 Overview and structure ......................................................................................................12 Population and level of analysis .........................................................................................12 Population factors ..............................................................................................................13 Need ................................................................................................................................................ 13 Resources ....................................................................................................................................... 13 Beliefs, preferences, culture and values ......................................................................................... 14 Health care system factors .................................................................................................14 Human resources ............................................................................................................................ 14 Material resources ........................................................................................................................... 14 Professional culture and practice style ........................................................................................... 14 Defining health care use, costs, and outcomes ..................................................................15 Health care use ............................................................................................................................... 15 Health care costs ............................................................................................................................ 15 Quality and appropriateness of care ............................................................................................... 16 Health outcomes ................................................................................................................16 Discussion ...........................................................................................................................16  v Chapter 3 - Geographic variation in Canadian health care ...................................... 21 Background .........................................................................................................................21 Methods ...............................................................................................................................22 Search strategy ..................................................................................................................22 Data extraction ...................................................................................................................23 Results .................................................................................................................................23 Study characteristics ..........................................................................................................24 Data sources and study populations ............................................................................................... 24 Health care outputs examined ........................................................................................................ 24 Methods used in identified studies .....................................................................................24 Geographic units and urban/rural definitions .................................................................................. 24 Methods to describe and quantify variation .................................................................................... 26 Controlling for relevant population characteristics .......................................................................... 27 Summary of observed variation by type of health care use and geographic unit ................27 Procedures and drugs ..................................................................................................................... 28 Service location or provider ............................................................................................................. 29 Quality/appropriateness of care ...................................................................................................... 32 Identified causes of variation ..............................................................................................33 Population causes examined .......................................................................................................... 33 Health system causes examined .................................................................................................... 34 Outcomes examined ....................................................................................................................... 36 Discussion ...........................................................................................................................37 Chapter 4 - Data sources and measures ................................................................... 45 Overview ..............................................................................................................................45 Data sources ........................................................................................................................45 Microdata accessed through Population Data BC ..............................................................45 Client registry (consolidation file) .................................................................................................... 45 MSP payment information file (physician payments) ...................................................................... 45 Discharge abstracts database (hospital separations) ..................................................................... 46 Vital Statistics death file .................................................................................................................. 46 Microdata access and linkage ......................................................................................................... 46 Publicly available data........................................................................................................46 Health System Matrix ...................................................................................................................... 47 BC Stats socio-economic profiles ................................................................................................... 47 Cost and outcome variables ...................................................................................... 47 Calculating cost of health care services .............................................................................47 Modeling costs ...................................................................................................................48 Measuring patient need or health status ............................................................................49 Choice of outcome measures .............................................................................................49 Chapter 5 - Defining multispecialty physician networks ......................................... 53 Background .........................................................................................................................53 Approach .............................................................................................................................55 Data sources and study population ....................................................................................55 MSP payment information file (physician payments) ...................................................................... 56  vi Discharge Abstracts Database (hospital separations) .................................................................... 56 Eligible residents ............................................................................................................................. 56 Eligible physicians ........................................................................................................................... 56 Eligible hospitals ............................................................................................................................. 56 Process for developing hospital networks ..........................................................................57 Linkage of residents to usual providers of primary care ................................................................. 57 Linkage of admitted residents to hospitals ...................................................................................... 57 Linkage of physicians to hospitals................................................................................................... 57 Final linkage of residents to hospitals ............................................................................................. 58 Aggregation of hospital networks .................................................................................................... 58 Comparison of costs across networks ............................................................................................ 59 Results .................................................................................................................................60 Study population and linkage characteristics ......................................................................60 Patients ........................................................................................................................................... 60 Physicians ....................................................................................................................................... 61 Hospitals .......................................................................................................................................... 61 Discussion ...........................................................................................................................63 Chapter 6 – Defining health service environments .................................................. 77 Background .........................................................................................................................77 Approach .............................................................................................................................78 Data ...................................................................................................................................79 Categories of service use ................................................................................................................ 79 Hierarchical cluster analysis ...............................................................................................81 Results .................................................................................................................................81 Discussion ...........................................................................................................................83 Chapter 7 – Variation in health care costs and population health outcomes in BC ...................................................................................................................................... 95 Background .........................................................................................................................95 Approach .............................................................................................................................97 Data sources ......................................................................................................................97 Client registry (consolidation file) .................................................................................................... 98 MSP payment information file (physician payments) ...................................................................... 98 Discharge abstracts database (hospital separations) ..................................................................... 98 Vital Statistics death file .................................................................................................................. 98 Study population ................................................................................................................98 Primary outcome: Total cost ........................................................................................................... 99 Population drivers ........................................................................................................................... 99 Health system drivers .................................................................................................................... 100 Health outcomes ........................................................................................................................... 101 Geographic units of analysis ......................................................................................................... 101 Model selection ................................................................................................................ 102 Costs ............................................................................................................................................. 102 Mortality outcomes ........................................................................................................................ 102 Results ............................................................................................................................... 103 Study population and exclusions ...................................................................................... 103  vii Models of health care costs ............................................................................................. 104 Unexplained variation in costs among networks and HSDAs ........................................... 105 Effect of unit of analysis ................................................................................................... 106 Observed and predicted outcomes................................................................................... 107 Relationship between costs and outcomes ...................................................................... 107 Discussion ......................................................................................................................... 108 Chapter 8 - Conclusion ............................................................................................. 139 Contributions ..................................................................................................................... 139 Conceptual framework and literature review .................................................................... 139 Multispecialty networks and Health Service Delivery Areas as units of analysis ............... 140 Distinct health service environments ................................................................................ 141 Limited unwarranted variation .......................................................................................... 142 No relationship between adjusted cost and health outcomes ........................................... 143 Strengths and limitations .................................................................................................. 144 Implications for research and policy ................................................................................ 145 Conclusions ....................................................................................................................... 146 References ................................................................................................................. 148 Appendices ................................................................................................................ 164 A Search strategy to identify Canadian geographic variations research .......... 164 B List of references consulted in scoping review ............................................... 165 C Network characteristics ...................................................................................... 190     viii List of Tables Table 1-1 Research questions and hypotheses.......................................................................... 8 Table 2-1 The Dartmouth framework for variation in health care services .................................19 Table 3-1 Terms used in the search strategy ............................................................................41 Table 3-2 Study characteristics .................................................................................................42 Table 3-3 Methods used ...........................................................................................................43 Table 3-4  Examination of causes of variation and outcomes ....................................................44 Table 4-1 Blocked modeling approach ......................................................................................52 Table 5-1 Example data used to aggregate provider network A to a larger network ..................68 Table 5-2 Characteristics of BC residents who were and were not linked for network development ......................................................................................................................69 Table 5-3 Service use characteristics of population used to define networks over three-year study period (N=3,993,149) ................................................................................................70 Table 5-4 Characteristics of physicians included in analysis (n=13,865) ...................................71 Table 5-5 Characteristics of 79 BC hospitals by number of linked patients ................................72 Table 5-6 Network characteristics by patient network size of largest linked hospital .................73 Table 6-1 Categories of service use examined in cluster analysis .............................................85 Table 6-2 Average per-capita health care costs by cluster (health service environment) and percent distribution by category .........................................................................................90 Table 6-3 Population characteristics by cluster, N(%) ...............................................................91 Table 6-4 Health system resource supply by cluster .................................................................92 Table 7-1 Blocked modeling approach .................................................................................... 113 Table 7-2 Models of average annual cost with network fixed effects ....................................... 114 Table 7-3 Average annual costs with fixed effects by HSDA ................................................... 119 Table 7-4 Comparing units of analysis .................................................................................... 124 Table 7-5 Logistic models of all-cause mortality ...................................................................... 125 Table 7-6 Logistic models of premature mortality .................................................................... 130  ix Table 7-7 Logistic models of treatable mortality ...................................................................... 134 Table B-1 Citation information and description of references consulted in scoping review ...... 165 Table C-1 Linked hospitals, patients, and physicians by network ............................................ 190 Table C-2 Loyalty measures, observed, and expected costs by network................................. 191 Table C-3 Breakdown of total costs by network ....................................................................... 192 Table C-4 Linked hospitals, patients, and physicians by satellite network ............................... 193 Table C-5 Loyalty measures, observed, and expected costs by satellite network .................... 193 Table C-6 Breakdown of total costs by satellite network .......................................................... 194     x List of Figures Figure 2-1 Factors shaping variation in health care service use ................................................20 Figure 3-1 Scoping review search results ..................................................................................41 Figure 5-1 Administrative health boundaries in BC ....................................................................67 Figure 5-2 Hospital networks mapped by Health Service Delivery Area (HSDA) .......................74 Figure 5-3 Observed and predicted costs for all networks .........................................................76 Figure 6-1 Dendogram of cluster analysis .................................................................................87 Figure 6-2 Map of cluster solution (health service environments) ..............................................88 Figure 6-3 Map of clusters (health service environments) and hospital networks ......................93 Figure 7-1 Observed and expected costs by networks and HSDAs (models 1-3) .................... 123 Figure 7-2 Plots of all-cause, premature, and treatable mortality ratios and cost ratios (models 1-3) ..................................................................................................................................... 138     xi List of Abbreviations ACO  Accountable Care Organizations ADG  Aggregated Diagnosis Group APP  Alternative Payment Program BC  British Columbia CA  Census Agglomeration CABG  Coronary Artery Bypass Graft CCHS  Canadian Community Health Survey CIHI  Canadian Institute for Health Information CMA  Census Metropolitan Area CMG  Case Mix Group CV  Coefficient of Variation DA  Dissemination Area DAD  Discharge Abstract Database ED  Emergency Department EQ  Extremal Quotient FFS  Fee-for-Service GLM  Generalized Linear Model GP  General Practitioner HA  Health Authority HCC  Home and Community Care HIV  Human Immunodeficiency Virus HRR  Hospital Referral Region HSDA  Health Service Delivery Area ICC  Intraclass Correlation Coefficient ICD  International Classification of Diseases  xii ICU  Intensive Care Unit LHA  Local Health Area MI  Myocardial Infarction MRI  Magnetic Resonance Imaging MSP  Medical Services Plan OECD  Organization for Economic Co-operation and Development PCI  Percutaneous Coronary Intervention PHCO  Primary Health Care Organization PN  PharmaNet PREM  Patient Reported Experience Measure PROM  Patient Reported Outcome Measure RIW  Resource Intensity Weight SCV  Systemic Component of Variation SES  Socioeconomic Status  UK  United Kingdom US  United States      xiii Acknowledgements  First and foremost, I want to thank my supervisor Dr. Kim McGrail for her mentorship, ready accessibility for guidance, and enthusiasm for research. Kim’s support has been unwavering from when I first considered a move to UBC, to the final revisions of this document. Thanks to Dr. Morris Barer, my secondary supervisor for his detailed and thoughtful feedback. Thanks also to my thesis committee members, Dr. Sabrina Wong, and Dr. Mike Law for their valuable input and advice.  I would also like to thank my colleagues at CHSPR. I wish to acknowledge the enormous contribution of Sandra Peterson to this research. Not only did she provide support for data preparation, but I have learned so much from her detailed knowledge of administrative data, programming genius, and analytical savvy. Thanks to Dawn Mooney for her cartographic expertise, and for all she’s taught me about data visualization and communication of results. Thanks to Dr. Charlyn Black for her helpful comments on my thesis proposal. Thanks also to Rachael McKendry and Lindsay Hedden for their wide-ranging knowledge of all things Health Services Research – your countless bits of input and advice have been invaluable. A special thank you to Saskia Sivananthan for being a terrific friend and colleague though all stages of our degree.  I also wish to recognize the Canadian Institutes for Health Research, who provided financial support for my doctoral studies.  Finally, thank you to my friends and family for your love and encouragement. Most of all, thanks to my husband Ian for his unfailing support and partnership.   1 Chapter 1 - Introduction Background For decades, researchers have documented marked variation in health care use and costs across regions, observing that higher spending does not necessarily correspond to better health outcomes (1,2).1 Such analysis is intended to inform health system performance: where population characteristics and health outcomes are comparable, higher-than-average costs suggest the potential for improved efficiency; where outcomes differ, questions about the equity, quality, and appropriateness of health care emerge.  In Canada, the potential to use the lens of regional variation in health care to find opportunities for performance improvement continues to receive attention (3–6). Amidst concerns about escalating health spending and the sustainability of the Canadian health care system, reduced variation in practice through evidence-based care is seen as a strategy to enhance value for money (6). However, much of the existing “variations” literature focuses on the United States (US). (1,2). Whether or not patterns and relationships documented in the US and international literature are also observed in Canada is unknown. Many of the mechanisms thought to explain marked regional variation observed in US studies—for example, differences in price, insurance coverage, degree of cost sharing, or medical malpractice risk (7)—may not apply within single-payer provincial health systems. At the same time, the geographic context in which health services are delivered may play a more important role in shaping service use in Canada, with some differences between urban, rural, and vast remote regions to be expected. In the remainder of this chapter, I provide an overview of my thesis research, and set out the rationale for this work. I briefly summarize the history of geographic variations research and its relevance for contemporary policy issues, and outline conceptual and methodological challenges that persist. I then describe how this thesis contributes to our understanding of how and why use and costs of health care services vary geographically, and the implications of variations research for assessing health system performance.                                                1 This research is sometimes described as “small area analysis.” Variation has also been examined across health care organizations, hospitals, and individual physician practices. Though these are not geographic units themselves, they are associated with definable patient populations and catchment areas, with potentially varied geographic characteristics. This thesis focuses on spatial areas that represent aggregations of individual service providers, and the terms “geographic variation” and “regional variation” are used interchangeably.   2 Thesis overview The overarching goals of this thesis are to critically examine the existing body of geographic variations research (focusing on the Canadian context), and to contribute new knowledge about how and why health care varies geographically in the province of British Columbia (BC). Specific research questions, and associated hypothesis are outlined in Table 1-1. Before undertaking new empirical work, I first consider how variations in health care use, cost, and outcomes have been conceptualized and categorized to date, and develop a conceptual framework used to structure subsequent analysis (Chapter 2). I then report the results of a scoping review of Canadian research examining geographic variation in health care, in order to summarize current knowledge and explore limitations in the relevant literature (Chapter 3). Chapter 4 describes the BC health administrative data used in analysis, as well as methods used to capture health care costs. As assessment of outcomes is important in interpreting any observed variation, I consider available health outcome measures based on administrative data sources. In support of my overarching objective of describing variation in costs, I completed preparatory analysis to help address identified methodological challenges. This consisted of developing networks of patients, physicians, and hospitals that are connected based on patterns of service use, and then comparing these to existing health regions, with the goal of identifying optimal units for analysis (Chapter 5). I also analyzed patterns of service use with the goal of identifying areas of the province that are distinct in their patterns of service use, reflecting different environments for health service delivery, for use in subsequent analysis (Chapter 6).  Finally, I describe variation in costs among health regions, as well as the networks developed in Chapter 5. I determine the degree to which variation is explained by “warranted” causes, namely population need, and the health service environment in which care is delivered. Finally, I examine the relationship between unexplained variation in costs, and population health outcomes (Chapter 7). The final chapter of this thesis (Chapter 8) then offers overall conclusions and directions for future research.   3 Rationale History of “variations” research and relevance for contemporary policy issues Research examining regional variation in health care has proliferated over the past four decades (1,2). In a seminal paper published in 1973, Wennberg and Gittelsohn found marked variation in utilization rates, as well as measures of health human resources, facilities, and expenditures, among hospital service areas in Vermont (8). In the following decades, hundreds of studies examined geographic variation in health care (1,2). Between 2000 and 2010 the annual number of published articles on variation in health care in OECD countries more than doubled (1). Canada follows only the US and UK in the number of studies published in this time period. This body of research garnered increased policy attention with the publishing of the Dartmouth Atlas of Health Care series, beginning in 1996 (9). The Dartmouth Atlas has documented marked regional variation in resource supply, hospital admissions, surgical procedures, and health care spending in the US. Notable research published by the Dartmouth Atlas group found that Medicare enrollees living in high-spending areas receive substantially more services than those living in low-spending areas, but do not have better health outcomes or greater satisfaction with care (10–12). These results have been interpreted as suggesting 30% of health care spending is waste, in that it contributes to neither improved quality nor population health. The US health reform debate, beginning in 2008, may have contributed to renewed interest and attention to geographic variations research, as embedded within such findings is the appealing idea that cost reduction and quality improvement can be achieved simultaneously by limiting wasteful and unnecessary spending (13,14).  The potential to use regional variation in health care to identify opportunities for improvement in both quality and efficiency continues to receive attention in Canada as well (3–6). In the context of concerns over escalating health spending, reduced variation in practice through evidence-based care is seen as a strategy to enhance value for money, while simultaneously improving quality of care, and ultimately patient outcomes (6). As is described in the chapters that follow, a large number of Canadian studies examine access to health services by tracking differences in use across regions, sometimes grouped by rural/urban characteristics. Often the goal of this research is to examine the accessibility of health care services, by identifying low-use areas that may signal barriers to access. Regardless of whether the goal of research is to inform accessibility, efficiency, or broader  4 health system performance, studies examining geographic variation in use or costs of health care are subject to similar conceptual and methodological challenges. I summarize these in the section that follows, and provide more detailed discussion in Chapters 2 and 3. Unresolved challenges   Despite the importance and prominence of issues related to the accessibility, quality, cost, and efficiency of health care systems, and the hope that variations research may offer useful information to address these, its impact on Canadian policy has thus far been limited. This may in part be due to the many limitations, both conceptual and methodological, in the existing literature.  How to interpret observed variation: which differences are warranted?  A central challenge throughout variations research is how observed differences should be interpreted. Defining “variation” as a problem in and of itself cannot inform an effective response and does not speak directly to policy changes that might achieve improved quality or reduced costs (15). In order to act on documented variation, we need to know something about what caused it. A recent systematic review of medical practice variation in OECD countries found a large body of research, but very few studies that examined causes of variation directly, or interpreted observed differences with reference to a clear theoretical framework (1).   Wennberg (16) conceptualizes variation due to patient need or preference as “warranted” and classifies all other variation as “unwarranted.” This means that only variation that remains after accounting for health-related patient characteristics may signal potential problems, but also that accurately accounting for patient characteristics is a central methodological challenge. However, much of the existing literature on variation in health spending relies on area-level measures of both population and health system factors, and lacks adequate measures of patient need or health status (7). Patient-level control for individual health status would offer greater protection against erroneous conclusions about the amount of unexplained, and potentially unwarranted, variation. Adding further complexity to Wennberg’s conceptualization is the fact that in a country as geographically diverse as Canada, we expect some variation in how services are delivered and accessed based on geographic context (for example, increased use of nursing stations or air transport in remote settings, or a wider scope of practice among generalist physicians in rural areas). Provided services are appropriate and of good quality, these differences may also be considered “warranted.” This is often tacitly acknowledged by controlling for urban or rural  5 context in health services research, but seldom explored in any depth. This means that in addition to population characteristics and outcomes, some understanding of how health system composition and organization differs across geographic contexts is needed for meaningful interpretation of observed variation, and crafting health policy that is appropriate to diverse settings. Of course, Canada is not unique in this respect, but its vast geography and expansive rural and remote areas place it in a small group of countries (including, for example, Australia and Finland) that are extreme examples. Describing variation accurately  If the geographic units compared do not reflect actual patterns of decision-making and service use within the health care system, any existing variation may be obscured, and the potential to inform policy is blunted (15,17). Though hospital referral regions (units commonly used in US research) reflect patterns of hospital use by Medicare beneficiaries (18), no similar units exist in Canadian research. The degree to which health administrative units commonly used in reporting regional variation actually correspond to patterns of service use is unknown, as is the implication of choice of unit for the amount of variation observed.  In addition, a pervasive issue in the variations literature is a focus on extreme cases in usage rates or costs, failing to account for the fact that sampling variability alone can generate extreme minimum and maximum values (19). Here also, studies relying only on area-level data make it impossible to compare variation across areas to variation within areas (7).  Measuring outcomes Measuring outcomes relevant to health system performance is central to interpreting variation in use. Only where outcomes are comparable can differences in use or spending suggest conclusions about efficiency. Mortality data is commonly used as it is routinely collected and reflects the health status of the entire population, but is limited by the fact that many factors beyond the health care system influence mortality rates. Recent Canadian research classified causes of death as “amenable” or “treatable,” in order to use mortality to more directly measure system performance (5). However, the extent to which this improves on other measures in isolating the impact of the health care system remains uncertain (20). Contributions of this work In addition to the overarching goal of contributing new knowledge about how and why health care costs vary in a Canadian context, this thesis helps develop conceptual and methodological tools to support future research. The conceptual model presented in Chapter 2  6 may be useful in structuring future analysis of health care variations, and the literature review in Chapter 3 provides a current picture of Canadian research in this area. The multispecialty physician networks developed in Chapter 5 may be useful in future analysis of patterns of care, and demonstrate that an approach first applied in Ontario has broader applicability (21). In contrast to my hypothesis, identified networks correspond closely to existing health regions (Health Service Delivery Areas (HSDAs). This provides useful validation that HSDAs do correspond to patterns of service use – of particular use as researchers can often only access data at the level of HSDA. This finding also supports the ongoing use of these units in national reporting (22). Cluster analysis described in Chapter 6 revealed four distinct areas of the province, which we labeled “health service environments.” This analysis used information on the distribution of health care costs across categories of care, defined based on service type (e.g. hospital, home supports, residential care, pharmaceutical) and provider (e.g. generalist, medical specialist, surgical specialist). Based on this information only, the metropolitan areas of Vancouver and Victoria were identified as distinct from non-metropolitan and remote communities. Non-metropolitan communities that contain large community hospitals and greater physician supply were further distinguished from those with smaller hospitals and fewer physicians. Differences in average total costs between health service environments were modest, while differences in individual categories of care much more marked, suggesting substitution of care across categories. The resulting health service environments proved useful in understanding variation in costs among regions, and offer an analytic approach of potential use in other jurisdictions that is more nuanced in the health care context than simple urban vs. rural dichotomies. Chapter 7 reports modeling of physician and hospital costs for people across BC. Adjusting only for age and sex, marked differences in costs are apparent. However, more complete adjustment for health status dramatically reduces observed variation. While not a major driver of individual-level costs, differences in the structure of the health care system, as are reflected in health service environments from Chapter 6, also helped explain differences in costs among hospital networks and health regions. This demonstrates the importance of accounting for warranted differences in context for service delivery (beyond crude urban-rural differences) in future variations research. In contrast to expectations, choice of geographic unit (networks vs. health regions) did not affect conclusions about the magnitude of variation.  7  Finally, in models that control for health status and non-medical determinants of health, we found no relationship between costs and mortality outcomes. However, at similar levels of spending, we did observe differences in outcomes, even in fully adjusted models. Further scrutiny of outliers (regions with much better than expected outcomes at similar levels of cost) are a potentially fruitful area for subsequent research.     8 Table 1-1 Research questions and hypotheses Research questions Hypotheses Chapter 5  Is it possible to define self-contained networks of patients, physicians, and hospitals based on patterns of service use? Networks can be defined, though service use will overlap among networks. Satellite networks exist in remote areas where services are commonly used at a larger hospital not located close by. Do identified networks correspond to existing health regions? Networks will not be contained within existing health regions commonly used for reporting and analysis.  Chapter 6  Are areas of BC distinct in their patterns of health care use? District areas can be identified on the basis of the distribution of costs across categories of care. How do geographic, population, and health system characteristics differ across identified areas? Areas similar in their patterns of service use will not be geographically contiguous and will not align health regions, but will share geographic, population, and health system characteristics. Is there evidence of substitution between categories of care? Substitution of services will mean that total spending will vary less than spending within individual categories of care. Chapter 7  Do physician and hospital costs vary across areas of BC? Variation in costs will be observed across networks and health regions. To what extent do population and health service environment explain observed variation? A larger proportion of variation will be explained by population characteristics than by health service environment, though unexplained variation will remain. How does choice of geographic unit affects conclusions about the magnitude of variation? Greater variation will be observed among networks identified on the basis of actual patterns of service use, than among health regions. What is the relationship between costs and population health outcomes? Higher cost areas will exhibit poorer health outcomes. Adjustment for population characteristics will attenuate the relationship.   9 Chapter 2 - What causes geographic variation in health care use, costs, and outcomes? In this chapter I describe my conceptual framework for what causes variation in health care use, costs, and outcomes. Based on existing literature, I identify factors both within and outside the health system that may contribute to variation, and consider how these operate at various potential levels of analysis. This conceptual framework is intended to help structure my review of Canadian research, and to inform empirical analysis, especially the choice of appropriate adjustment variables and units of analysis. Background  A recent systematic review found that research examining medical practice variation was rarely based on a theoretical construct, and only about 10% of studies explored causes of observed variation (1). Authors concluded that a conceptual framework is needed to support a more focused body of research, which can uncover the causes and consequences of variation, and support policy and clinical remedies. In the absence of other options, they suggest a framework developed by the Dartmouth group with the categories of effective, supply-sensitive, and preference-sensitive care (16). Effective, supply-sensitive, and preference-sensitive care Dartmouth researchers define three categories of health care (10). Effective care refers to services with a strong body of evidence supporting their effectiveness (Table 2-1). All patients with specific medical need for effective services should receive them. Preference-sensitive care involves trade-offs, where decisions should be based on patients’ preferences and values. Clinicians may also have strong preferences, despite scientific evidence being equivocal about a preferred option in all cases. Supply-sensitive care typically does not have strong theory or evidence to guide decisions about use. For example, medical texts often provide minimal guidance on when to schedule a return visit, perform a diagnostic test, hospitalize, or admit to intensive care (1). In these situations, utilization rates may be strongly influenced by supplier preferences, with the potential for supplier-induced demand. Patients’ preferences and values may also play a role. Based on the Dartmouth framework, variation can be considered warranted or unwarranted (Table 2-1) (10). Effective services should be provided to all patients who need them, as in this category of service there is clear evidence that benefits outweigh any harms.  10 Any variation in effective care, other than that attributable to population need, is unwarranted. In the context of clinical uncertainty, patient preferences (revealed through informed patient choice) may shape warranted variation in preference or supply sensitive care, though provider preferences should not. Variation driven by any other factors is considered unwarranted.  While the Dartmouth framework is a useful heuristic, not all types of care can be cleanly assigned to these categories. While patient need, and the preferences of both patients and physicians clearly drive variation, these are each multifaceted concepts that require further scrutiny. In particular, it is not clear what factors shape “physician preferences,” and this category may be inadequate in capturing sources of variation other than patient need and preference. While physicians may hold certain preferences based on their training, the views of mentors, or colleagues (23–25), other opportunities and constraints can also shape the context in which physicians practice and resulting practice patterns (26). Especially in Canada, local availability of health personnel, equipment, and facilities and other physical resources may play an important role in decisions about care and thus the services patients receive. A more complete conceptual framework of causes of variation must incorporate aspects of the geographic settings where patients live and where services are provided. Existing conceptual models of health care use Several conceptual models of health care use exist, and may inform understanding of geographic variations. However, most existing models provide detail only on one or the other of population factors (also called demand-side in the economics literature), or health system (supply-side) characteristics. In addition, none explicitly recognizes that factors driving variation may act at multiple spatial scales, ranging from individual patients and providers through to large populations and health care organizations. Finally, some frameworks have focused on health care use, without going the step further to make connections to population health outcomes, as well as the outcomes of differences in service use within the health system (most notably differences in cost and ultimately efficiency). In their 1987 review of small area analysis literature, Paul-Shaheen, Clark, and Williams simplify an earlier model by Anderson (27) to identify major factors that contribute to small area variation in health services use (2). They conceptualize community determinants (such as poverty, unemployment, socioeconomic status, and physical environment) as shaping health services system determinants (facilities, services, and personnel), and both, in turn, shaping individual determinants (predisposing, enabling, and illness), and resulting health services use.  11 These identified determinants are consistent with subsequent conceptual models of health care use, though Paul-Shaheen et al. do not describe health outcomes as part of their framework, nor do they consider various spatial scales, beyond “individual,” “community,” and “health services system.” Andersen (28) and Evans and Stoddart (29,30) include outcomes, namely health status, and emphasize that multiple factors outside the health care system also shape health outcomes. They both also note a cyclical relationship, where changes in health status lead to changes in population characteristics. They do not, however, make explicit whether identified factors act at individual physician or patient levels, or aggregate levels, or both. Bernstein, Reschovsky, and White propose a detailed framework specific to variation in Medicare spending in the US (15). In addition to population, environmental, and economic factors, they highlight that cost of services and supplies also vary and that regulatory environment, provider organization, payment structures, and professional culture all drive practice patterns. Though they list many factors, only a subset is relevant within the Canadian health system. Moreover, they provide no discussion of what level of aggregation is relevant to listed environmental, population, and economic factors, as well as health care markets, or how these might correspond. Certainly economic or market factors like tax policy or regulatory environment operate across much larger units than travel time or air and water quality. Manning, Norton, and Wilk outline a conceptual framework for what causes geographic variation in health care use, that starts from the distinction between demand-side and supply-side factors (7). Demand is said to depend on a person’s health status, preferences for care, and budget constraints. Supply encompasses decisions made by individual care providers, but also differences in input costs, provider supply, and broader policy context. Importantly, Manning, Norton, and Wilk also discuss how interactions between patient and provider may fall between aggregate supply and demand. Though their discussion is by far the most detailed, it is not depicted visually, and the relationships among the many listed factors are sometimes unclear. Here as well, many of the listed factors are not relevant in the Canadian context. Skinner also distinguishes demand- and supply-side causes of regional variations, though he focuses on the Dartmouth typology of effective, preference-sensitive, and supply-sensitive care in categorizing health care (18). In the sections that follow I first provide an overview of the rationale underpinning my framework and its structure. I then discuss in greater detail how health care use, cost, and outcomes are defined, before elaborating on factors that may shape their variation, drawing on  12 earlier conceptual models. I conclude with discussion of implications for analysis, most notably choice of units, and identification of control variables. Conceptual framework Overview and structure The conceptual framework depicted in Figure 2-1 is divided into population factors on the left, and health care system factors on the right, with multiple levels of analysis nested within each. These range horizontally from the highest levels of aggregation on the outside of the page, to the patient-provider encounter at the centre. From top to bottom, the model identifies factors shaping variation, the characteristics of service use itself, and resulting costs and health outcomes. Identified factors are influenced by policy, governance, and funding mechanisms, but I do not describe these in detail. The clinical activities and decisions that ultimately determine regional variation in health care use occur through individual contacts between patients and providers. However, factors that shape what transpires at this point of contact (and, if contact occurs at all) are complex and can be described at multiple levels of analysis. Factors shaping health care use can be divided into those within and those outside the health care system. These have also been conceptualized as demand-side and supply-side factors, with demand shaped by the health status, resources, and preferences of the population (shown on the left), and supply shaped by resources and culture within the health care system (shown on the right) (7,18). Provided quality is good and service use is appropriate, contacts with the health care system should lead to reduced risk, effects, and/or duration of health conditions, and ultimately improved health of individuals and populations. Of course factors outside the health care system also shape health, and must be accounted for in evaluating health system performance. Health care use also has a cost, a function of service volume and price. As with health outcomes, this can be described at multiple levels of aggregation. Population and level of analysis The horizontal structure of the model reflects the fact that factors driving variation operate at multiple scales, and variation in health care use, costs, and outcomes may be observed at different levels of aggregation. Health care units on the right (for example, health system, health care organization, and service provider), mirror the populations they serve on the  13 left (country, region or community, and patient). At the core of the model, health care activities and outputs occur when individual patients (who are nested within larger populations) interact with individual service providers (who are embedded within larger organizations and systems). The patient population in research may represent all residents of a defined geographic area, or may be subdivided by demographics, medical condition, health insurance, or other factors. Regardless of the unit examined, geographic setting has the potential to shape both population and health system characteristics. Often geographic setting is described simply as “urban” or “rural,” categories that serve as shorthand for a range of unspecified characteristics. For example, population size (a function of area and population density) influences what resources are available locally. This is true for both human and material resources within the health care system and non-medical determinants of health in the community. Distance and connectivity determine accessibility of resources in other areas, and may influence culture—both styles of practice among health practitioners, and attitudes and values related to health in the population. Population factors Need The most important driver of “warranted” variation in service use is patient need. Age and sex are often used as predictors of need, as are more direct measures of health status (7). We also know that income, education, occupation, and ethnicity may also shape need, and predispose people to seeking health services (28). Factors contributing to need at the individual level (for example, age, sex, health status, ethnicity) also contribute to aggregate need based on the composition of the population (31). Resources Even in a context where there is no explicit financial barrier to medical and hospital services, travel, ability to take time off work, and other factors may all constrain contact with the health care system. Income, stable employment, and access to transportation, are therefore resources that enable use (28). Beyond the simple aggregation of individuals, factors that support use may include collective resources outside the health care system, such as education, transportation systems, and other welfare services (31).  14 Beliefs, preferences, culture and values  Finally, patients with the same physiological characteristics may still vary in their self-assessed health status, beliefs about the importance of health care, priorities and preferences both for accessing care, and choices regarding the use of specific procedures or drugs (28). Patients differ in how they weigh benefits of treatment against costs (if not financial, then time needed to receive care, and side effects of treatment). For example, some patients and families embrace palliative care approaches earlier, while others prefer intensive treatment until the end of life. Norms and values related to health and health care may be shared by populations living in the same area (32,33), though other factors such as religion, ethnicity, or culture, that do not necessarily correspond to geographic regions, may be more important (7). Empirical studies have found patient preference is relatively unimportant in explaining regional variation (34–36), which is not to say it is not an important cause of individual-level variation, but simply that it may not be correlated across geographic units, so once aggregated within regions, individual-level differences disappear. Health care system factors Running parallel to sources of variation related to the characteristics of individuals or populations are factors within the health care system. Human resources Fundamental to the provision of health care are the providers themselves. There is considerable variation in the supply and mix of human resources (e.g. primary care, medical specialists, surgical specialists, nurses, allied health care providers), which can influence the volume and mix of services delivered (2,7,15). Material resources Material resources are broadly conceived. At the practice level this could include laboratory and imaging equipment, or other investments in medical technology. At the hospital or region level, this refers to facilities for inpatient care, as well as rehabilitation and long-term care, but also specific resources such as cardiac units, cancer centers, diagnostic imaging, and emergency response systems (7,15).  Professional culture and practice style Multiple factors have been identified as shaping practice culture (37). Physicians may differ in their belief about the marginal effectiveness of treatments (18). Chassin (1993) hypothesized that authoritative teachers enthusiastic for certain procedures may influence  15 clustering of practice style (23). Wright et al. describe processes of attraction and retention that lead to groups with shared practice style (24). Physician decisions regarding practice location also represent a form of self-selection, as prospective income, as well as preferred practice setting, may lead to physicians with beliefs and practice styles to co-locate (7). De Jong approaches medical practice variation from a sociological perspective, using a constraint- rather than preference-centred approach to determine how social organization influences medical decisions (37). In addition to selection and mutual adaptation as mechanisms whereby physicians within the same practice resemble each other, she notes the importance of circumstances (26). Since individual service providers operate within the contexts of organizations and systems, their practice styles may be shaped not only by their individual training and skills, or compensation arrangements, but also by the resources available to them (15).  Defining health care use, costs, and outcomes Health care use Encounters with the health care system can be categorized by what service was provided (for example, type of procedure or drug), who was the service provider (for example, generalist physician, specialist physician, or nurse), and where the service was provided (for example, community, emergency department, or acute inpatient facility). Within the body of literature describing medical practice variation (1,2), some studies describe rates of specific services, while others report service volume categorized by provider or location. Though not depicted in the diagram, in some systems and for some types of services it may also be useful to categorize services by source of payment (e.g. public insurer, private insurer, out of pocket) (7). Health care costs  Health care costs are a function of service volume and unit costs. In some health care systems (notably the US), prices are extremely variable, with different prices for different insurers even within one institution. In studies only examining the Medicare population, spending must be adjusted for differences in reimbursements for cost of living, hospitals serving low-income patients, and medical training (38). In British Columbia fees paid to physicians are standardized across the province, as are methods for determining the cost of hospital stays. Reimbursement for medical training is administered separately, so price will not shape any observed variation. As with health status and use of health services, cyclical relationships may  16 also be present here. Health care costs may influence policy priorities and governance decisions, as well as available funding going forward.  Quality and appropriateness of care The framework depicts health outcomes as a function of service use and the quality and appropriateness of care. Only if the provided service is of good quality and appropriate to the patient will it positively impact health outcomes. This could include skill at specific procedures, or delivery of appropriate, evidence-informed care. While the body of scientific evidence is in theory universal (and so this is not included in the framework as a potential cause of variation), providers may differ in their knowledge and interpretation of evidence (7). Health outcomes  A range of potential individual or population health outcomes could be conceptualized depending on the unit of analysis, population, and type of service examined. Ideally outcomes would take into account morbidity and prognosis (7) and could include measures of effectiveness for individual services. However, mortality-based measures are much more commonly reported (if outcomes are reported at all) (1). Though not shown, there could perhaps be an arrow connecting health outcomes to population factors shaping variation, because as health care cures disease and contributes to improved health status, it modifies need for health care (29,30).  Most importantly, while health outcomes are shaped by service use (and the quality and appropriateness of services), population factors and health supporting activities external to the health care system also determine health status. Isolating the impact of the health care system represents a central methodological challenge throughout this body of work. Discussion A better understanding of factors that drive variation is important in crafting and evaluating health care policy (15,17). The conceptual framework described above illustrates that observed variation in use of health care may be driven by underlying variation in any number of contributing factors, tracing through multiple pathways, both inside and outside the health care system. This framework thereby provides a more comprehensive picture of sources and levels of variation than embodied in existing conceptions of causes of variation. Up to this point, it has been common practice to adjust for known (and measurable) sources of variation, often with population characteristics considered “warranted” sources of that  17 variation. Any remaining unexplained and thus “unwarranted” variation is then attributed to the health care system, though this is neither measured directly nor named specifically. Inadequate measurement and control of sources of warranted variation can thus lead to inaccurate conclusions about the magnitude or causes of unwarranted variation through residual confounding. Existing frameworks tend to focus on one or the other of population or health system characteristics, and so lack sufficient detail to structure analysis that sufficiently accounts for both population and health system causes, and the ways in which they might interact. If tracking regional variation in health care is to inform policy interventions, units of analysis must appropriately reflect levels of decision-making and patterns of service use within the health system (17). There may be variation in individual provider practice, but these variations will cancel out over large numbers of physicians, unless some drivers are correlated systematically within regions (18). Up to this point, there has been a fair amount of description of geographic variation (1,2), but relatively less consideration of the mechanisms by which these variations arise (37). Existing conceptual frameworks only rarely distinguish between individual and area-level factors, let alone consider how the two interact.  Finally, though not included in most existing frameworks, interpreting whether or not observed variation in use points to inefficiency or inequity hinges on better assessment of the outcomes of care. Individual or population health measures are only partially determined by health care, and careful control of factors external to the health care system is required. The Dartmouth group has used the approach of comparing Medicare beneficiaries who died, reasoning that the outcome is certain and constant across populations (39). However, even just prior to death, health status and expected resource use can still vary substantially (40). More direct measures of service quality, outcomes (especially changes in morbidity), and patient experience are needed, but not routinely collected. This chapter has examined prominent international research focused on geographic variation in health care. While the resulting framework and discussion are intended to apply to Canadian research, additional conceptual and methodological challenges may be of particular concern in the Canadian context. On one hand, Canada’s rural and remote settings present unique challenges for service delivery, and may themselves contribute to observed variation. On the other hand, Canada’s provincially administered health systems may reduce or eliminate variation driven by factors such as price, insurance status, and provider organization. In the  18 following chapter I review Canadian literature, and consider methodological challenges important within the Canadian context in greater depth.    19 Table 2-1 The Dartmouth framework for variation in health care services  Effective Preference sensitive Supply sensitive Evidence Strong theory and evidence benefits outweigh risks for patients in need Clinical uncertainty / multiple options with different risks and benefits Weak theory and evidence Warranted variation  Patient need  Patient need  Patient preference  Patient need  Patient preference Unwarranted variation  All other factors  Physician preference  Physician preference Examples  Childhood immunizations  Beta-blockers following heart attack  Bypass surgery for heart disease  Physician visits  Diagnostic imaging services  20  Figure 2-1 Factors shaping variation in health care service use 21  Chapter 3 - Geographic variation in Canadian health care Background There is an extensive body of research examining variation in health care use in the US and internationally (1,2,7). This research is based on the core idea that examining variations in use or cost provides useful information, signaling barriers to access, or the potential for improved efficiency. A recent systematic review of literature on variation in health care in OECD countries found large variations across regions, hospitals, and physician practices (1). However, authors noted that of the 836 studies reviewed, only 10% had explored causes of variation, and the research was rarely based on a clearly articulated theoretical construct or framework.  The central idea that examining variation may provide useful insight has also been applied across a wide range of health conditions and categories of health care services in Canada. In moving from a US or international to Canadian context, the relative importance of factors driving variation may change. For example, individual enabling resources in the form of insurance coverage and health system funding arrangements are less variable in Canada. At the same time, in a country as geographically diverse as Canada, we may expect to see more extreme variation in the supply and mix of resources available, and resulting practice scope and style. A large Canadian body of research comparing service use across the rural-urban continuum shares the same core goal of identifying barriers to access (or, in some cases, potential for waste) through geographic comparisons. This chapter describes a scoping review I undertook to map research examining geographic variation in health care in Canada, summarize key findings, and identify limitations. An additional objective was to document the methods used to describe variation, which are themselves quite variable, and may shape conclusions about observed differences. Methods were not discussed in the existing international reviews (1,2). A scoping review was chosen instead of a systematic review because of the nature of the topic of geographic variations. There are no simple subject headings or search termss that would make it easy to find all relevant literature. More importantly, almost any piece of research could be seen to be relevant to “variations” in one form or another. Even focusing on geographic variations, as this work does, leaves a very broad frame and the likelihood that hundreds if not thousands of relevant papers could be identified. Given that the interest here was to identify themes and areas of concentration, a scoping review was deemed a manageable approach. I  22  focus on geographic variation, rather than variation observed at the level of individual patients or providers (which comprised 26% and 24% of the studies in the most recent international review, respectively) (1). More details on inclusion and exclusion criteria are provided below. Methods Search strategy I identified studies published between January 1, 1973 and January 1, 2013 using a Medline search. I chose 1973 as the start date because a seminal article by Wennberg and Gittelsohn describing small area variation in surgical rates was published in that year (8). I included all studies focusing on Canada that examined health care services use or cost, and made comparisons across geographic areas (for example, health regions, counties, provinces, or areas defined as urban and rural). Studies comparing across hospitals where a catchment area was implied were also included. The following were excluded:  Studies with no geographic dimension of comparison. There are many studies examining variation across income, education, ethnicity, and other factors. While many of these factors may be related to geography, the aim of my review was only to examine papers where geography was a focus, rather than a correlate;  International comparisons with no Canadian sub-national analysis;  Studies examining variation in incidence or prevalence of disease or health behaviours, with no examination of health care service use or cost;  Articles describing variation in physician training or physician-level reported practice patterns, with no measure of service use or cost;  Articles describing variation in the distribution of material or health human resources, with no measure of service use or cost; and  Clinical practice guidelines, articles on research methodology, editorials, or review articles that reported no original research. There is no specific Medical Subject Heading (MeSH) term for geographic variation in use of health care services, so I used an iterative process to identify search terms, first examining terms from known relevant articles, and then taking additional terms from newly-identified articles. I included terms to focus the search on health care, rather than health status, to identify that variation was a focus, and to flag that comparisons were made on the basis of geography (Table 3-1). I required three-word adjacency between terms capturing “geography”  23  and “variations” to restrict results to articles where these terms were closely associated in text. The final search was multi-purpose (.mp), identifying terms in the title, abstract, and key word fields. The detailed search strategy is provided in Appendix A. French language studies (identified using their translated title and abstracts) were retained. No studies were identified in other languages. I did not formally search the grey literature, but reviewed health service atlases published by Canadian research centres. Data extraction I extracted information on the data source used and population examined in each study. I classified studies based on how they capture health care use, distinguishing between those that examined rates of specific procedures and drugs and those that looked at service volume by provider (e.g. GP or specialist) or location (e.g. home or hospital). I also tracked the small subset of studies that assessed variation in the quality or appropriateness of services provided. I describe the methods used in several ways, using the conceptual framework from the previous chapter as a guide. First, I recorded the geographic units used to make comparisons. These range from sub-provincial counties or health regions to provinces or groups of provinces. Some studies make comparisons on the basis or urban or rural locations. As definitions of urban and rural places also subdivide the country into discrete areas, these studies are included and their results described separately. Second, I identified the approaches used to describe and quantify variation. Third, I tracked methods used to control for population characteristics that are considered to represent “warranted” sources of variation (18).  I also recorded if any studies identified causes of variation. Population (or supply-side) causes include age and sex; measures of health status; socioeconomic status; race, ethnicity, visible minority, or new immigrant status; and beliefs or values. Health system causes include supply of health human resources, health facilities (for example, diagnostic or acute care), and professional culture. Finally, I recorded whether the study examined health outcomes, and if so what measure was used. Results This search yielded 359 results when executed in Ovid Medline on January 14, 2013. Of these articles 7 were duplicates, and 170 were excluded based on review of title and abstract (Figure 3-1). Examining full text, an additional 71 were excluded, leaving 111 articles for in-depth review.  24  Study characteristics Data sources and study populations The majority of studies (77, 69%) used health administrative data (Table 3-2). A subset (26, 23%) used surveys, including population-based health surveys (for example, the National Population Health Survey and Canadian Community Health Survey), project-specific patient surveys, and/or surveys of individual service providers (included only if used to solicit patient-level data maintained in their practice). Eight studies (7%) used a combination of survey and health administrative data. Some studies were restricted to a population with a defined condition (43, 39%), while the remainder reflected the whole population of the study areas. Health care outputs examined  The majority of studies (68, 61%) examined health care services defined by a provider or location, for example family physician visits, specialist visits, hospitalizations, home care, or emergency services (Table 3-2). Fifty (45%) looked at variation in specific procedures or drugs, often within a defined patient population. Examples include rates of arthroplasty, revascularization after an acute myocardial infarction (MI), angiography, epidural, caesarian section, various screening tests, and use of classes of drugs like antihypertensives, statins, and opioids. Wait times for a specific service (e.g. arthroplasty or cardiac interventions) or length of stay in a given location (e.g. hospital, ICU) were also examined in 13 studies (12%). Six studies (5%) defined and assessed variation in quality or appropriateness of service use. Methods used in identified studies Geographic units and urban/rural definitions Geographic units were defined in multiple ways (Table 3-3). Health regions were the most common choice (40, 36%), with a few using catchment areas for hospitals or clinics rather than delineated administrative regions (8, 7%). County or municipal geography and census units were also commonly used (12, 11% and 9, 8% respectively), while a handful of studies (5, 5%) used other researcher-defined units. Provinces or groups of provinces (e.g. Atlantic Canada, the prairie provinces) were compared in 14 studies (13%). A system of classifying urban and rural areas was used in 36 studies (32%). In 23 of these studies, comparisons were made only between urban and rural areas and not also among other units described above. Only a handful of studies compared multiple units of analysis, but there is some evidence that the unit selected shapes conclusions. For example, Feasby found that provincial  25  rates of carotid endarterectomy ranged from 25.7 to 82.8 per 100,000, but at the level of census divisions ranged from a low of 0 to a high of 179 (41). There is evidently a balancing act, where identified units need to be small enough that significant and important between-area variation is not masked, but large enough that stable estimates for utilization rates and outcome measures can be calculated. Units must also appropriately correspond to the level at which factors driving variation operate. Variation was observed across provinces (or grouped provinces) in kidney transplantation from a deceased donor (which is provincially organized), but within regions rates were not affected by distance from the closest transplant centre (42). Jimenez-Rubio et al. found that inequalities in provincially funded health care use (measured by GP visits, specialist visits, and inpatients stays) were mainly driven by variation between provinces, with higher rates observed in provinces with greater economic resources. Inequality in self-reported health (measured by the health utilities index) was due to differences in socioeconomic status between individuals within, rather than between, provinces (43). Of the 38 studies examining urban/rural differences, most used a dichotomous definition. Two studies used Census Metropolitan Areas and Agglomerations (CMA/CA) to identify urban areas, two used the middle digit of the postal code (0 if rural), while in 19 studies urban areas were researcher-defined (generally by identifying major cities in each province, for example, Winnipeg in Manitoba, Edmonton and Calgary in Alberta, or Halifax and Sydney in Nova Scotia), or the approach was not stated. Twelve studies used the census definition of rurality, where urban areas have a core population greater than 1000, with population density of more than 400 people per square kilometer. Five of these studies further distinguished between larger urban centres (CMA/CAs), and smaller urban areas captured by the census definition, yielding three categories for analysis. Three studies used the Metropolitan Influenced Zones classification, where areas outside of CMAs and CAs are subdivided into four categories based on the degree of metropolitan “influence,” measured using commuting patterns from census data (44). Ten studies grouped areas by proximity to medical care or health care resources available locally (for example tertiary care, or in some cases condition-specific services such as cancer centres or diagnostic testing facilities). In a few cases authors referred to areas far from hospitals or with low health resource availability as “rural.” Unless another definition of  26  urban/rural was also used, I classified these studies as having assessed availability of resources as a potential driver, but not as making comparisons on the basis of rurality. Methods to describe and quantify variation  Most studies reported means, proportions, or rates stratified across areas, sometimes reporting ratios or differences for high and low areas (104, 94%). The ratio of the highest to lowest age- and sex-adjusted rate is sometimes called the Extremal quotient (EQ). Occasionally Χ2 statistics were used for the comparison of standardized area rates to the overall rate (in some cases adjusted for multiple comparisons).  Summary measures of variation were used in 16 studies (14%). Chosen measures include the Coefficient of Variation (weighted, CVw or unweighted, CV), and Systemic Component of Variation (SCV). The CV is the ratio of the standard deviation of health area rates, divided by the mean health area rate (and multiplied by 100). The EQ, CV, and CVw all use directly age- and sex-standardized rates for each area. The SCV measures variation in rates adjusting for random variation within regions, and is based on observed (yi) and expected (ei) cases for each area i. Expected cases may be predicted using multivariate models (with age, sex, and sometimes health status), or using indirectly standardized rates from a reference population. More detailed definitions of each measure are discussed by Ibanez et al. (45) and others (19,45–49).  Only four studies (4%) used multilevel models to partition individual- and area-level variation. Resulting individual- and area-level variance components can be used to calculate the intraclass correlation coefficient (ICC) which reflects the proportion of variance accounted for by areas, or related measures of area heterogeneity appropriate to logistic regression (50). Non-hierarchical regression approaches were also used, assigning dummy variables to study geographic areas. As this approach does not quantify variation directly, these studies are grouped with those reporting stratified results.  Comparing magnitude of variation across studies is difficult. Some degree of fluctuation in rates is to be expected due to chance alone. Differences in geographic units compared, time periods examined, and condition prevalence all affect the stability of rates, and therefore the degree of variation observed. While the SCV attempts to compensate for variation due to chance alone, only a subset of studies reported this. It is therefore not possible to quantify systematically and compare the degree of variation across studies. In the section that follows, ratios of the highest to lowest age and sex-adjusted rates are reported. These are sometimes  27  reported as the EQ, but can also be calculated in studies that only report values by strata. This measure is problematic in that it is highly sensitive to outliers, but allows for some degree of comparison across all studies. Controlling for relevant population characteristics The health care needs of populations vary, and (appropriately) drive variation in the use of health services. Therefore, controlling for population composition (or some other appropriate proxy for area-level need) is a key methodological challenge throughout the body of variations research. Social and economic factors that enable access, and shape patient preferences may be factors researchers wish to study, or may require control to isolate sources of variation attributable to the health care system. What constitutes appropriate measurement and adjustment for these factors depends on the goal of analysis. All studies reporting summary measures of variation (EQ, CV, SCV) used some form of standardization, and so results are adjusted for population age and sex at a minimum. Similarly, all studies using multilevel models adjusted for some variables capturing population composition. Of the 91 studies describing variation by strata only (i.e. no modeling or other explicit quantification of variation), 25 used age- and sex-standardized values, 31 reported multivariate-adjusted values, and 35 reported only crude effect measures. Eleven of the crude-only studies are of condition-defined populations, so assumptions that they are relatively homogenous with respect to health status and health care needs may be reasonable. The lack of any adjustment in the remaining 24 crude-only studies makes the meaning of observed variation especially difficult to interpret. Even in regression-based approaches, the complement of control variables included in analysis differed markedly among studies, making it difficult to interpret adjusted estimates and compare across studies. Summary of observed variation by type of health care use and geographic unit  Where relevant, in the sections that follow the ratios of high to low rates are reported in parentheses (adjusted for population characteristics, if provided), in order to facilitate comparison among multiple studies examining similar services. Once again, this measure is highly sensitive to outliers, especially in analysis of small areas or rare events.  28  Procedures and drugs Sub-provincial regions  Marked variation in rates of individual procedures has been observed across sub-provincial areas. A number of studies were conducted in Ontario at the county level, allowing direct comparisons of the magnitude of variation. A range of procedures was found to have more than five-fold variation in rates among Ontario counties, including insertion of tympanostomy tubes (9.6) (51), anti-reflux surgeries (5.7) (52), knee replacements (5.2) (24), echocardiography (5.4) (53), and degenerative lumbar spinal surgery (5.0) (54). An Ontario atlas also reported county-level variation in Ontario, and found more modest ratios for hip replacement (1.6), knee replacement (2.4), cholecystectomy (1.9), and hysterectomy (2.5) (55).  At the level of health regions (for example, District Health Councils in Ontario or Health Service Delivery Areas in BC) researchers found two- and three-fold variation for a variety of procedures, including revascularization rates post-MI (3.08) (56), tonsillectomy rates (3.0) (57), invasive cardiac procedure (1.7) (58), pediatric appendectomy (positive for appendicitis 1.7, negative for appendicitis 3.8, and perforated appendicitis 1.7), caesarian section (1.7) (59), and arthroplasty for femoral neck fracture (1.4) (60). An earlier study in Quebec also found ratios close to two for a number of procedures among Quebec community health districts (cholesystectomy 2.3, total hip replacement 2.5, coronary artery bypass surgery (CABG) 3.0, inguinal hernia repair 2.0, hysterectomy 2.5, appendectomy 2.7, prostatectomy 2.3, varicose vein stripping 2.6), with the exception of tonsillectomy which varied more markedly (5.2) (61). Screening and diagnostic procedures were frequently measured at the health region level, including flow-study use (3.6) (62), mammography (screening and diagnostic combined, 2.5) (63), gastroscopy (2.9), colonoscopy (2.0), sigmoidoscopy (2.3) (64), and endoscopic retrograde cholangiopancreatography (diagnostic and therapeutic, 2.0) (65).  Variation in rates of drug use in relevant patient populations was generally more modest than for procedures, including statin use in Ontario (1.8) (53), asthma drug use in Quebec (2.0) (66), and a range of drugs examined in BC (statins 2.9, antidepressants 2.3, opiods 2.1, acid-reducing drugs 1.7, and antihypertensives 1.3, adjusted for individual age, sex, income, and health status) (67). Prescribing of high-risk drugs to patients age 65+ was very low in the northern health care region of Quebec, but elsewhere ratios ranged from 1.2 for rational prescribing of NSAID combinations to 1.6 for rational psychotropic drug combinations (whether prescribing was rational was determined by contraindications and duration) (68).  29  Provinces and larger regions  Variation has also been observed for specific procedures and drugs across provinces and subnational regions. The kidney transplantation rate (from deceased donors) varied widely by province, with the rate in Alberta 3.7 times that of neighbouring Manitoba (42). Mastectomy rates were found to vary 3-fold (69), and the same for carotid endarterectomy (3.2) (41). Rates of any revascularization procedure showed a ratio of 6.9 among provinces (CABG alone was 3.9, while percutaneous coronary intervention (PCI) was 8.3, though Prince Edward Island was a very low outlier) (70). A study of medication use among Canadians aged 65 and older found significant regional differences for acetaminophen (1.5), benzodiazepines (1.9), nitrates (2.9), and complementary and alternative medicines (1.8) (71), though not for other medications examined. Modest variation (1.5) was observed for HIV antibody testing among gay and bisexual men (72). Total expenditures per active tuberculosis case ranged from $28,259 (Atlantic provinces) to $72,441 (Northern Territories) (73). Comparisons across the urban-rural continuum  Few specific procedures were examined across the urban-rural continuum. Differences in drug treatment have generally been found to be modest, with no consistent trend, though hormone replacement therapy was slightly higher in smaller cities and rural areas of Quebec (74). Rates of bone mineral density testing were lower in rural areas and did not increase following publication of guidelines, while bisphosphonate treatment rates did not exhibit an urban-rural gradient and increased following guideline publication (75). Service location or provider Sub-provincial regions  Variation in use by class of service has also been reported. Given that these studies are more often based on entire populations, and that frequency of use is high, variation was less extreme than was observed for individual procedures.  Variation tends to be more modest for primary care than for specialist services. For example, a Manitoba study documented the number of family physician visits per child ranged from less than two to greater than four (two-fold variation), while specialist visits ranged from 0.51 to 1.82 per child (a ratio of 3.6) (76). A Nova Scotia study found the average number of primary care visits ranged from 3.1 to 4.5 per person (for a ratio of 1.4), whereas specialist visits ranged from 0.8 to 1.5 (a ratio of 1.8) (77). In a later study, the average number of visits to primary care ranged from 3.3 to 5.5 per person among Nova Scotia areas (census consolidated  30  subdivisions in rural areas, enumeration area-based neighbourhoods in urban areas), for a ratio of 1.7 (78). Use of muscluloskeletal specialist physicians among patients with arthritis and rheumatism ranged from 4.2% to 15.4% among Ontario health areas, for a ratio of 3.7 (79). The number of dental services provided to people age 65+ in Ontario communities ranged from 9.1 to 15.3 (ratio of 1.7) (80).  Hospital use has been found to vary more markedly than physician services use, with a range of 0.16 to 0.35 visits per person per among areas of Nova Scotia (ratio of 2.2) (77), and a range of 30 to 114 (ratio of 3.8) among Manitoba children (76). The degree of variation may also depend on the type of hospital services examined. Ratios as high as 13.0 were found for non-surgical hospitalization rates for back and neck problems across Ontario counties, while the ratio for surgical rates was 2.3 (81). The percentage of ambulatory care provided by emergency departments ranged from 2.6 to 10.8 among Winnipeg neighbourhoods (82).   Two studies examined homecare use among sub-provincial regions. Coyte and Young found 3.5-fold regional variation following inpatient care (8.6-29.9% of separations) and a 7-fold variation in rates of use following same-day surgery in Ontario (1.7-11.9% of separations) (83). On the other hand, only modest variation in home care use after hospitalization (a ratio of 1.6) was found among Manitoba regional health authorities (6.3-10.3% of hospitalizations) (84).  Provinces and larger regions  Variations across service categories are also apparent at the provincial level. A study reporting findings from the 2001 Canadian Community Health Survey found the average number of per-capita GP visits per year ranged from 1.0 in Newfoundland to 2.5 in Quebec, specialist visits ranged from 1.1 in Newfoundland to 1.6 in Ontario, and nights in hospital ranged from 0.65 in BC to 1.14 in New Brunswick (43). Aboriginal residents of the three northern territories were more likely to have consulted a nurse or traditional healer, and less likely to have consulted a doctor than aboriginal residents of the ten provinces (85). Variation was found in surgical rates, hospitalization, and drug use among patients with schizophrenia (86) and length of hospital stay for births (87).  Comparisons across the urban-rural continuum  Many studies have compared volume of service use across the urban-rural continuum. Frequency of contact with specialist service providers is consistently higher in urban areas. For example, 57.4% of Quebec urban residents with myocardial infarction saw a cardiologist, compared to 31.5% in small towns and 37.0% in rural areas (88). Residents of Winnipeg (age  31  65+) were more likely to have a hip or knee replacement or cataract extraction than other Manitobans, though rates of total physician visits did not differ (89). Among patients with chronic disease in Quebec, consultation with specialists was highest in census metropolitan areas, and declined with decreasing metropolitan influence, such that residents of rural and remote areas were less than half as likely to receive specialist services (74). In a study with sites in Alberta and Ontario, rural residence was associated with higher rates of appendectomy, cholecystectomy, and carpal tunnel release, but there was no difference for closed hip fracture repair, rectal cancer surgery, thyroidectomy, or inguinal herniorrhaphy (90), Rural residence was also associated with higher rates of pediatric appendectomy and lower ultrasound/CT use (91).  Among seniors with mental health problems, a higher proportion of rural Canadian seniors saw their family doctor more than 6 times, though a higher proportion also did not see a GP at all, compared to urban residents (92). Another study found that 16.1% of rural residents with probable anxiety or moods disorders consulted mental health specialists, compared to 22.2% of urban residents (93).  Differences in use of hospitals and emergency departments were evident across the urban-rural continuum. In a national study, hospitalization rates were found to increase with increasing rurality, while average length of stay decreased (94). Provincial studies found high rates of hospitalization among rural residents (89,95,96), though a Nova Scotia study found that the average number of days in hospital was lower in non-metropolitan areas (outside Halifax and Sydney) (77). A greater proportion of rural residents were found to receive care in emergency departments or outpatient clinics, and to have more limited access to community-based ambulatory care facilities (94). In Quebec 51.4% of rural residents reported at least one emergency department visit compared to 32.5% of urban residents (97). In Alberta, urban patients with incident heart failure had more office-based physician visits, fewer hospitalizations, fewer emergency department visits, and greater use of efficacious pharmaceuticals, than rural patients, though volume and pattern of ambulatory care use did not differ greatly (98). Similarly, while asthma standards of care were found to be comparable, urban asthma patients made fewer emergency department and hospitals visits (99). Rural and remote diabetes patients were more likely to visit an emergency department for diabetes care, and to be hospitalized for diabetes-related complications (100). Olatunde et al. found that emergency department, after-hours services, and hospital inpatient care were a significantly greater part of rural primary care physician practice (101), which supports the idea that rural emergency departments may be filling a different role in providing basic primary care than their urban counterparts (97).  32  Quality/appropriateness of care Sub-provincial regions  Services indicative of quality care were found to vary at the sub-provincial level. Variation has been found for chemotherapy in the last two weeks of life, dying in an acute care bed, physician home visits (102), and radiotherapy use (by region, and unrelated to travel time) (103). The percentage of decedents hospitalized within the last month of life (an indicator of potentially inappropriate care) ranged from 45.9% to 70.2% among Manitoba regions (104). Variation was higher for negative pediatric appendectomy (where no appendicitis was found) (3.8) than positive (1.7) in Ontario (91).The percent of births with inadequate prenatal care ranged from 1.1-21.5% in Manitoba (105). Use of Intensive Care Unit clinical practice guidelines for nutrition ranged from 20-70% and from 20-80% for weaning from mechanical ventilation among BC health authorities (106). Variation in wait times following abnormal mammography screening (2.0-4.7 weeks) (107), and for coronary angiography (3-7 days among urgent patients, 22-69 days among elective) (108) were also been reported.  Provinces and larger regions  A handful of studies examined indicators of quality of care, wait times, and outcomes across provinces. The percentage of adult CCHS cycle 3.1 respondents not receiving treatment for diagnosed hypertension (used as an indicator of quality of care) ranged from 8.8 in the Atlantic region to 29.2 in the territories (109). Mean length of hospital stay following childbirth ranged from 2.6 days in Alberta to 4.3 days in Newfoundland and Prince Edward Island, and was not explained by rates of medical complications. Median wait times for revascularization procedures ranged from 8 days in Alberta to 29 in Nova Scotia (70). Readmission rates (30-day) following heart failure ranged from 6.5% in Quebec to 10.4% in BC, while in-hospital mortality rates ranged from 7.5 per 100 patients in Saskatchewan to 11.9 in Newfoundland/Labrador (110). Mortality rates were standardized for age and sex, but there was no other adjustment for case mix. Comparisons across the urban-rural continuum  Some differences in quality of care have been reported. Rural/remote residents with terminal lung cancer or COPD were less likely to receive physician visits, home care, and home palliative care (111). Longer waits for non-emergency follow-up were found among patients presenting to the emergency department with COPD in smaller urban centres than in major metropolitan areas (112). Urban residents of Manitoba were more than twice as likely as rural  33  residents to receive home care (113), and urban residents received more hours of home care per year in Alberta (114). Identified causes of variation Population causes examined  Sixty-four (55%) studies examined population causes (beyond age and sex of the population) (Table 3-4). Of these, 24 (38%, or 21% of the total) measured health status in some way, including self-reported health, years with illness, specific conditions, comorbidities, or condition-specific measures of disease progression. A measure of socioeconomic status was included in 41 studies (64%), most often operationalized as average area income, education, and/or employment. Seventeen studies (11%) included a variable for race, ethnicity, visible minority, or new immigrant status, while seven (11%) examined some measure of beliefs or values related to health care services use.  Findings regarding the impact of population causes differed. Both individual and neighbourhood factors were found to influence mental health service use (115). Strong regional variation in hospital admissions was also largely explained by socioeconomic factors (116). Social demography and needs explained a greater proportion of variation in mental health service use than physician supply (117,118). In a unique study Hawker et al. directly surveyed residents of counties with high and low arthroplasty rates, and found both need and patient willingness to undergo surgery were higher in the high rate county. Health system characteristics or physician preferences were not examined (36). Another study reviewed the medical records of patients from high and low rate regions and did not find differences in the rate of inappropriate procedures (119).  Other research found that substantial regional caesarian use was not explained by patient illness or preference (59). Population need also did not explain marked (30-60 fold) variation in distribution of harm reduction products in BC (120) or four-fold variation in fee-for-service expenditures for mental health in Ontario (121). Case mix did not explain variation in total surgery rates among Manitoba hospital areas, suggesting decision-making around surgery, not population characteristics, drives differences in rates (122). GP visits were found to track area-level premature mortality (interpreted as showing that those with greater need receive services), but specialist visits, high-profile procedures, screening, and preventative services did not bear any relationship to premature mortality (123). Per-capita mental health spending  34  ranged from $13.70 in the North to over $50 in the central east region of Ontario, even though the rate of mental disorder was highest in the north and lowest in central east (121).  As age, sex, health status, and other proxies for patient need are considered warranted drivers of health care use, we expect them to explain a substantial proportion of variation. However, results typically show a substantial proportion of variation remains unexplained. This is often interpreted as suggesting that factors in the health care system may be driving variation in rates, especially for use of more specialized services. Health system causes examined  Forty-four studies (39%) examined health system causes of variation (Table 3-4). Of these, 22 (50%) examined supply of health human resources, while 18 (41%) looked at health facilities or other material resources (for example, equipment or acute care facilities). Five (11%) measured some aspect of professional culture, such as propensity to refer, or enthusiasm for specific procedures. Health Human Resources Supply  Supply of specialists tended to explain variation in use more strongly than other health care resources (79,124,125). Higher specialist supply was associated with higher rates of cardiac intervention (88,124), orthopaedic office visits (126) and higher specialist service use for arthritis and rheumatism-related diagnoses (79). While need for pediatric mental health services did not vary significantly across regions of Quebec, distribution of service providers did, and rates of use appeared to correlate with provider supply (127). Invasive cardiologist supply was associated with lower wait times for coronary angiography among urgent cases, but not semi-urgent or elective (catheterization laboratory was a stronger determinant) (108). One study found that greater surgeon availability was related to lower appendectomy rates in children, but no other variables were controlled for (91).   Having a regular source of primary care was associated with PAP-testing (128) but area-level primary care physician use was not related to the diagnosis and management of hypertension (78), or to prescription drug use (though population characteristics were) (67). Health region GP supply was a significant predictor of use of GP services for mental health reasons, while specialist supply was predictive of use of psychiatric services (129,130). Another study found that physician supply was significant in explaining regional variation in prevalence for mental disorder (determined by physician treatment), though variables reflecting patient need explained a larger proportion of variance (117).  35  This is not to say that differences in supply have always or everywhere been found to correlate strongly with differences in use or service provision. While the per-capita supply of specialist physician varied two-fold across Manitoba regions, services use varied more modestly (125). In Ontario, supply of otolaryngologist was not correlated with tympanostomy tube in insertion (51), nor was surgeon supply associated with rates of spinal surgery (54). Pediatric neurologist supply (0.7 to 4.0 specialists per capita) was not correlated with differences in the hours providing patient consultations in clinic (which ranged from 3 to 20) or wait times (3 to 72 weeks) (131). In general, physician supply appeared to have a greater effect on service use for specialist services than for primary care. However, the links between physician supply and service volume are not straightforward. For example, orthopedic workforce supply is associated with number of office visits, but not with hours in surgery (126). In addition to human resource supply, other health system resources, such as operating room availability, shape patterns of use. Material resources/Health facilities  Findings concerning the impact of material resources (such diagnostic facilities and acute care beds) are mixed. Higher rates of surgery for degenerative lumbar spine conditions were found in counties with more MRI scanners (54), and reduced wait times for coronary angiography were observed in areas of higher catheterization laboratory supply (108). Variation in access to radiotherapy is thought to drive provincial differences in mastectomy vs. breast conservation surgery rates (69). Hospital resources (on-site catheterization lab and revascularization facilities), and status as an academic centre were associated with higher 30-day revascularization rates, but accounted for only a modest amount of regional variation in acute myocardial infarction mortality outcomes (124). Distance from bone mineral density device was highly predictive of bone mineral density testing, but variation in testing was not related to rates of osteoporosis treatment (132).  Availability of pediatric beds was associated with high admission rates for gastroenteritis, and may indicate unnecessary hospitalization (133). Hospital bed supply was the only significant factor explaining 14-fold variation in hospitalization rates for gastroenteritis among children (133). Hospital bed and physician supply did not predict GP services use, but did predict specialist services use (which was lower in areas with fewer physicians and hospital  36  beds) as well as emergency department and hospital admissions (which were higher in areas with fewer physicians and hospital beds) (134).    Distance to services, often used as a proxy for access, did not explain wait times for pediatric cancer treatment or rates of kidney transplantation (from a deceased donor), or invasive cardiac procedures (42,135). Distance to a cardiology center was associated with invasive cardiac procedure rates, but did not fully explain variation in those rates (58). Access to a teaching hospital was associated with higher rates of CABG, but lower rates of appendectomy, cholecystectomy, hysterectomy and tonsillectomy (61), and regional rates of ultrasound use were not related to rates of pediatric appendectomy (positive or negative) (91). The percentage of patients reporting barriers to post-acute rehabilitation services did not differ among regions with different levels of rehabilitation services in Quebec (136).  Differences in use corresponding to resource availability suggests the potential for inequities in access, however, without examining outcomes of care it is difficult to distinguish between cases of under- or over-use, or differences based on local resource supply that do not lead to poorer patient outcomes. Professional culture In the absence of other clear explanations for regional differences, physician “practice patterns” or professional culture have been cited as determinants, though not assessed directly (61,122,137). An early study of hysterectomy rates found that residents living in high-rate areas were more likely to have “hysterectomy-prone” physicians (138). Three studies found no association between specialist supply and use of specific procedures, but found that positive physician attitudes towards a form of care were predictive of higher rates of that care (24,51,54). The relationship between resource supply and enthusiasm for procedures was not examined in these studies. However, it may be that local practice context, and especially resource availability, shapes professional culture. In a study that surveyed family physicians directly, substantial variation in reported propensity to refer was found to be related to resources available (hospital beds, lab tests, admitting privileges), and practice style (including treatment policies and colleagues’ expectations about referral - which were themselves shaped by resources available) (139).  Outcomes examined  Only 16 studies (14%) examined variation in outcomes in relation to service use (Table 3-4). Mortality was the primary outcome (56,74,76,88,94,98,106,110,116,140–143) though in  37  some cases premature mortality was employed, counting only deaths before age 75 (76,123). A handful of studies also looked at hospitalization (or re-hospitalization) as an adverse patient outcome (74,88,98,124). Patient-reported outcome measures were examined in one study (80). The only study that reported on patient experience of care paradoxically found more positive assessments of care in rural settings, despite relatively lower physician supply than in urban areas (144). Only rarely do differences in use appear to correlate strongly with differences in outcomes measured, though this may reflect the fact that the outcomes examined aren’t sensitive to more subtle changes in morbidity. One study that examined patient reported outcomes for dental services did find better outcomes in communities with higher volume of services (80). Another notable exception is the finding of a threefold increase in risk of pre-hospital death in regions with limited access to a trauma centre (143). Otherwise, Kwon et al. found that while variation in treatment for gynecologic cancers exist (the ratio of highest to lowest regions in the rate of surgical staging by a gynecologic oncologist was 11.2, and adjuvant pelvic radiotherapy was 2.3) there was no significant regional difference in overall survival among Ontario health regions (142). Cohen et al. found that high use of specialists was associated with revascularizations but not with mortality (88). Similarly, Alter et al. found that hospital resources (on-site catheterization lab, revascularization facilities), and status as an academic centre were associated with higher 30-day revascularization rates, but accounted for only a modest amount of regional variation in acute myocardial infarction mortality outcomes (124). Variation in ICU length of stay for COPD patients (medians ranged from one to three days among hospitals) showed no relationship with mortality (145). Lower use of material resources and specialized services among rural residents did not translate into differences in mortality and morbidity among chronic disease cohorts (74).  Discussion In general, it appears that where Canadian researchers have looked for geographic variation in health care use, they’ve found it. While geographic variation may be of academic interest, determining the “right” level of use remains is a challenge. There is a tendency to assume that lower rates in some areas (if appropriate adjustment has been done) indicate overuse in higher-rate areas (146). But to determine if lower use or spending is more efficient as opposed to potentially compromising patient health, variation in service use must be connected to relevant measures of health outcomes. To this point in time, Canadian research has been  38  limited by reliance on mortality-based outcome measures which in most cases are too crude to detect the impact of subtly different patterns of health care services use. Wennberg’s typology of effective, preference, and supply-sensitive care (16) may be helpful in understanding why some procedures of less certain evidence, such as insertion of tympanostomy tubes (51), anti-reflux surgeries (52), knee replacements, and degenerative lumbar spinal surgery (54) showed such high variation. In contrast, variations in situations of more certainty, such as hip replacements following fracture, show less variation (59,60). Interpreting differences in types of services, especially the larger variation observed for specialist and hospital services is more challenging. One possibility is that variations across geography in hospital or specialist use occur because of substitution of one service provider or location for another. In this case if the substitute services were added together the apparent variation may be attenuated. Certainly, some degree of substitution between primary and specialist care, or ambulatory and acute care may be expected, depending on geographic context such as urban vs. rural areas. The difficulty this raises in interpreting variations is that the line between acceptable substitution and unacceptable waste or inequity is unclear.  For example, it may be that the greater average distance to hospital for individuals who live in rural areas leads to appropriate substitution of hospital care for community-based care, as patients are kept longer for monitoring and recuperation to avoid the need for a return visit (147). On the other hand, hospitalization in other situations may be a less than ideal substitute for the lack of primary care availability in the home community (148–150). Olatunde et al. found that emergency department, after-hours services, and hospital inpatient care were a significantly greater part of rural primary care physician practice (101), which supports the idea that rural emergency departments simply fill a different role in providing basic primary care than their urban counterparts (97). The fact that another study found high variation in a variety of service use, but not for reported unmet health care needs (149), also suggests the possibility of acceptable substitution. Only one study (134) sought to define health service environments across multiple dimensions of supply. This means that, at least in the Canadian context, it is not clear that all health-system or supply-side sources of variation can or should be eliminated, though up to this point. Still, up to this point, it has been common practice to adjust for measurable population characteristics, and then attribute remaining variation to “unwarranted” sources within the health system.   39  The relationship between resource supply and service use remains unclear, and our understanding is limited by the use of study units that may not reflect actual patterns of use. In general, while large differences in distribution if health care resources exist, smaller differences in use are typically observed (96,130,131,142). Since populations can move to access care, local availability of resources is just one factor that may shape access.  The choice of geography clearly makes a difference in the extent of variation observed. Choosing smaller units will typically reveal larger variations that tend to be muted in higher levels of aggregation. Another possible attenuating factor not discussed in any significant way in this literature is that the choice of geography itself. The boundaries of small areas established for administrative purposes, even health regions, may not reflect actual patterns of the use or delivery of health care services (151). The role of health regions in the provision of health services varies by province (22), and in many cases regions have only limited influence over delivery of hospital or physicians services. Patterns of use were not assessed directly in any of the reviewed literature. In contrast to the Dartmouth group that built units from empirical examination of service use, Canadian research tends to start with existing geographic boundaries without consideration of whether those are the most meaningful. Research examining differences across the urban-rural continuum is limited by potentially inappropriate definitions or artificial or arbitrary distinctions between urban and rural. In the literature examined above, how measures of rurality were related to health care provision was not usually described. Major cities identified in dichotomous measures may represent locations of tertiary care. Census-based definitions are more challenging to interpret in ways that have relevance to health care services use, as relatively small, isolated towns are often included as “urban” and may not differ much from adjacent rural areas in population or health system characteristics. In any case, such binary divisions likely mask important within-category variation. More careful consideration is needed of the mechanisms and pathways through which geographic context is related to health care, so that the effects of geography can be parsed more directly.  This review has clear limitations. The search has likely missed relevant papers due to the wide range of potentially relevant search terms. In particular, reliance on search terms “health care,” "health care." or "health service*” to narrow search results may mean that some studies that mentioned only the name of specific procedures were missed. Only Medline was searched. Though other databases, notably CINAHL, may have offered more complete  40  coverage, as the goal was to may the field of research, and not identify all potentially-relevant studies, Medline was deemed sufficient. A previous systematic review found a similar number of Canadian papers published from 2000 to March 2011 (1), but also included studies examining variation at both patient and physician levels. The search did not include gray literature, although an overview of provincial atlases confirmed similar findings to the academic literature, as well as similar limitations in the extent to which causes of variation are identified and outcomes examined (152–156). Finally, in interpreting this body of research, it is important to consider the potential for publication bias towards studies that found marked variation. I can only conclude that, in the studies examined, there is clear evidence that variation in health care exists.  In conclusion, marked geographic variation in the use of procedures, drugs, and types of services has been observed. However, the relationship between variations in service use and health outcomes is only rarely examined, and what research has been done is limited by crude mortality-based outcome measures. Isolating the impact of health care service use among the many factors that influence population health remains a methodological challenge. Approaches are also needed to summarize patterns of service use in a way that captures complexity within the health care system, in order to define meaningful units of analysis. While considerable attention has been devoted to describing variation, its causes remain unclear. The literature is replete with examples of variation crying out for explanation and, potentially, abatement, but policy directions or proposed solutions are rare. Many policies aimed at addressing variation in health care use target supply factors, despite the fact that the literature suggests that changing supply may not be the only (or best) way to influence ultimate goals of improved quality of care, health outcomes, and system efficiency. Finally, whether or not areas that have high use of individual procedures of types of services, have high use across board (i.e. whether there are places with systematically more intensive patterns of service use, and correspondingly higher total costs), remains unclear. More careful analysis is needed to unravel the complex causes of variation, and to determine the appropriate levels at which to target interventions.    41  Table 3-1 Terms used in the search strategy Concept Terms Health care use  health care or "health care" “healthcare” or "health service*” Variations examined variation* or differen* or disparit*  Comparisons on the basis of geography rural* or geography* or region* or area* or neighbo*rhood* Canadian studies Canad* or "British Columbia*" or Alberta* or Saskatchewan or Manitoba* or Ontari* or Quebec* or "New Brunswick" or "Nova Scotia*" or Prince Edward Island or Yukon or Northwest Territories or Nunavut   Figure 3-1 Scoping review search results   Full text review Title and abstract review Medline search (January  1, 1973-2013) 352 articles Retained: 182  Retained: 111 Excluded: 71 Excluded: 170  42  Table 3-2 Study characteristics Study characteristics N (%)  Data source(s)  Health Administrative data 77 (69%) Survey 26 (23%) Combination 8 (7%) Study population  Complete area population 68 (61%) Defined by health condition 43 (39%) Health care output examined (may be more than one)  Services defined by provider or location (e.g. GP contacts, hospitalizations) 68 (61%) Specific procedure(s) or drug(s) 50 (45%) Wait times or length of stay 13 (12%) Quality or appropriateness of service 6 (5%)      43  Table 3-3 Methods used Geographic units compared  Health region 40 (36%) Hospital/clinic catchment 8 (7%) County or regional municipality 12 (11%) Census 9 (8%) Researcher-defined sub-provincial regions 5 (5%) Province or sub-national regions 14 (13%) Urban/rural 23 (21%)   Examines rural/urban differences 38 (34%)   Method of describing variation (may use more than one)  Stratified measures (may also report ratios/differences) 104 (94%) Extremal Quotient (EQ), Coefficient of Variation (CV), Systematic Component of Variation (SCV)* 16 (14%) Multilevel models (partitioning individual- and area-level variation) 4 (4%)   Methods of adjustment  Crude only 35 (32%) Age/sex standardization 36 (32%) Multivariate 40 (36%) *Extremal Quotient (EQ) is the ratio of highest to lowest values. The Coefficient of Variation (CV) is the ratio of the standard deviation to the mean. The Systematic Component of Variation (SCV) adjusts for variability within areas, based on the number of observed observations relative to the number expected (given age/sex distribution). More detail is provided by Ibanez et al. (45)  44  Table 3-4  Examination of causes of variation and outcomes  N (%) ) Examines any population causes 62 (55%) Population driver(s) examined (of those that examined any, may be more than one): Health status 23 (38%) Socioeconomic status 40 (64%) Race, ethnicity, visible minority, or new immigrant status 17 (27%)  Beliefs or values 7 (11%) Examines health system causes 45 (39%) Health system driver(s) examined (of those that examined any, may be more than one): Health human resources 23 (51%) Material resources of health care facilities 18 (40%) Professional cultures 5 (11%) Examines health outcomes 16 (14%)     45  Chapter 4 - Data sources and measures Overview  To meet my research objectives I used a combination of data on individual patients, providers, and hospitals, accessed through the BC Ministry of Health and Population Data BC, as well as area-level sociodemographic and health system data that are publicly available. In this chapter I describe each of the data sources used, and outline how health care costs were measured and modeled statistically. As measurement of health system outcomes remains a central challenge for this research, in the later part of this chapter I consider the strengths and weaknesses of available options.  Data sources Microdata accessed through Population Data BC  This research uses health administrative data from British Columbia for April 1 2008 through to March 31, 2011, maintained by the BC Ministry of Health and accessed through Population Data BC. The study population was drawn from all BC residents who were enrolled in BC’s Medical Services Plan in that time period. The following sources of data were linked. Client registry (consolidation file)  Population Data BC maintains a central file with data on all individuals who are eligible to receive health services, and/or actually receive services in BC (157). It is a comprehensive data source including almost the entire population of the province in any given year. The data are regularly cleaned and checked for errors by the programmers at Population Data BC. I used this file data to obtain sex, age, neighbourhood income decile, and Local Health Area (LHA) of residence. The point in the year registration with MSP started, and total days registered were also recorded. MSP payment information file (physician payments)  The Medical Services Plan (MSP) payment data captures all fee-for-service payment records for physicians in the province (158). Each record includes the date of service, the fee item indicating the service provided, and diagnostic codes for each visit (ICD-9 codes). An important limitation of the physicians’ payment data is that they do not include detailed information on the use of services provided by physicians paid using alternate (non-fee-for-service) payment plans (APPs).   46  Discharge abstracts database (hospital separations)  Any patient who is admitted to hospital for either inpatient care of same-day surgical care is entered into the hospital separation records at the time of their discharge or death (159). The hospital data provide information on each patient’s hospital stay, including the project-specific study ID of the patient, the level of care (which differentiates inpatient stays from surgical day care), dates of admission and separation, the responsible physician, codes for patient diagnoses (ICD-10) and any procedures provided, and the resource intensity weight indicating the relative service intensity of the stay, which is used for costing hospital stays.  Vital Statistics death file  BC vital statistics death data contain information on the year and underlying cause of death (ICD-10) for all deaths occurring in the province, or among BC residents. Hospital code was also obtained and made it possible to identify the location of death when accessing hospital care (and determine when this was outside the geographic area of residence).  Microdata access and linkage  Data from Population Data BC are available to researchers who meet access criteria, including ethics review, peer review, and review by all relevant data stewards. Peer review for this thesis was deemed met by review by the UBC School of Population and Public Health Thesis Screening Panel, and my supervisory committee. I obtained ethics approval from the Behavioural Research Ethics Board at the University of British Columbia, and I obtained permission to obtain extracted anonymized data to be published in this dissertation from data stewards at the BC Ministry of Health.   Data were linked across these datasets using the individual-specific Personal Health Number assigned to all permanent residents of BC. Population Data BC uses this identification number to link individuals across datasets and then replaced it with a project-specific anonymized study identification number, which I used for analysis. Physicians were similarly linked using their practitioner number, replaced with a study ID. Publicly available data  Publicly available data describing population and health system characteristics at the level of Local Health Area (LHAs) were obtained from the following resources:  47  Health System Matrix  A critical source of data for complete health services use data is the Health System Matrix produced by the BC Ministry of Health. This contains data from all of the above-listed sources, as well as PharmaNet (data on prescriptions dispensed from community pharmacies), and Home and Community Care (residential care, assisted living, home care and support, community support, and adult day programs), aggregated across LHAs and population categories. Services provided by the Provincial Health Services Authority, most notably the BC Cancer and Renal Agencies, are not included, nor are some Emergency Department services, Community Health and Substance Use Services, and Population Health and Wellness services. Data on these services omitted in the Health System Matrix are also not currently available through Population Data BC. BC Stats socio-economic profiles  BC Stats maintains data on population, demographic, and socioeconomic characteristics and are publicly available at http://www.bcstats.gov.bc.ca/statisticsbysubject/SocialStatistics/SocioEconomicProfilesIndices/Profiles.aspx. The primary data source for the variables used is Statistics Canada’s 2006 Canadian census. BC Stats population projections were used in calculation of rates for non-census years. Cost and outcome variables Calculating cost of health care services  Fee-for-service physician payment records from the MSP payment information file include exact paid amounts for all services. All paid amounts were adjusted to 2010/11 constant dollars using a specialty-specific fee index built and maintained by the BC Medical Association and used by the Canadian Institute for Health Information. There are no differences in the amount paid for individual fee items across the province.  All hospital discharges records include a Resource Intensity Weight (RIW) that corresponds to the average expected resource use given the diagnoses, any procedures performed, and the age and sex of the patient. RIWs were converted to dollars using a standard cost per weighted case provided by the BC Ministry of Health, and price-adjusted by using the same (2010/11) amount for every year of data.  48   Costs were then summed for each patient and year, to yield an annual patient-level total expenditure, as well as seven sub-totals:  General and family practitioner services  Medical specialists (dermatologists, neurologists, psychiatrists, neuropsychiatrists, pediatricians, pediatric cardiologists, and specialists in internal medicine, physical medicine, emergency medicine or osteopathy)  Surgical specialists (obstetrician/gynecologists, ophthalmologists, otolaryngologists, general surgeons, neurosurgeons, orthopedic surgeons, plastic surgeons, thoracic and cardiovascular surgeons, urologists and anesthesiologists)  Laboratory specialists (pathologists and medical microbiologists)  Imaging specialists (radiologists and nuclear medicine specialists)  Acute inpatient care  Day surgery Modeling costs  The cost outcome variable was not normally distributed. To account for this, I used a generalized linear model (GLM) framework to identify an appropriate model. GLMs directly model both the mean and variance functions on the original scale of the outcome variable. The GLM framework includes a link function (μ) that relates the observed y to the linear predictor (x’ β). The functional form was determined using the Box-Cox Test. Results supported selection of log link (which means covariates act multiplicatively on the mean). A distribution within the linear exponential family specifies the relationship between the variance and the mean, with variance functions of the form:  v(x) = κ(µ(x’β))λ  The Park test was then used to determine the relationship between the mean and variance (160). This regresses the natural logarithm of squared residuals on the natural logarithm of predicted vales and a constant. The coefficient on the predicted values gives the family. Several distributional families are possible:  – Gaussian: constant variance; λ=0 – Poisson: variance proportional to the mean; λ =1  – Gamma: variance proportional to the square of the mean; λ =2  – Inverse Gaussian: variance proportional to cube of the mean λ =3  49   Results supported either a Poisson or Gamma distribution. Poisson had the highest correlation between the observed response and model predicted values (161). However, there was some evidence of over-dispersion, and the potential for correlation within units of analyses. A robust variance estimator was used, that accounts for clustering within geographic units, thereby relaxing the assumption of independence among observations.  Ideally, random intercepts for geographic units would have been included in the linear predictor. Modeling clustering within geographic units would have provided unbiased estimates of coefficient standard errors, and allow for examination of sources of variations statistically, with the potential to simultaneously model variables at different levels. However, given the very large sample size and relatively small number of geographic units, random-intercept models could not be estimated in standard statistical packages (SAS, Stata, and R).  Regression modeling followed a blocked approach (Table 4-1), first estimating models with area-level units only (HSDAs and networks), then examining drivers of “warranted” variation or markers of patient need (age, sex, and health status (ADG)), and then differences in the health service environment. I used these models to predict expected individual cost based on chosen covariates. I then aggregated predicted individual costs and observed costs and calculated ratios of observed to expected costs for each unit examined. Measuring patient need or health status  In analysis exploring causes of variation (Chapter 7), health status was measured using Aggregated Diagnosis Group (ADG) indicator variables generated by the Johns Hopkins Adjusted Clinical Group Case-Mix System (162). This uses diagnoses recorded in encounters with physicians and hospitals over a one-year period to assign patients to groups with similar expectations of health care use (such as follow-up visits or referral to a specialist). By aggregating diagnostic codes and grouping them into categories (for example, time limited minor or major, chronic medical unstable or stable) ADGs reflect groups similar in terms of severity and persistence of health conditions, not simply volume of health service use. This coding system has been validated for use in British Columbia (163). ADGs combined with patient age and sex were used in adjustment, as this has been found to be more predictive of service use than Aggregated Clinical Groups (ACGs) alone (163,164).  Choice of outcome measures  As has been established previously, it is necessary to connect service use to relevant outcomes in order to determine if differences in use or costs represent issues of inequity, or opportunities for efficiency gains. My review of Canadian literature described in Chapter 2, as  50  well as an earlier international review (1) highlight the need for better assessment of the outcomes of health care service use, as this was only rarely done in reviewed literature.  While the use of health services is captured in routinely collected health administrative data, corresponding measures of its effect on the health of individuals or populations are harder to define. Survey data can offer a rich array of potential outcome indicators, but at present in the BC context it is not possible to link these data at individual level to actual use of health care services. In addition, small numbers, even for large surveys like the Canadian Community Health Survey, limit ability of survey data to capture small-area variation.   Mortality data has the benefit that it is routinely collected and reflects the health status of the entire population. Its chief limitation is that many factors beyond the health system influence mortality rates. It is also relatively rare, and irrelevant for many types of services, where the goal of treatment is reduced morbidity, not changes in mortality (7).  There has been resurgence in interest in the concept of amenable or avoidable mortality (165,166), defined as deaths from causes that are theoretically preventable with timely access to quality health care. Classifying a condition as amenable to health care is based on a judgment that once the condition has developed, treatment is available that can be reasonably expected to prevent death (165). This renewed interest in amenable mortality as a health care system performance indicator likely comes because this capitalizes on existing mortality data while appearing to correct for its known limitations.  This concept was first put forth in 1976 by Rutstein et al. (167), who consulted with an expert panel to compile a list of health conditions from which deaths were considered “untimely and unnecessary.” They describe each death as “a warning signal, a sentinel health event, [indicating] that the quality of care might need to be improved.” Charlton et al. (168) were the first to apply this concept at the population level, selecting 14 disease groups from Rutstein’s list for which mortality in a high-income country should be avoidable. Since then, studies have modified the list of health conditions to reflect advances in health care, increased the upper age limit for deaths to reflect improved life expectancy; and in some cases extended the concept to include conditions preventable by public health interventions (which may or may not also be amenable to health care).  In its 2012 Health Indicators report, the Canadian Institute for Health Information (CIHI) put forth definitions of “treatable” and “preventable” mortality, developed by a panel of experts  51  for use in the Canadian context (169). They first subset deaths occurring under age 75, classifying these as premature. The choice of 75 as the age cut-off is of course essentially arbitrary and does not imply that some deaths among those 75 years and over could not also be avoided. Of premature deaths, CIHI identified causes of death that could have been potentially avoided through provision of timely and effective health care, and classified these as “treatable.” Deaths that could have been potentially avoided through prevention practices and public health policies are considered “preventable.” The remaining subset of premature deaths is classified as unavoidable.  “Treatable” mortality has been put forth as an outcome measure that more directly captures health system performance than all cause mortality (170). However, earlier measures of amenable mortality (on which the CIHI definition was based), are highly correlated with both premature and all-cause mortality, and lower numbers mean rates are less stable among small areas (20). For these reasons, in addition to treatable mortality, premature and all-cause mortality will also be retained as outcome measures in this thesis (Chapter 7), though statistical control for individual underlying health status and sociodemographic characteristics will be critical in this analysis (7). 52  Table 4-1 Blocked modeling approach Model Healthcare costs Mortality outcomes 1 Age, sex Age, sex 2 Age, sex, ADG Age, sex, income 3 Age, sex, ADG, health service environment (metro, non-metro (a), non-metro (b), remote) Age, sex, income, census variables: % no high school, % recent immigrant, % aboriginal, 4 All covariates All covariates       53  Chapter 5 - Defining multispecialty physician networks Background  Canadian research has used many different units of analysis to compare health care use across geographic areas (Chapter 3). Health regions are commonly used, as is census geography, or larger units of provinces or regions. Some studies make comparisons by hospital catchment, though these are typically limited to patients hospitalized within the study period, rather than the whole population potentially served by the hospital. Throughout this research, the inability to define precisely the service provider group responsible for observed patterns of care, either because multiple hospitals and physician groups are located within units of analysis, or because patients cross unit boundaries to access care (9,171) remains a persistent limitation. If chosen units do not align with patterns of care, the magnitude of variation in use or costs may be masked; differences of interest will be unseen, not because they do not exist but purely as a result of the level of aggregation chosen for reporting.  Health planning geography is typically used to make comparisons in BC (153), as well as in national health indicator reporting (22). In national comparative research, the province’s 16 Health Services Delivery areas (HSDAs) are considered “health regions.” This unit of analysis is justified on the grounds that health regions are responsible for providing health services to residents (5). In fact, few services are directly planned (or delivered) by HSDAs, and there are no structures that make these sub-units of BC’s five geographic Health Authorities (HAs) accountable for care received. Physicians billing fee-for-service are paid directly by the province, and neither HAs nor HSDAs have direct control over physician activities. Hospital services are planned by the five regional HAs (in which the HSDAs are nested) and by one province-wide authority responsible for higher-level acute care services and special services such as the BC Centre for Disease Control and cancer care centres.   Health Authorities are thus the true administrative boundary for the health care system. However, HAs encompass large regions, and important differences may be missed in analysis at this level. The sub-units available are HSDAs (n=16) and Local Health Areas (LHAs, n=89) both of which are nested within HAs (Figure 5-1). LHAs enable the examination of small-area differences. While initially conceived of as key planning units, over time planning has grown more centralized, and whether LHAS or HSDAs reflect current patterns of needs or use is unknown.  54   In short, the boundaries of small areas established but no longer explicitly used for administrative purposes within the health care system may not reflect actual patterns of the use or delivery of health care services (151). Moreover, while most individuals access primary care services close to home, it is not uncommon that patients travel to larger centres for more specialized services (94). Distances travelled vary among urban, rural, and remote areas, but this is not directly accounted for in any of the available administrative units.  A large body of American health services research uses the analytic unit of Hospital Referral Regions (HRRs), developed by analysts at Dartmouth (9). These 306 geographic units are intended to represent regional health care markets for tertiary medical care, based on where US Medicare beneficiaries were referred for major cardiovascular surgical procedures and neurosurgery. A recent Institute of Medicine report observed that health care decision making occurs at the level of individual practitioner or organization, and found substantial variation in units within HRRs (including hospital service area, hospital, practice, and individual provider) (17). Originally based on 1992/93 discharge data, the 306 identified HRRs remain unchanged (18). Keeping boundaries stable makes temporal comparisons straightforward, but secular changes in hospital catchment may mean they are now less accurate representations of patterns of health care seeking and referral. HRRs also have the limitation that they focus only on tertiary services, and use data only from the regular Medicare population (a subset of the population aged 65+ who are not enrolled in Medicare Advantage, health care plans offered by private companies contracting with Medicare).   In response to the limits of HRRs, researchers in the US developed an approach to define “multispecialty physician networks” based on observed patterns of primary and specialty physician services use, as well as acute care (171–174). A similar approach was more recently applied in Ontario (21), and is adapted here for the BC health care system context and available health administrative data. However, the objective of previous analysis was not to describe variation in service use, but rather to define units that might form the basis of accountable care structures. In previous analyses developing and using multispecialty physician networks, the focus was on comparing health care resources, not use or costs of care. Formal multispecialty physician networks do not currently exist in BC. This analysis works from the premise that informal networks of patients, care providers, and hospitals exist, and can be revealed using health administrative data capturing service use. These informal networks behave as “self-organizing” systems, developing through referral patterns, information- 55  sharing, and patient travel patterns (21). If shared practice style develops as a result of influential leaders (23), attraction and retention (24), self-selection (7), or mutual adaptation to the shared contexts of organizations, systems, and resources available (26), naturally-existing networks may reflect this most closely. As such, they represent a unit of analysis that has the potential to reveal variations, if such variations exist. The intention is to define patient populations for which primary care physicians, specialists, and hospitals work together to provide care, as this may be the optimal unit of analysis to compare patterns of use and outcomes.  Consistent with previous research, we use acute care hospitals as a central feature of networks, linking patients and physicians to individual hospitals, and then aggregating hospitals together (as necessary, using rules described in detail below). The choice to work from hospitals reflects both practical and theoretical considerations. The goal is to develop relatively self-contained networks that provide a reasonably comprehensive range of health care services. It is natural, then, for such units to contain an acute care facility. Importantly, while networks are based around hospitals, the study population is not limited to those hospitalized in a given year, as individuals are linked to the hospital where they would likely be admitted, based on the patterns of other patients who share their ambulatory care provider.  The goal of analysis in this chapter is to determine if we can define self-contained networks of patients, physicians, and hospitals based on patterns of service use. We then examine the extent to which identified networks correspond to existing health regions. In BC it is not generally possible for researchers to access data at levels of aggregation below HSDAs or sometimes LHAs, and so findings may be of use in future research. Identified networks are subsequently compared to health regions in Chapter 7 with respect to descriptive and analytic power for understanding health care services use and costs. Approach Data sources and study population  This research used administrative data from British Columbia for April 1 2008 through to March 31, 2011, accessed through Population Data BC. Two sources of data were linked using project-specific unique identifiers for all patients and physicians as well as unique (unencrypted) hospital numbers.  56  MSP payment information file (physician payments)  The physician payment database in Population Data BC captures fee‐for service payment records for all physicians in the province who receive any income in the form of fees for clinical services (158). Each record includes the date of service, a fee item indicating the service provided, the specialty code for the physician billing the services, the location in which the service was provided (e.g., practitioner’s office, residential care, hospital) and the study ID of the patient receiving the service. Discharge Abstracts Database (hospital separations)  Hospital separations data provide information on each hospital stay, including the project‐specific study ID of the patient, the level of care (which differentiates inpatient stays from surgical day care, and patients awaiting an alternate level of care), dates of admission and separation, the responsible physician, physicians responsible for procedures or anesthesia, codes for patient diagnoses (ICD-10), and codes for any procedures performed on the patient during the hospital stay (159).  Eligible residents  The study population includes all BC residents who were continuously registered in the provincial health insurance program over the three-year study period (registered for >75% of possible days), and used physician or hospital services at any point. Individuals who were born or died during the study period were included if they were continuously registered up to death, or following birth. Eligible physicians  All physicians who provided services to patients paid for via fee-for-service, or who were listed in the hospital Discharge Abstract Database (DAD), were included in analysis. Physician specialty (primary care or specialist) was identified by functional specialty, as evidenced by the specialty code billed most frequently in the MSP payment data. In the case that a physician appeared only in hospital data, specialty was determined based on practice specialty most frequently recorded in discharge records. Eligible hospitals  All acute care hospitals in the province of BC were included. Rehabilitation and chronic care facilities, sometimes coded as hospitals, were excluded as these do not provide acute care. The BC Cancer Agency and psychiatric hospital were also excluded, as these facilities  57  treat patients from across the province, and populations defined around them would represent only patients with relevant conditions. Process for developing hospital networks Linkage of residents to usual providers of primary care  There is no formal rostering of patients in BC. There is also no policy support for group- or team-based practices, though these do exist in some places. For this analysis, patients were ‘assigned’ to the individual primary care physician responsible for the plurality of their primary care over the study period. This was measured by fee-adjusted dollars billed within general practice service codes, summed over the study period (21,175). In the case of a tie, the patients were assigned to the primary care provider with the most recent visit. Any remaining unlinked residents were assigned to the physician (primary care or specialist) who provided the highest total volume of ambulatory physician services. This was measured by dollars billed for visits, laboratory tests, and diagnostic tests provided in locations other than emergency departments, inpatient hospital or day surgery (determined by service location code). Services provided in outpatient and/or ambulatory clinics located in hospitals were included. Residents who had no ambulatory contact with a physician were not linked to a usual provider of primary care, but could be linked to a network if they had hospital service use. Linkage of admitted residents to hospitals  All non-maternal medical admissions were examined. Maternal admissions were identified using the CIHI Case Mix Group (CMG) codes for vaginal and caesarian delivery and neonates. Medical admissions included all other diagnostic CMG codes. The rationale for excluding surgery and maternal admissions was that these services are more often regionalized, and are less likely to correspond to physician referral networks (21). Information on patient admissions to hospital is used in the following step to link primary care physicians to hospitals. For residents admitted to more than one hospital, each admission was assigned a weight, so that all admissions sum to one. This was done so that individuals with multiple hospital admissions would not bias assignment of primary care physicians. Linkage of physicians to hospitals  Each specialist was linked to the acute care hospital where he or she provided the highest volume of services. If a physician was designated in the “most responsible,” “intervention provider,” or “anesthetist” fields of the DAD (in this case including maternal and surgical admissions), this was counted as services for one hospital separation. We also  58  examined MSP billing records to identify physicians who billed for the care of patients during periods of hospitalization recorded in the DAD. Any inpatient services billed during a hospital stay were counted as one hospital separation. If a physician was both recorded in hospital data, and also had MSP billings corresponding to the same patient during the hospital stay, this was counted as care for just one hospitalization.  As only a subset of primary care physicians provide services in hospitals (176), primary care physicians were linked to hospitals on the basis of where their patients were admitted. Each primary care physician was linked to the hospital where the plurality of the patients for whom the physician was the usual provider of care was admitted. As is described above, only non-maternal, medical admissions were counted, and for patients admitted to multiple hospitals, visits were weighted weights to sum to one. A few primary care physicians whose assigned patients were not hospitalized in the study period were assigned to the hospital used most frequently by all unique patients seen, not just those assigned to that physician. Physicians with no hospitalized patients who did not appear in the DAD could not be assigned. Final linkage of residents to hospitals  Each BC resident was assigned to the hospital where his or her usual provider of care was linked. Any unlinked residents (those who had no physician services use) who used hospital services were assigned directly to that hospital. Obviously, residents who used neither physician nor hospital services could not be assigned based on service use data. For later analysis, these individuals were assigned to a network based on where the majority of other individuals in their LHA of residence were assigned.  Individual hospital networks include all physicians and residents linked to each individual hospital. In other words, there is a network of residents and physicians for each acute care hospital in the province, and each resident and physician is linked to only one hospital. Aggregation of hospital networks  Individual networks were aggregated, with the goal that all final networks would include a minimum of 50,000 patients. This number is essentially arbitrary, but has been suggested as a sufficient size to provide a range of primary and specialist health care services (21). We use three ”loyalty” indices developed by Stukel et al. (21) to measure the extent to which hospital networks are self-contained:  59   Acute care admission loyalty: the percentage of total non-maternal medical admissions for all patients linked to the network occurring at the network hospital and at each other hospital  Primary care loyalty: the percentage of primary care visits (service codes 1-9) by patients in the network, to primary care physicians linked to the same network and to primary care physicians linked to each other network  Specialty physician loyalty: the percentage of other ambulatory visits (for specialist services) by patients in the network that were to specialist physicians linked to the same network, and to specialist physicians linked to each other network  The above loyalty indices were calculated for each network to all other networks, and the top four on each index retained for further examination. In the example shown in Table 5-1, Network A has fewer than 50,000 residents (and few specialists), and has correspondingly moderate admission and specialty physician loyalty. Patients are frequently admitted to larger nearby hospitals, with Network B ranked most highly. Networks A and B were therefore linked to form a network. Whether or not Networks C and D were also linked would be determined on the basis of similar examination of loyalty indices calculated for each of those networks (i.e. Table 1 is repeated for each network with less than 50,000 patients).  Some remote networks with fewer that 50,000 patients could not be linked to larger networks within five hours travel time. Patients in these areas with needs that cannot be met locally are typically transported by air to tertiary care centres. These are labeled satellite networks.  Finally, in urban areas with multiple hospitals in close proximity, we expect considerable overlap in service use across hospitals. However, if each individual hospital has more than 50,000 patients, these would not normally be linked. To capture service use adequately in these situations, I combined hospital networks to ensure that all final networks had a minimum acute care loyalty of 60%. Comparison of costs across networks  For the purpose of comparing costs across areas, 268,371 individuals with no service use who were not assigned to a network directly, were assigned based on their LHA of residence. In this way, all networks contained the complete population of users and non-users.  60   I calculated total costs of hospital and physician services for each patient using fee-for-service billing data, and hospital discharge data. I used Poisson GLM models to predict expected costs for all patients based on age, sex, and Aggregated Diagnosis Groups (ADGs), as a measure of health status. More detail on calculation of costs, model selection and fit is presented in Chapter 4. Observed and model-predicted expected costs are compared across all networks. Results Study population and linkage characteristics Patients  The initial study population included 4,272,094 individuals who were continuously enrolled in MSP during the study period. Of these, 259,139 (6%) used neither physician nor hospital services during the study period, and so could not be linked based on service use. An additional 19,806 (0.4%) used some physician services, but could not be assigned a physician, or were assigned to a physician with no hospitalized patients or hospital services, and so could not be linked to a hospital (Table 5-2). Of those not included in the network development, a greater percentage was male, and very small percentages were born or died during the study period (Table 5-2). Those not included were also slightly younger than those who were. These individuals were subsequently assigned to the network where the majority of residents of their LHA were also assigned.  With respect to physicians receiving income through non-fee-for-service avenues, 717,213 (18%) of the linked patients received some services from a non fee-for-service primary care physician. While this means data on their primary care service use is incomplete, these patients were retained because the data on their physician services paid for via fee-for-service and hospital use could still contribute useful information for network development. Some of these patients were excluded in cost models, as is described in Chapter 7.  Only 2.5% of those included had no primary care visits, and 8.5% had no ambulatory care visits (Table 5-3). Roughly one third of patients had between 1 and 6 primary care visits, and visits for other ambulatory physician services, over the three-year study period. Most patients had no non-maternal medical hospitalizations (91.6%). Very few patients (<1%) were hospitalized at more than two unique hospitals.   61   Of the 3,993,149 linked patients, 3,892,257 (97.5%) were assigned to a network based on their assigned primary care provider. An additional 98,172 (2.5%) were assigned based on other ambulatory service use. The remaining 2,720 (<0.1%) were assigned based on their own hospital service use, rather than the hospital most commonly used by all patients of an assigned physician. Physicians  13,865 unique physician IDs appear in the data. 55.7% were in both physician MSP and hospital data, 20.2% in MSP only, and 24.0% in hospital only2 (Table 5-4). The total number includes physicians registered with the BC College of Physicians and Surgeons and also physicians registered in other provinces, who were paid through MSP or who treated BC patients in out-of-province hospitals. The highest percentage (44.8%) were general practice physicians, while 28.2% were medical specialists, and 22.9% were surgical specialists (Table 5-4).  Of the GPs, 4,759 (76.6%) were assigned to the hospital where most of their usual patients were hospitalized (i.e. the plurality of weighted hospitalizations). An additional 909 (14.6%) were not the usual provider of care to any hospitalized patients, and so were assigned to the hospital used most frequently by all unique patients seen over the study period. 544 (8.8%) appeared in hospital data only, and were assigned based on hospital activity. Of specialists, 7,117 (93.0%) were assigned to the hospital where they had the highest volume of activity (based on DAD and/or MSP data). An additional 536 (7.0%) specialists could not be associated with hospitalizations, and were assigned to the hospital where all unique patients they’d seen were most frequently hospitalized, as for primary care physicians. 50 GPs and 36 specialists could not be linked to a hospital. These were observed in MSP only and none of their assigned UPC patients were hospitalized during the study period. In total, 2,845 (20.5%) physicians were linked to an out-of-province hospital and so were not included in subsequent analyses. Hospitals  79 BC hospitals were used in analysis. Table 5-5 presents the characteristics of hospital networks grouped by the number of assigned patients. On average, more physicians were assigned to larger hospitals, and a larger proportion of them were specialists. It should be noted                                                2 These may include salaried hospital physicians or physicians paid by other alternate payment plans, but also any out-of-province physicians who cared for BC patients in hospital.  62  that fewer than 15,000 patients were assigned to BC Children’s as their regular source of care, and fewer than 50,000 to BC Women’s. However, as these are also tertiary centres for populations not linked directly to the hospital, a large number of physicians were linked. This is what lies behind the wide range of assigned physicians in the categories with ≤15,000 population and 15,001-50,000 population.  Smaller hospital networks were generally less self-contained than larger hospitals on all loyalty measures, reflecting the fact that patients must go elsewhere to access a full range of services. The largest hospital networks were also less self-contained for acute hospitalizations, reflecting the fact that these tend to be in urban areas, with several other hospitals nearby.   Through the linkage process described above, 79 individual hospitals networks were combined into 23 final networks and 6 “satellite” networks. A detailed list of all networks and their characteristics is provided in Appendix 4. Table 5-6 shows final network characteristics stratified by the size of the largest linked hospital or “hub” (defined as the number of linked patients). Five individual hospitals and one four-hospital network had fewer than 50,000 patients, but would next be linked to large hospitals more than five hours away. These are included as satellites in Table 5-6, and are coloured blue in Figure 5-1. Dashed lines in the figure indicate the larger hospital networks where satellite residents next most commonly use services. These hospital networks include as few as 117 patients and 5 physicians; for these, loyalty measures are low and cost estimates are unreliable.  Linked networks all contain GPs and medical and surgical specialists, though not all had lab or imaging specialists. The median percentage of specialist physicians ranged from 45% in the networks surrounding large hospitals to 36% in the networks surrounding small ones. This is not surprising, as the large hospitals were in urban centres, and the specialist physicians located there may serve patients across the province.  Following linkage of individual hospitals to networks, specialty, and acute care loyalty measures increased for all hospitals (Table 5-6). For the smallest hospitals, primary care loyalty fell slightly following linkage, but increased for all others. This likely reflects the fact that smaller hospitals are relatively remote, and it makes sense that there is little overlap in primary care services. Increases were especially marked for the large hospitals, reflecting the high degree of overlap among the linked facilities. Loyalty measures were highest for networks with small hub hospitals, likely reflecting the fact that these tended to be in rural areas, and services outside of the network would be less commonly accessed.  63   Identified networks were completely contained within the five regional health authorities, with the exceptions of Masset and Queen Charlotte Islands, remote satellite hospitals in the Northern Health Authority which were more closely linked to Vancouver than to Prince George (the largest hospital in the Northern Region). However, the population size was small, and this relationship was based on relatively few services. Alignment of the networks with Health Service Delivery Areas was moderate (Figure 5-2). While a few linked hospitals crossed boundaries (for example, Bulkley Valley, Shuswap Lake, and Queen Victoria), on average 85% of the population linked to a network lived within the HSDA where the network hub was located. So contrary to our initial hypotheses, HSDAs appear to correspond fairly closely to observed patterns of service use.  Per-capita costs for hospital and physician services (based on average values over the three years, and truncated at the 99th percentile) ranged from $889 to $1,477 among linked networks (Table 5-6). However, predicted costs also varied widely, reflecting markedly different demographics and health status across network populations. Differences between observed and expected ranged only from $111 per-capita less than expected to $88 per capita more (excluding the satellite networks, for which cost estimates are unreliable).  On average, small and medium hospital networks had slightly higher than expected costs, and large networks slightly lower. This may be explained by the fact that hospital services formed a higher proportion of total spending for small and medium network patients. Overall, variation is remarkably limited, and the plot of observed and expected costs is near linear (rho=0.86) (Figure 5-3). Discussion  Identified networks were remarkably self-contained, and show promise as units of analysis with multiple potential applications. Of course we do not expect all linked physicians consult directly with one another, or with the same patients, but they share material and human resources (hospital facilities, specialist expertise). We also hypothesized that if differences in professional culture exist within the province, it would make sense these develop within institutions, and that physicians connected in this way may be more similar in practice style.   The face validity of network assignment is high, and the fact that loyalty was generally highest for primary care, followed by acute care, and finally other ambulatory physician services is consistent with the fact that primary care is most localized, and specialty physician resources may be shared across large areas. Analysis also demonstrates that patterns of use do not  64  necessarily correspond to existing organizational structures. For example, St. Paul’s and Mount Saint Joseph Hospital are both administered by the non-profit organization Providence Health Care, but Mount Saint Joseph was more closely linked to Vancouver General Hospital than to St. Paul’s.  Contrary to expectations, networks were largely nested within existing health administrative units (HSDAs). This may reflect the planning process at the time HSDAs were established. It was recognized that many of the 52 (then) existing Regional Health Authorities did not have the population or budget to support the full range of services needed by residents. HSDAs were introduced as sub-units of the five Health Authorities, taking into account provincial geography, such as road networks that would shape referral patterns (177). Regardless of the explicit process undertaken then, the present analysis reveals that the HSDA boundaries are largely consistent with data-revealed patterns of networked care. Though networks offer a slightly finer level of aggregation (as multiple networks (n=29) are nested within some HSDAs (n=16), our findings support the utility of HSDAs in future analyses. This may be particularly useful as data can often only be accessed at HSDA-level.   Though adjusted per-capita costs ranged widely across networks, differences were largely explained by patient demographics and health status. There is reason to believe that remaining variation is related to different patterns of use of hospital and specialist services, which in turn are likely shaped by the geographic context in which different networks are located. This is not the focus of the present chapter, but will be examined in greater detail in Chapter 7. Still, the current findings suggest that the magnitude of geographic variation in health spending is much smaller in BC than was found in US research using similar methods (171).  This analysis has some limitations that warrant discussion. It does not include home and community care services. However, unlike physician and hospital services, these are organized and delivered by Health Authorities, and patterns are therefore much more likely to correspond to existing health system administrative units. The fact that this analysis used complete data on hospital and fee-for-service physician services still represents an enormous improvement over previous studies that used data on subsets of US Medicare beneficiaries (171–174). While we were missing data on payments to physicians via alternate payment plans, we could still assign patients and physicians based on hospital data, and do not believe that network assignment would be different if APP data were available.  65   We lacked information on college registration of physicians, and some out-of-province physicians may have been linked to BC hospitals. The number of such physicians would be small, and would not affect final network assignment of BC patients. Similarly, over the course of the three-year period, some patients and providers would have moved location, and some patients would have changed primary care providers. While this adds noise to the data, it would not change basic network patterns. The Ontario study on which this analysis is based found similar networks using two different three-year periods of data (21).   Using hospitals as the basis for defining networks is potentially problematic if physician attachment to hospitals is non-exclusive and overlapping (173). However, by combining the individual hospital networks, overall loyalty measures increased markedly. In all but one case, small hospitals linked unambiguously to larger hubs – either regional centres, or in the case of satellites, major tertiary centres at greater distance. In one case did we find a “chaining” of hospitals, where the second most highly ranked hospital for Port Hardy was Port McNeil, but the second ranked for Port McNeil was Campbell River. Given that these hospitals are along a single highway with Port Hardy at the end, these relationships make sense, and there was sufficient overlap between Port Hardy and Campbell River that once linked, overall loyalty measures still increased.  The choice to define networks around hospitals also has advantages. Hospitals are major resources in any local care system, and while hospital care spending as a proportion of all health care spending has been falling for decades, it still represents the largest single component. In order to provide seamless care, and avoid unnecessary admissions or readmissions, health professionals caring for patients in the community must work in collaboration with hospital-based providers. At the same time, the fact that non-hospitalized patients are included in these networks provides a denominator that reflects all patients in a given area, and allows population-based analysis of performance indicators not solely based on acute hospitalizations (171).  The choice to set minimum network size at 50,000 patients is essentially arbitrary, and does have a clear effect on the final networks. Based on this criterion alone, if a hospital had more than 50,000 patients, unless a smaller hospital network linked to it, it remained an individual network, even if there was considerable overlap with another larger hospital. The additional requirement of 60% of acute care visits was helpful in these situations. This meant that St. Paul’s and Burnaby linked in with Vancouver General, Eagle Ridge linked to Royal  66  Columbian, and Victoria General linked to Royal Jubilee. If a larger minimum population size is deemed more appropriate for another purpose, the same approach and loyalty metrics could be used to further link identified networks.  Previous studies seeking to identify provider networks have framed their objectives in relation to identifying potential Accountable Care Organizations (ACOs) (21,172,173). ACOs are funded to provide care for a defined patient population, and represent a policy option currently being implemented in the US by the Centers for Medicare and Medicaid Services (21,173), though they have not gained traction in the Canadian context. Nevertheless, the networks identified in a manner similar to what has been described here could serve as building blocks for the vertical integration of primary specialty and hospital care, should such policies be explored in the future in a Canadian context. Even without a move to formal ACOs, the identified networks may represent promising starting points for interventions that seek to improve quality or lower costs, such as health information systems, or care management protocols (172). Other than hospital privileging, there is currently no mechanism that aligns fee-for-service physicians to broader networks of service providers.   In conclusion, the identified networks of patients, physicians, and hospitals correspond closely to existing health regions (HSDAs). In cases where more than one network is nested within an HSDA they add some additional detail. Networks may be used to structure accountability around existing patterns of care, and may have more resonance with providers than administrative boundaries. This finding may be useful in future research where data are only accessibly at HSDA level, by confirming that HSDAs correspond closely to patterns of service provision by defined groups of providers, to defined populations.  However, the geographic contexts in which these networks operate are diverse, ranging from networks centered on individual hospitals in the densely populated lower mainland, to expansive networks of multiple small hospitals stretching over hundreds of kilometers in the Interior and Northern Health Authorities. The socioeconomic characteristics of the populations served also differ markedly. Any comparisons must still grapple with whether differences in use or outcomes are driven by geographic necessity, population characteristics, or whether they truly signal unwarranted variation in health care delivery. The following chapter scrutinizes differences in patterns of service use by geography more closely. In Chapter 7, factors driving variation in service use and outcomes among networks, as well as among existing health system units, are examined in greater detail. 67   Figure 5-1 Administrative health boundaries in BC   68  Table 5-1 Example data used to aggregate provider network A to a larger network Hospital  Individuals assigned to the hospital  Medical non-maternal admissions for network A residents  Ambulatory visits to specialist physicians by network A residents  Visits to primary care physicians by network A residents  # patients # primary care physicians # other physicians  # of admits Admission loyalty Rank  # of visits Physician loyalty Rank  # of visits Primary care loyalty Rank A 26,000 26 30  2,400 54% 1  280,000 70% 1  280,000 88% 1 B 180,000 250 600  1,200 27% 2  90,000 20% 2  22,000 7% 2 C 120,000 100 200  280 4% 3  22,600 5% 3  7,800 2% 3 D 60,000 200 100  200 5% 4  4,800 1% 4  700 0.5% 4   69  Table 5-2 Characteristics of BC residents who were and were not linked for network development  Linked Not linked  N=3,993,149 N=278,946 Female, n(%) 2,062,274 (51.6%) 103,847 (37.2%) Age, mean (sd) 42 (22.9) 41 (21.8) Newborn during study period, n(%) 136,418 (3.4%) 1,462 (0.5%) Died during study period, n(%) 67,795 (1.7%) 606 (0.2%)    70  Table 5-3 Service use characteristics of population used to define networks over three-year study period (N=3,993,149) Primary care visits, N(%) 0 100,892 (2.5) 1-6 1,333,562 (33.4) 7-12 998,124 (25.0) 13-18 637,932 (16.0) 19-24 372,906 (9.3) 25+ 549,733 (13.8) Visits for specialty services (ambulatory), N(%) 0 340,315 (8.5) 1-6 1,406,074 (35.2) 7-12 746,619 (18.7) 13-18 452,946 (11.3) 19-24 296,773 (7.4) 25+ 750,422 (18.8) Non-maternal medical hospitalizations, N(%) 0 3,658,003 (91.6) 1 217,201 (5.4) 2 63,421 (1.6) 3 26,126 (0.7) 4+ 28,398 (0.7)  71  Table 5-4 Characteristics of physicians included in analysis (n=13,865)  n (%) Data source Both MSP and DAD 7,727 (55.7) MSP only 2,804 (20.2) DAD only 3,334 (24.0) Practitioner specialty General Practice 6,212 (44.8) Medical specialists 3,908 (28.2) Surgical Specialists 3,175 (22.9) Imaging specialist 417 (3.0) Laboratory specialist 153 (1.1)    72  Table 5-5 Characteristics of 79 BC hospitals by number of linked patients Hospital network size ≤15,000 15,001-50,000 50,001-150,000 >150,000 Number of hospitals 36 20 15 8 Hospital network characteristics Median Range  Median Range Median Range Median Range # patients 5126 117- 14,245 23,748 15,085- 45,605 85,977 56,631- 143,752 177,661 164,364- 505,374 # GPs 12 1-28 38 24-77 121 88-224 250 166-630 # medical specialists 1 0-184 6 1-58 36 20-235 119 61-423 # surgical specialists 0 0-77 7 0-83 45 25-111 94 62-282 # lab specialists 0 0-15 1 0-9 7 0-26 24 3-57 # imaging specialists 0 0-5 0 0-3 3 0-7 9 1-22 Loyalty measures Median Range  Median Range Median Range Median Range % primary care visits within network  83.3 29.1-96.2 92.6 66.0-96.7 92.2 74.8- 94.9 87.2 81.9-95.8 % specialty physician visits within network 19.5 3.6-42.9 38.5 18.5-81.6 42.9 23.0-79.2 52.6 36.3-88.5 % acute medical hospitalizations within network 67.6 7.2-86.4 70.4 10.3-83.5 78.6 47.3-86.6 61.9 45.2-90.0  73  Table 5-6 Network characteristics by patient network size of largest linked hospital Hospital hub size Satellite (<50,000 total) Small 15,001-50,000 Medium 50,001-150,000 Large >150,000 # networks 6 4 12 7 Hospital network characteristics Median Range  Median Range Median Range Median Range # hospitals 1 1-4 4 4-5 3 1-8 2 1-7 # patients 323 117- 45,993 63,886 51,623- 68,837 102,093 56,631- 172,559 267,536 167,193- 823,312 # GPs 11 5-79 119 70-150 134 88-359 352 166-1,133 # medical specialists 1 0-16 23 8-26 39 23-115 141 73-1031 # surgical specialists 0 0-22 24 10-32 43 24-122 112 62-717 # lab specialists 0 0-4 19 6-10 8 0-27 27 3-114 # imaging specialists 0 0-2 2 0-3 3 0-9 11 1-30 Loyalty measures Median Range  Median Range Median Range Median Range % primary care visits within network  59.9 35.6-94.5 94.9 91.1-96.5 92.4 84.2-94.8 90.1 87.2-95.8 % specialty physician visits within network 10.6 3.6-58.6 66.3 46.6-84.9 55.8 41.6-79.3 55.0 36.3-88.5 % acute medical hospitalizations within network 80.7 76.7-86.4 83.1 78.0-86.1 80.9 66.6-89.6 79.3 66.9-92.7 Costs* Mean Range Mean Range Mean Range Mean Range Observed costs 869 340-1,473 1,174 959-1,285 1,268 1,050-1,477 1,106 889-1,253 Predicted costs 871 589-1,420 1,106 920-1,217 1,237 10,083-1,436 1,131 1,000-1,251 Difference between observed and predicted -2 -249- 173 69 38 – 93 31 -75 – 88 -25 -111-44.7 % of total costs from hospital services 66.1 57.6-74.4 59.8 56.3-62.4 57.8 55.8-59.5 54.0 52.5-56.9 % of physician costs from specialists 70.0 52.9-82.2 53.2 45.2-57.4 59.2 56.8-62.4 62.5 60.9-65.2 *Mean, minimum and maximum values for cost variables do not necessarily correspond to the same networks   74    Figure 5-2a Hospital networks mapped by Health Service Delivery Area (HSDA)   75  Figure 5-2b BC hospital networks mapped by Health Service Delivery Area (HSDA) (inset of lower mainland)    76   Figure 5-3 Observed and predicted costs for all networks     77  Chapter 6 – Defining health service environments  Background  Marked geographic variation has been observed in the use of health care, including specific procedures and drugs, as well as in broader categories of services. The review described in Chapter 3 found that while considerable attention has been devoted to describing variation in health care service use in Canada, we are still without clear conclusions about factors that shape observed variations, or the implications of variation for patient outcomes and health system costs.  Up to this point, much of the Canadian literature has focused on categories of services in isolation, for example, visits with family physicians, contact with specialists, hospitalization rates, or use of long-term care services. This fails to take into account the complexity of the health care system and the potential for substitution of health care services in different geographic contexts. For example, it may be that greater distance to hospital leads to appropriate substitution of hospital care for community-based care, as patients are kept longer for monitoring and recuperation to avoid the need for a return visit (147). On the other hand, hospitalization in other contexts may be a poor substitute for primary care not available in the home community (148–150). Analysis of Canadian Community Health Survey data found large reported variation in categories of service use, but not in unmet health care needs (149), which also supports the possibility of appropriate substitution.   A considerable body of literature has also examined urban/rural differences in health services use (once again, rarely assessed comprehensively). Most often, areas are grouped according to various definitions of urban and rural places (e.g. postal code, census urban/rural, Metropolitan Influenced Zones) and then measures of service use are compared. Some studies have developed new methods of classifying urban and rural areas, specific to health service delivery (101,178,179). A particularly interesting example is that of Rosenberg & Hanlon (134), who created health service environments using cluster analysis in Ontario. These were defined at the level of 43 Public Health Units, and incorporated measures of hospital, acute, and chronic care bed supply, number of total physicians, GPs, and specialists, nursing full-time equivalents, as well as population and population density (134).  The present analysis is informed by this research, though the approach is reversed. Rather than first classifying areas by geographic characteristics related to rurality (e.g. distance,  78  density, population) or availability of resources for health services (e.g. health human resources, hospitals, imaging facilities), and then comparing patterns of service use, we examine service use directly, and then seek to uncover factors which may explain any observed differences. While other chapters of this thesis focus on physician and hospital services, this analysis incorporates data describing pharmaceutical use, and home and community care as well. This expanded analysis was made possible by the availability of a data set constructed by the BC Ministry of Health (described in more detail below). These data were available at a geographic rather than individual level, and for that reason could not be used in the previous chapter or later analyses of individual-level costs. However, their examination at the small geography of Local Health Area (LHA) (n=89) provides a more complete picture of health care service use and potential patterns of substitution in the province.  The primary research questions for this work is whether LHAs can be grouped according to revealed patterns of health care services used and if so whether these groupings align with the existing planning areas of Health Service Delivery Areas. A secondary question is whether the differences in service use are related to particular features of demography or geography and whether understanding these differences might be a useful lens for health care services policy and planning. Approach  This analysis describes patterns of health service use across BC, from the patient perspective. This means all services used by an area’s residents are captured, regardless of where they were accessed. Average per-capita costs of services were standardized to adjust for population age and sex, and areas that were similar in their volume and mix of health services were grouped into clusters.  Data were accessed at the level of LHA, the smallest unit of geography commonly used in the context of health care. There are 89 LHAs in BC, with populations ranging from 3,480 (Kettle Valley) to 336,369 (Surrey). LHAs are nested within the 16 Health Services Delivery areas (HSDAs) and the five geographically based Health Authorities (HAs). As such, they represent geographic boundaries relevant to health care managers and policy-makers, and were the smallest unit at which data could be accessed. Results of the previous chapter confirm that the larger HSDAs correspond closely to patterns of physician and hospital service use. For the present analysis the smaller geographic areas are helpful because they create areas that  79  are somewhat more homogenous in their population, geographic and health care service use characteristics. Data   This analysis used the Health System Matrix, produced by the Modeling and Analysis Team, within the Health System Planning Division of BC’s Ministry of Health Services. These data include:   Medical Services Plan (MSP) Fee-For-Service (FFS) payments to physicians  Hospital Discharge Abstract Database (DAD)  Home and Community Care (HCC)  PharmaNet (PN) pharmaceutical spending  The matrix tabulates service use at the level of LHA, and across sex and age groups. It represents an extremely useful resource in that it captures the major areas of public health care spending, as well as complete data on pharmaceutical spending available through PharmaNet (i.e. this data set includes both public and private payments for pharmaceuticals), in one data source. There are notable gaps in the available data, including cancer services (chemotherapy and radiation therapy) covered through the BC Cancer Agency, services provided by the BC Renal Agency, and some payments to physicians through alternate payment plans. Other health-related services not captured include public and environmental health, health emergency management, emergency transport, and case management, all of which are more population-focused rather than individual-focused expenditures. Many of these data sources are not routinely available to researchers through Population Data BC, and so the Health System Matrix captures health care service use as completely as is now possible in BC. Categories of service use  For the present analysis, service use (cost) was grouped into the categories listed in Table 6-1. These represent distinct types of services, defined either by provider (GP, medical specialist, surgical specialist), location of care (office, hospital, home/community, residential care facility), or in the case of hospital services, type (medical/surgical) and circumstances of care (day/inpatient surgery, elective/emergent). Home and community care includes community-based supports for daily living as well as short- and long-term residential care. Pharmaceuticals include all payments for pharmaceuticals, regardless of source (public, private insurance, and out-of-pocket).  80   Data were available on more specific classes of services (i.e. obstetrics, gynaecology, mental health, oncology, palliative care, rehabilitation, and paediatrics), however these correspond to defined patient groups, and use may be more directly shaped by area population characteristics. Data were also available on fee-for-service payments to physicians in emergency departments. However, in some hospitals these physicians are paid by salary, which is missing in the matrix data. For this reason, emergency department use was not included as a separate category, as similarities between areas may be driven more by payment arrangements then by real patterns of service use. Day surgeries performed on BC residents in Alberta are also missing from the hospital data, though other acute care is captured. We also miss drugs used by patients in hospital, and by residential care clients who are in extended care facilities attached to acute care hospitals, because they use the hospital pharmacy for their drugs, rather than community pharmacies, therefore are not in pharmanet/pharmacare data.  Five years of data were averaged from 2004-5 to 2008-9 to ensure stable values in less populous LHAs. Per-capita costs of care for each category were calculated by sex and age strata (0-17, 18-49, 50-64, 65-74, 75+) in each LHA. Costs were indirectly standardized using the age and sex distribution of the entire BC population (averaged over the five years of pooled data) as the standard population.  For the purpose of describing health resources available, hospitals were classified by peer group (academic, large, medium and small community) according to the Canadian Institute for Health Information’s methodology (180). CIHI groups hospitals into four peer groups: teaching hospitals, large community hospitals, medium community hospitals, and small community hospitals. Teaching hospitals are defined as hospitals with full membership in the Association of Canadian Academic Health care Organizations (ACAHO), and with formal partnerships with universities to provide undergraduate and post-graduate medical education. Hospitals are categorized as large community hospitals if they meet two of the following three criteria:   • More than 8,000 inpatient cases   • More than 10,000 weighted cases   • More than 50,000 inpatient days  Hospitals that do not meet the above criteria are classified as medium or small depending on the hospital’s total weighted cases (medium: 2,000 weighted cases or more, small: fewer than 2,000 weighted cases).  81  Hierarchical cluster analysis  The intent of this analysis was to identify groups of LHAs that have similar patterns of health care service use, considering that use in the narrow and detailed areas described above. For this reason, we refer to the identified clusters as “health service environments.” This is a descriptive analysis, attempting to use detailed data to reveal patterns of similarity.   Given this interest we chose to use hierarchical cluster analysis. Hierarchical clustering starts from the point where each data point (in this case LHA) is a single “group” and then successive linkages are made to minimize within group differences. A pictorial representation of the revealed clusters is provided with a dendogram (Figure 6-1). This visually represents information about how LHAs are grouped. At the bottom of the figure, each LHA is treated as its own cluster. The LHAs are combined at various levels of similarity (joined by horizontal lines), until at the top of the figure the whole province is grouped together. The height of the lines gives information about the similarity of the clusters. Long vertical lines (for example, separating “metropolitan” and “remote” LHAs) indicate a distinct separation between clusters. Shorter lines (for example, separating “non-metropolitan” clusters a and b) reflect clusters that are not as distinct.  Adjusted LHA-level average per-capita spending for each category of care was standardized with a mean of zero and standard deviation of one. Hierarchical clustering was done using Stata’s cluster command and Ward’s minimum variance method (minimizing the total within-cluster sum of squares) (181). The resulting dendogram illustrates the arrangement of clusters. The height of each link between branches reflects the calculated dissimilarity between the two linked groups. We also report each cluster’s average per-capita cost for all categories of care.  To confirm the consistency and reliability of the cluster solution, the analysis was re-run using earlier years of data (2002/3-2003/4), as well as with alternate variable specification (collapsing or subdividing within categories – for example, subdividing GP services into office, hospital and other, subdividing home and community care services, or collapsing all physician services by specialty but not location-of services). Results  Figure 6-1 shows the dendogram produced in cluster analysis. There are multiple possible cluster solutions. There are three distinct high-level clusters, which we label metropolitan, non-metropolitan, and remote health service environments. This labeling is based  82  on the fact that all LHAs in the “metropolitan cluster” are located in densely populated urban areas surrounding Vancouver and Victoria (including Nanaimo), the “remote” LHAs are in isolated coastal and northern LHAs, and the “non-metropolitan” LHAs encompass the remaining area of BC outside the Lower Mainland, and southern Vancouver Island (Figure 4). The metropolitan cluster could in theory be subdivided into LHAs in metro Vancouver on the right branch, and Victoria plus suburban LHAs on the left branch, if that degree of detail was required (Figure 6-1). The non-metropolitan LHAs has been subdivided into two smaller clusters (a) and (b) (which could also be broken down further). These are mapped in Figure 6-2.  Table 6-2 describes average per-capita costs for each category of service use by the chosen four clusters. Overall the highest percent of spending is for hospital services, followed by fee-for-service payments to physicians, home and community care, and pharmaceuticals. Metropolitan LHAs used the most physician services overall. They also used more diagnostic and imaging services than non-metropolitan LHAs, and slightly more home and community care and pharmaceuticals. Non-metropolitan LHAs had notably higher spending on GP services provided in hospital, as well as medical, day surgery, and elective hospital services. Remote LHAs had extremely high use of hospital services, and much lower use of other physician services. Despite these differences by categories of care, total per-capita spending only differed by roughly $68 between metropolitan and non-metropolitan LHAs, and $153 between metropolitan and remote LHAs. Looking within the non-metropolitan areas we observe that cluster (a) had slightly higher use of specialist services, laboratory, imaging, and pharmaceuticals, and cluster (b) had higher use of GP services in hospital and medical hospital services.  The majority of British Columbians live in metropolitan LHAs (58.0%) (Table 6-3). A higher percentage of metropolitan residents are female, and the percentage female declines with increasing remoteness. Residents of non-metropolitan cluster (a) are, on average, older than both metropolitan residents, and residents of remote LHAs, and also more concentrated in higher income quintiles. The majority of remote residents are in the lowest income quintile.  Though population characteristics vary, what more clearly differentiate the identified clusters are the health system resources available. Not surprisingly, metropolitan LHAs have more total physicians per capita, and particularly more medical and surgical specialists (Table 6-4). All of the province’s teaching hospitals are located within the metropolitan cluster, which helps to explain the higher supply of specialist physicians in these areas. The high Resource  83  Intensity Weights (RIW) per hospital separation among metropolitan hospitals reflects the fact that many of these are tertiary care centres. This measure reflects all hospitalizations at these facilities, and is not restricted to cluster residents. This also means that the “supply” of physicians is in a sense over-stated as physicians also serve patients from outside the metropolitan LHAs.  Non-metropolitan cluster (a) can be distinguished from (b) by the fact that it has relatively greater physician supply (especially specialist), and is the location of several large community hospitals. The distinction between non-metropolitan clusters is evident in Figure 6-3, which shows that with the exception of the small hospital in Fort St. John, cluster (a) contains all of the hub hospitals for non-metropolitan networks identified in the previous chapter. The implication of this is that networks tend to include more than one cluster type. The most common boundary crossing is networks that include non-metropolitan (a) and non-metropolitan (b) areas. No networks cross into three cluster types.   When examining cluster consistency by altering variable specifications or years of data, a handful of LHAs moved between the two non-metropolitan clusters (a) and (b), but the overall cluster structure was robust to variable specification and years of data included. Discussion  The face validity of this clustering exercise, based only on observed patterns of use, is remarkably high. It groups areas that are similar in their geography and access to health care resources. It is clear that the primary distinction is between metropolitan and non-metropolitan areas. However, differences in patterns of use were evident among non-metropolitan areas as well. Remote communities had much lower use of GP services, home and community care, and pharmaceuticals, while all categories of hospital service use were high. However, some of these communities receive services (especially primary care) through nursing stations and first nations health services, and payments for medical transport or evacuation are not captured, so total spending for those communities is underestimated.  There are some other notable data limitations. The Health System Matrix omits all physicians on alternate payment plans. This affects specialists more than GPs, but is not likely to vary geographically. Of greater concern is the fact that BC residents’ day surgeries done in Alberta are not in the DAD (hospital separations) data, though other hospital services are captured. This may explain some of the lower day surgery rates in non-metropolitan (b).  84   The percentage of total physician spending on specialist services, and the percentage of total spending on hospital services show the largest difference across areas. Areas with greater supply of specialist physicians make greater use of specialist services. On the other hand, spending on hospital services, particularly medical admissions and inpatient elective surgery, is inversely related to hospital supply. This reflects the fact that residents of low-supply regions also use hospital services elsewhere.  A considerable body of literature has debated what “rural” means in the context of health and health services. Some have developed new measures, specific to aspects of health service provision (101,134,178). The objective here was not to propose a new or more comprehensive way to classify urban and rural communities, but rather to unpack and better understand differences previously described.  Rural areas naturally have a different supply structure, and this is evident in the distinct patterns of physician and hospital service use revealed in this analysis. Definitions of urban and rural developed for general purposes offer, at best, an indirect proxy of this structure (182). Where possible, it makes sense to either examine health system resource supply directly or, where available, choose a definition of urban and rural areas designed to capture health system resources (101,134,178). If the question of interest has to do with differences in health care services use, another option is to abandon the notion of urban vs. rural entirely and instead use categories of either health system structure or health services use. It seems that a good deal of research uses “rural” or “urban” as a catch-all code to signify difference but without being explicit about what those differences are and what they might mean for health care planning or health care services use.  This analysis builds on the findings of the previous chapter. Apart from networks completely contained within dense urban centres, identified networks spanned clusters, with member hospitals in non-metropolitan cluster (b) linked to hub hospitals in non-metropolitan cluster (a). Relatively little unexplained variation in spending was observed across networks, but the cluster analysis suggests that we may have missed differences within networks and by implication within HSDAs. Individuals connected to the same network hospital may live hundreds of kilometers apart, and experience very different local health service environments. The role of identified clusters or health service environments in explaining individual and network/HSDA-level variation will be explored in greater detail in the following chapter.    85  Table 6-1 Categories of service use examined in cluster analysis Category Source Description General Practitioners GP office visits MSP GP and family physician services in office and locations other than hospital or residential care (based on service location codes)* GP other MSP All other GP services, including services provided in hospital, visits to residential care facilities, in-hospital surgery, emergency department services, and obstetrics, gynecology, mental health, oncology, palliative care, rehabilitation, and pediatric services Medical specialists Medical specialist office visits MSP Medical specialist physician services provided in office and locations other than hospital (based on service location codes)* Medical specialist in hospital MSP Medical specialist physician services provided in hospital and emergency departments, and obstetrics, gynecology, mental health, oncology, palliative care, rehabilitation, and pediatric services Surgical specialists Surgical specialist office visits MSP Physicians services by surgeons in office and all locations other than hospital (based on service location codes)* Surgical specialist in hospital MSP Hospital care, and services for obstetrics, gynecology, mental health, oncology, palliative care, rehabilitation, and pediatric services Category Source Description Diagnostics Pathology/laboratory MSP Laboratory services (physician services based on FFS billings) Diagnostic imaging MSP Diagnostic imaging (physician services based on FFS billings)    86  Category Source Description Hospital Medical care DAD Inpatient hospital care for medical patients (i.e. non-surgical patients), excluding obstetrics, gynecology, mental health, oncology, palliative care, rehab, and pediatric services, which in other categories Day surgery DAD Day procedures performed in hospitals. Includes all day procedures, therapeutic and diagnostic, elective and emergency Inpatient elective surgery DAD Elective inpatient surgeries (elective surgeries determined based on admission code) Trauma and emergency surgery DAD Non-elective inpatient surgery and trauma (based on admission code) Other hospital DAD All other hospital services Home and community care Home supports HCC Nursing care, home support services, rehabilitation services, and adult day care, provided by health authorities, and public funding to Community Services for Independent Living clients Residential care and assisted living  HCC Publicly funded support services provided in assisted living settings and residential care, including convalescent care, transition care, and respite Pharmaceuticals   PharmaNet PN Total value of prescription drugs from community pharmacies, including drugs paid by PharmaCare, extended health plans and patient. *Does not include obstetrics, gynaecology, mental health, oncology, palliative care, rehabilitation, and paediatric services or services provided in the emergency department, which are grouped in the “other” categories.  87   Figure 6-1 Dendogram of cluster analysis   88   Figure 6-2a Map of cluster solution (health service environments)     89   Figure 6-2b Map of cluster solution (health service environments)    90  Table 6-2 Average per-capita health care costs by cluster (health service environment) and percent distribution by category  Metropolitan Non-metropolitan Remote a b Total spending $2,424.2 $2,356.1 $2,357.8 $2,271.0 % of spending by category Physician services GP Office 5.9 5.5 5.1 1.4 GP Hospital 1.7 2.9 3.7 1.6 Medical Office 1.8 1.3 0.9 1.5 Medical Hospital 2.6 1.4 1.0 1.3 Surgical Office 0.6 0.6 0.5 0.5 Surgical Hospital 4.2 4.1 3.1 4.4 Diagnostic services Laboratory/pathology 2.8 2.5 2.4 3.2 Diagnostic imaging 2.9 2.7 2.3 2.8 Hospital services Medical 9.7 11.3 14.9 25.7 Day surgery 2.7 3.1 2.8 3.7 Inpatient elective surgery 3.4 4.4 4.2 5.0 Trauma/emergency 6.2 5.7 5.4 7.0 All other hospital 15.3 14.2 14.5 17.1 Home and community care 21.0 21.2 22.2 17.3 Pharmaceuticals 19.1 19.0 17.1 7.7 Note: Darker shading corresponds to higher spending    91  Table 6-3 Population characteristics by cluster, N(%)  Metropolitan Non-metropolitan Remote a b Cluster population 2,466,656 (58.0) 323,074 (34.0) 323,074 (7.6) 18,058 (0.4) Sex Female   1,282,606 (51.1)  732,143 (50.7) 159,792 (49.3)  8,900 (48.0) Age 0-19 449.082 (17.9) 271,993 (18.9) 66,101 (20.4) 4,184 (22.5) 20-39 676,882 (27.0) 350,956 (24.3) 80,955 (25.0) 4,762 (25.7) 40-59 805,771 (32.1) 432,984 (30.0) 99,809 (30.8) 5,720 (30.8) 60-79 443,646 (17.7) 298,405 (20.7) 63,129 (19.5) 3,271 (17.6) 80+ 132,341 (3.4)  87,933 (5.3) 13,968 (4.3) 618 (3.3) Income quintile 1 (highest) 477,675 (18.3) 322,603 (21.5) 66,493 (20.9) 717 (4.3) 2 493,406 (18.9) 322,382 (21.5) 70,513 (18.8) 914 (5.5) 3 528,686 (20.3) 305,227 (20.3) 64,558 (19.3) 2,094 (12.6) 4  549,503 (21.1) 276,490 (18.4) 62,711 (21.1) 4,388 (26.3) 5 (lowest) 556,982 (21.4) 275,611 (18.3) 69,752 (19.9) 8,542 (51.3) *Missing information on sex for 1,794 individuals, on age for 127,895, and income for 69,609       92  Table 6-4 Health system resource supply by cluster  Metropolitan Non-metropolitan Remote a b Physician supply per 100,000 residents GPs 118.0 132.2 134.9 165.5 Medical specialists 64.3 35.7 10.8 20.7 Surgical specialists 44.8 41.8 10.8 0.0 Imaging specialists 9.3 8.8 3.8 0.0 Laboratory specialists 3.2 3.3 0.3 0.0 Unknown* 27.0 21.1 20.4 51.7 # Hospitals (by CIHI peer group) Teaching 6 0 0 0 Large 8 8 0 0 Medium 4 10 6 0 Small 1 13 19 5 Other 4 2 0 0 Hospital resources Separations (within cluster, per 1,000 residents) 258 294 218 262 Resource Intensity Weight (RIW) per 1,000 residents 410 357 199 222 Resource Intensity Weight (RIW) per 1,000 hospital separations 1589 1214 911 850  93   Figure 6-3a Map of clusters (health service environments) and hospital networks  94   Figure 6-3b Map of clusters (health service environments) and hospital networks (inset of lower mainland)  95  Chapter 7 – Variation in health care costs and population health outcomes in BC Background A considerable body of research has highlighted geographic variation in health care service use and spending (1,2). Influential US research found that even after controlling for differences in age, sex, and health status, a substantial amount of variation in spending on Medicare beneficiaries remains unexplained (17). Results of the scoping review (Chapter 2) confirm that variation in the use of procedures, drugs, and types of services is observed in Canada. Whether or not this corresponds to areas that are systematically higher or lower cost is less clear. Some understanding of the factors driving observed variation is needed in order to interpret and potentially act on it. Most basically, an individual patient’s use should be determined by need and preference. Only variations which remain after accounting for patient need and preference may be considered unwarranted under the Dartmouth framework (16). The results of analyses in Chapters 5 and 6 add to this, suggesting that there may be health system differences driven by geography that create potentially warranted variations in the components of total care if not in total costs. Even in this case, complete and accurate accounting for need is central to variations research that seeks to inform health system efficiency. The review of Canadian literature reported in Chapter 3 found that two thirds of studies presented only crude or age/sex standardized values, and only 20% had a measure of health status. A recent systematic review in OECD countries found large variations across regions, hospitals, and physician practices (1). However, authors noted that of the 836 studies reviewed only 10% explored causes of variation. Moreover, existing international research often lacks detailed individual-level data needed to adjust for patient characteristics (1). The potential to use regional variation in health care to identify opportunities for performance improvement continues to receive attention (3,4). Analysis of regional variation in spending is central to the Canadian Institute for Health Information’s recent work aimed at measuring health system efficiency (5,183). This analysis, based on area-level data, suggests overall system efficiency is between 0.65 and 0.82 (where 1.0 represents the efficiency of the “best performing” region) and substantial gains are possible for most regions. Estimates of the maximum possible efficiency are based on the idea that if one health region can produce more “output” (better health outcomes) than another with the same level of inputs (spending), the latter is not efficient. This conclusion depends on not just quantifying the extent of variation  96  (adjusted for need) but also on relating that variation to differences in outcomes. This is little existing variations literature in this area, with only 14% of Canadian studies reviewed in Chapter 3 examining health outcomes of any kind.  The fact that outcomes are rarely examined may in part be because patient or population-level outcomes clearly attributable to health care services use are difficult to define. While the use of health services is captured in routinely collected administrative data, corresponding measures of its effect on the health of individuals or populations are harder to obtain. Researchers often resort to population-based measures such as admissions for ambulatory care sensitive conditions as outcomes of care, but at best these are proxies for ultimate interest in changes in health status attributable to health care use. Survey data can offer a rich array of potential individual-level outcome indicators, but at present in the BC context it is not possible to link these data at an individual level to actual use of health care services. In addition, small regional numbers, even for large surveys like the Canadian Community Health Survey, limit the ability of survey data to capture small-area variation.   Mortality data have the benefit that they are routinely collected and reflect the health status of the entire population. A limitation is that many factors beyond the health system influence mortality rates. Death is also a relatively rare event, and irrelevant for many types of services, where the goal of treatment is reduced morbidity, not changes in mortality (7).  There has been resurgence in interest in the concept of amenable or avoidable mortality (166,184), defined as deaths that are theoretically preventable with timely access to quality health care. Classifying a condition as amenable to health care is based on a judgment that once the condition has developed, treatment is available that can be reasonably expected to prevent death (185). This renewed interest in amenable mortality as a health care system performance indicator likely comes because this capitalizes on existing mortality data while appearing to correct for its known limitations.  In its 2012 Health Indicators report, the Canadian Institute for Health Information (CIHI) put forth definitions of “treatable” and “preventable” mortality, developed by a panel of experts for use in the Canadian context (169). They first subset deaths occurring under age 75, classifying these as premature. The choice of 75 as the age cut-off is of course essentially arbitrary and does not imply that some deaths among those 75 years and over could not have been further delayed through treatment. Of premature deaths, CIHI identified causes of death that could have been potentially avoided through provision of timely and effective health care, and classified these as “treatable.” Deaths that could have been potentially avoided through  97  prevention practices and public health policies are considered “preventable.” The remaining subset of premature deaths is classified as unavoidable.  “Treatable” mortality is therefore considered an outcome measure that more directly measures health system performance than all cause mortality (170). However, earlier measures of amenable mortality (on which the CIHI definition was based), are highly correlated with both premature and all-cause mortality, and lower numbers mean rates are less stable among small areas (20). For these reasons, in addition to treatable mortality, premature and all-cause mortality will also be retained as outcome measures. Statistical control for individual underlying health status and sociodemographic characteristics is critical in this analysis (7).  The analyses described in this chapter are guided by the conceptual framework developed in Chapter 2 and have several objectives. First, I seek simply to document the amount of variation in health care costs within the province of British Columbia. I then explore the extent to which “warranted” causes explain any observed variation. This includes measures of patient need (age, sex, and health status). I also examine the degree to which health service environment (defined based on the results of cluster analysis in Chapter 6) explains variation in spending.  We also build on the analysis in Chapter 5, comparing measured variation using Health Service Delivery Areas (HSDAs) to variation among hospital networks defined in Chapter 5. Despite the apparent concordance between HSDAs and networks, we wish to determine if choice of geographic unit has any effect on the magnitude of variation observed.  Finally, we explore the extent to which variation in spending corresponds to variations in available measures of population health (all-cause mortality, as well as premature and treatable mortality). We examine the degree to which adjustment for warranted drivers of health care costs and factors outside the health care system that shape health outcomes alters the relationship observed. Considering the idea of efficiency, we seek to determine the extent to which there are similar health outcomes, achieved at different levels of health spending (5,12).  Approach Data sources  This research used administrative data from British Columbia for April 1 2008 through to March 31, 2011, accessed through Population Data BC. The study population was drawn from all provincial residents who were enrolled in BC’s Medical Services Plan in that time period. The following sources of data were linked.  98  Client registry (consolidation file)  Population Data BC maintains a central file with data on all individuals who are eligible to receive health services, and/or actually receive services in BC (157). It includes detailed demographic information and area of residence for each individual, and the data are regularly cleaned and validated by the programmers at Population Data BC. We used this file to obtain sex, age, neighbourhood income quintile (based on dissemination area level data), and Local Health Area of residence. The point in the year registration with MSP started, and total days registered were also obtained and used in defining the study cohort. MSP payment information file (physician payments)  The physician payment data in Population Data BC captures fee-for-service payment records for physicians in the province (158).  Discharge abstracts database (hospital separations)  Any patient who is admitted to hospital for either inpatient care or same‐ day surgical care is entered into the hospital separation records at the time of their discharge or death (159). The hospital data provide the resource intensity weight indicating the relative service intensity of the stay, which is used for costing hospital stays.  Vital Statistics death file  BC vital statistics death data contain information on the year and underlying cause of death for all deaths occurring in the province, or among BC residents. The underlying cause of death is coded using ICD‐ 10 codes. Hospital code was also obtained and made it possible to identify the location of death when accessing hospital care (and determine when this was outside the geographic area of residence). Study population  The study population was drawn from BC residents who were continuously-enrolled in MSP in all 3 study years (>=275 days in fiscal years 2008/9, 2009/10, and 2010/11) or who were born or died during the study period but were registered for at least 75% of days after birth or before death. MSP covers all permanent residents of BC, except for approximately 4% of the population that is covered under federal health insurance programs, including First Nations and federal employees (e.g. RCMP or Armed Forces).  Some physicians are compensated through alternate payment plans (APP), which are not included in the fee-for-service data file. Importantly for this analysis, APP-compensated primary care physicians are not evenly distributed throughout the province. Primary care  99  physicians who practice in Primary Health Care Organizations (PHCOs) are paid under an alternative payment scheme and submit “encounter claims,” a form of shadow billing. These records are included in the MSP payment information file, and we used these records to identify patients whose primary care would not be fully captured in fee-for-service costs and exclude them from analysis. This does not capture all patients with APP payments for their care. Many specialists are also paid via APP. However, this APP primary care is clustered around a handful of clinics in only a few parts of the province, and is therefore more likely to influence observed variation.  Because of small population sizes, costs and outcome measures for individual hospital satellite networks could not be reliably estimated (see Chapter 5). For this reason, individuals linked to the five individual hospital satellites (all in remote coastal BC) were excluded. Finally, individuals for whom data on location of residence or other covariates were missing were also excluded. Primary outcome: Total cost  Fee-for-service physician payment records include exact paid amounts for all services. All paid amounts were adjusted to 2010/11 constant dollars using a specialty-specific fee index built and maintained by the BC Medical Association and used by the Canadian Institute for Health Information. All hospital separations include a Resource Intensity Weight (RIW) that corresponds to the average expected resource use given the diagnoses, any procedures performed, and the age and sex of the patient. RIWs were converted to dollars using a standard cost per weighted case provided by the BC Ministry of Health, and price-adjusted by using the same (2010/11) amount for every year of data. Average total annual cost over the three-year study period (total cost over the study period, divided by the number of months each individual was registered, multiplied by 12) was modeled as the primary outcome variable. Population drivers  Chosen variables were intended to capture and control for factors outside the health care system identified in the logic model (Chapter 1) that are likely to drive variation in total spending. Patient need for health care/population composition: This was captured using age category, sex, and health status. Health status was measured using indicator variables for Aggregated Diagnosis Groups (ADGs), assigned using the Johns Hopkins Adjusted Clinical Group Case-Mix System in the first year of the study period. This represents people’s need for health care at the beginning of the study period. This system uses diagnoses recorded in encounters with physicians and hospitals over a one-year period to assign patients to ADGs. This coding system  100  has been validated for use in British Columbia (163). Health status may change over the three-year period, but there is no reason to believe that changes will take place differentially in populations in different parts of the province. While imperfect as a measure of health status over time, this choice is not likely to introduce bias to the analysis. Enabling resources/community resources to support health: Resources to support health were measured using neighbourhood income quintile and high school completion at the level of LHA. Neighbourhood income is a household size-adjusted measure of household income, based on 2006 census summary data at the census dissemination area (DA) level. Where DA income data were suppressed because of small sample size, imputations based on reported income from adjacent DAs were substituted (186). The percent aged 19-54 with no high school was examined at the level of LHA, the smallest level for which this could be attained. Patient beliefs and preferences/culture and values for health care It is not possible to measure patient beliefs or preferences using available data. We examined two variables reflecting area-level ethnic composition based on 2006 census data: the percent of people who immigrated to Canada within the past 10 years (4.4% of the BC population) and the percent of people who self-identify as aboriginal (4.8% of the BC population). These groups differ in both their values and preferences for health care use, as well as health status (187–189). Health system drivers  Health system characteristics were explored in two different ways. First, we examined individual variables intended to capture health system resources. These are as follows:  Supply and mix of material resources: Available measures were limited to hospital resources. Average RIW per separation captures the average complexity of patients seen within the hospital, and indirectly reflects resources available to handle complex patients. Average number of hospital separations occurring within an area or network (but not limited to area or network patients) reflects service volume, and indirectly captures local capacity. Finally, the type of hospital captured using the CIHI peer group system (teaching, large community, medium, and small community) reflects both resources and capacity.  Supply and mix of human resources: Measures of available human resources were limited to physician supply by HSDA or connected to hospitals or hospital networks. Per-capita supply of GP and specialist physicians was examined, as well as the ratio of specialist to primary care physicians.  The second approach was to include markers for the four health service environments (metropolitan, non-metropolitan (a) and (b), and remote) identified in Chapter 6, assigned based  101  on individual LHA of residence. These represent distinct health service environments, and are distinguished by the variables mentioned above (supply of hospital and physician resources). Policy, governance and funding: As health care is provincially administered, the overarching policy, governance and funding structure will be the same across areas. The five geographically defined health authorities (HAs) represent jurisdictions independently responsible for planning programs and services (though they do not pay physicians directly, and highly specialized services are delivered through the Provincial Health Services Authority). As HSDAs and networks are nested within HAs, some remaining variation after controlling for population and health service environment may reflect regional differences in policy and governance.  Service and compensation models are somewhat variable (see above for discussion of Alternate Payment Plans). In addition to differences in physician and hospital supply described above, residents of very remote locations may receive care from nursing stations, or other sources not captured in Fee for Service physician payment data (94). In BC, nursing stations staffed by out-post nurses or nurse practitioners offer a wide range of health services, including diagnosis and treatment of minor diseases, with backup support from physicians. An indicator variable for residents of LHAs containing a nursing station was included to control for this. Professional culture: As with patient beliefs and preferences, this cannot be measured directly using administrative data. Any remaining observed variation after having controlled for the population and health system drivers above may be partly attributed to area variations in professional culture. Health outcomes  As has been established previously, it is necessary to connect service use to relevant outcomes in order to determine if differences in use or costs represent issues of inequity, or opportunities for efficiency gains. We examine treatable mortality as defined in the 2012 Health Indicators report (169), as well as premature (deaths occurring under age 75), and all-cause mortality. Geographic units of analysis  This analysis uses the hospital networks developed in Chapter 5 which link together patients, primary care physicians, specialists, and hospitals based on patterns of care. This approach is based on earlier work in Ontario and the US (21,171,172), with small modifications that recognize the constraints of BC data. We also examine variation among Health Service Delivery Areas (HSDAs), the administrative health units typically used in national reporting (22). We compare findings between hospital networks and HSDAs.  102  Model selection Costs  The cost outcome variable was not normally distributed. To account for this, we used a generalized linear model (GLM) with log link and Poisson distribution (variance proportional to the mean). Details of the model selection procedure are described in Chapter 4.  Regression modeling followed a blocked approach first estimating models with HSDAs or network fixed effects only, then adding covariates following a blocked approach (Table 7-1). Here the goal of modeling was to adjust for “warranted” causes of variation in costs. Model 1 included only age and sex, as in many studies this is the only adjustment presented. Model 2 added ADGs as a measure of health status to adjust for patient need more completely. Model 3 added indicators for area-level health service environment. Finally, Model 4 also included all remaining variables, reflecting enabling resources (income and education) and culture (% aboriginal and % recent immigrant). Modeling was repeated for both Health Service Delivery areas and hospital networks.  McFadden’s pseudo R-squared (R2) is reported. This treats the log likelihood of the intercepts only model as a total sum of squares, and the log likelihood of the full model as the sum of squared errors. This measures the improvement from the null model to the model with covariates, and approximates explained variability as would be estimated in Ordinary Least Squares regression. Differences in R2 were compared between models with HSDAs and networks, and at each stage in blocked regression.  Differences among networks and HSDAs were described by fixed effect coefficient estimates for these units. We also used blocked models with the covariates described in Table 7-1 (but without network and HSDA fixed effects) to predict individual costs, and averaged these predictions to create expected costs for each network and HSDAs. Individual observed costs were also averaged, and ratios of observed to expected costs calculated for each of the four models (Table 7-1) with progressively greater adjustment. Mortality outcomes  Logistic regression was used to model the three outcomes (all-cause mortality, premature mortality, and treatable mortality), with variance estimates adjusted for clustering. Here again, models were first estimated with area-level units only (HSDAs and networks), and then covariates entered in a blocked approach (Table 7-1). However, covariates were entered in a different order, reflecting the slightly different goal of modeling. Rather than adjusting for “warranted” causes of variation in costs, we sought to adjust for factors outside the health care system that influence population health. Model 1 included only patient age and sex (the same  103  variables as in models of costs). Model 2 then included dissemination-area level income quintile, the closest we had to individual-level resources to support health. Model 3 added in area-level education, percentage aboriginal, and recent immigrant, all factors known to shape health behavior and health outcomes independent of medical care. Model 4 finally included individual ADGs. While ADGs are a direct measure of current health status, they may also reflect the combined effect of non-medical determinants on health status, as well as previous health care. This final model could thus be interpreted as over-controlling given the goal of adjusting for only factors outside the health care system. On the other hand, Model 3 (without ADGs) could be viewed as under-controlling for population health status. Results will be interpreted accordingly.  Individual probabilities of each outcome were estimated with each model to allow the calculation of Standardized Mortality Ratios (SMRs) for each network and HSDA.  Results Study population and exclusions  There were 4,272,094 individuals continuously enrolled (>75% of days while alive) in BC MSP during the study period. This includes 68,401 who died during the study period, and 68,353 newborns. Individuals who were born and died during the study years were evenly distributed across the study areas. Though these events were associated with increased spending (despite receiving services for a subset of the study period), including dummy variables indicating a person was born or died during the study period in models od costs had no impact on regional variations, as these events were balanced over the study period and across HSDAs and networks.  There were 707,584 individuals recorded as living in more than one LHA over the study period, or missing LHA of residence in one or two study years. They were assigned to the location where they lived in two of the three study years, or else the known LHA where they lived most recently. The 268,371 individuals who had no health service use were assigned to a hospital network assigned to the majority of other residents of their LHA of residence. No LHAs contain more than one hospital and there was no ambiguity in this assignment.  We excluded 15,739 (0.37%) people from analysis because we had no information on their location of residence, 87,305 (2%) because of missing age, sex, or income, and 5,605 (0.13%) because of missing census variables. These people were evenly distributed across the province.   An additional 8,572 (0.20%) people were excluded because they received care from a Primary Health Care Organization (PHCO), which was paid via alternate payment  104  arrangements. This means not all of their service use would be captured in the fee-for-service records examined. These individuals were concentrated in a handful of LHAs where such clinics operate.   Finally, 19,852 (0.45%) people who were connected to the five individual satellite hospitals (Massett, Queen Charlotte Islands, R.W Large, Bella Coola, and Powell River) were excluded. These satellite networks were too small as a basis for reliable outcome estimates. Though classified as a satellite network, the 45,993 people connected to Mills Memorial were retained in analysis, as this network is of sufficient size for reliable reporting of costs and outcomes. The final study population included 4,135,021 people. Models of health care costs  Tables 7-2 and 7-3 show the results of cost models for networks and HSDAs. Coefficients (cost ratios) are first presented for networks and HDSAs only. Then models 1-4 each successively add more adjustment variables (listed in Table 1). For all network and HSDA ratios, the reference category is Richmond hospital or the Richmond HSDA, an outlier with both low spending and excellent health outcomes.  Without adjusting for any population characteristics, costs for the patients of the highest spending hospital network, Penticton, were 1.66 times the lowest, Richmond (Table 7-2). Costs in the highest spending HSDA, Okanagan (which contains Penticton hospital), were 1.50 times those of Richmond HSDA (Table 7-3). However, following adjustment for age, sex, health status, and health service environment (Model 3), spending in the highest cost network (Mills Memorial) and HSDA (Northwest) was only 1.22 and 1.21 times that of Richmond (Tables 7-2 and 7-3).  By far the most important predictors of differences in costs were variables indicative of patient need. Individual age, sex and health status were significant predictors of health care costs in all models. Adjusting for age only, women cost 10% more on average (Model 1). However, adjusting for age and health status, including the ADG for pregnancy, women cost slightly (3%) less, and this effect remained significant (Model 2). Costs were elevated among children ages 0-4 (compared to the reference 20-24 year olds), and climbed again from age 25 onward. ADGs associated with high costs included: Time Limited: Major (e.g. subarachnoid hemorrhage or acute stomach ulcer with perforation), Chronic Medical: Unstable (e.g. carotid artery occlusion or thyrotoxicosis), Psychosocial: Recurrent or Persistent, Unstable (e.g. cocaine dependence or opioid abuse), and Pregnancy. The inclusion of an individual-level measure of health status (ADGs) contributed a huge improvement in predictive power over  105  models with age and sex alone, with R2 values doubling from roughly 0.21 to 0.42 between Models 1 and 2.  The addition of health service environment (Model 3) explained a much smaller amount of individual-level variation in costs. Compared to the reference (metropolitan), residents of non-metropolitan (a) & (b), and remote areas had slightly higher costs (7- 2 and 7-3), as did residents of LHAs with a remote First Nations nursing station. As was shown in Chapter 6, this is attributable to higher spending on hospital services. These variables only accounted for a small change in R2 in individual-level models and not all categories reached statistical significance in models that also contained network or HSDA-level fixed effects.  We also estimated models with individual variables capturing resource supply (per-capita physician supply, proportion of specialist physicians, hospital supply), rather than the health service environments based on cluster analysis (results not shown). In this analysis only the proportion of specialist physicians was significant. The change in R2 was even smaller for these variables than for the four categories of health service environment. For this reason, and for parsimony, only health service environment is included in the presented models (Tables 7-2 and 7-3).  Finally, Model 4 added all remaining variables, reflecting enabling resources (income and education) and culture (% aboriginal and % recent immigrant). With the exception of percent high school completion, all of these variables were significant. A gradient was observed by income quintile, with costs in the lowest quintile 9% higher than the highest. A 10% increase in the percent recent immigrants corresponded to roughly 10% lower costs, while a 10% increase in the percent aboriginal corresponded to a roughly 4% increase in costs. These adjustments further attenuated differences among areas. Adding these variables also had the effect of reversing the relationship observed by health service environment – with lowest costs now seen in remote areas. To the extent that these variables reflect need or preference for health services, then this final model may offer the most complete adjustment for “warranted” causes of variation. However, having adjusted for individual-level health status, it could also be argued that these variables reflect barriers to accessibility of appropriate care, in which case differences are unwarranted, and Model 3 would offer the most appropriate adjustment. Unexplained variation in costs among networks and HSDAs  To visually explore variation in costs among networks and HSDAs, we plotted observed costs against costs predicted by models 1-3, averaged among all residents of each area (Figure 7-1). A much higher proportion of area-level variation was explained, suggesting much of the  106  unexplained individual variation occurs within and not between networks and HSDAs. The R2 for models with age, sex, and ADGs was 0.78 among networks, and 0.71 among HSDAs. In contrast to individual-level models, including health service environment resulted in a further jump in R2 for area-level correlations to 0.89 and 0.93 respectively (Figure 7-1). Because of the possibility that Model 4 represents “over-controlling” for warranted sources of variation, these results aren’t plotted, though R2 further increased to 0.93 and 0.94.  These plots show that at area-level the majority of variation is still explained by population need, though health service environment also plays a substantial role. Put differently, geographically based health service environment is a significant predictor of relatively small differences in costs observed among networks or HSDAs, but not a significant driver of costs at the individual level.  Model 3 represents our best attempt at adjusting for “warranted“ sources of variation. To the extent that we were successful in this effort, very little unexplained (or “unwarranted”) variation remains among networks or HSDAs. However, in the absence of data on patient preference, the potential for incomplete measures of health status, and the very subtle question of how much variation based on health service environment is necessary, it remains impossible to completely distinguish between “warranted” and “unwarranted” remaining variation. Regardless, the magnitude of variation that may be explained by true inefficiency, such as the discretionary provision of inefficient care (17), is small. Effect of unit of analysis  Choice of unit of analysis (newly-developed networks vs. existing HSDAs) appears to have had no impact on conclusions. R2 for the model of costs with only HSDA was 0.0045, compared to 0.0058 in the model with only networks (Table7-4). This difference is likely largely a function of the number of units of analysis (24 vs.16). Hospitals (n=76) and LHAs (n=84) also explained slightly more variation. Though differences persist in models with added covariates, these very low R2 values reflect that while differences in cost ratios among areas are significant, both networks and HSDAs explain only a very small amount of individual variation in cost.  We took care to develop networks that reflect patterns of service use, but find that conclusions about the magnitude or causes of variation do not differ markedly between models including networks and those with HSDAs (Tables 7-2 and 7-3). In the following section examining population health outcomes only results among multispecialty physician networks are reported, but findings were the same among HSDAs.  107  Observed and predicted outcomes  Compared to variations in costs, variations in outcomes were more marked, and a smaller proportion of variation can be explained with available data. Adjusting for age and sex, odds of all-cause and premature mortality for Fort St. John were 1.93 and 2.07 times those of Richmond, respectively (Tables 7-5 and 7-6). Age-sex-adjusted odds of treatable mortality were highest for Ridge Meadows: 1.64 compared to Richmond (Table 7-7). Adjusting for income and other census variables lowers odds of all-cause mortality to 1.69 in Fort St. John. Ridge Meadows becomes highest for premature mortality at 1.75 and remains highest for treatable morality at 1.60. However given smaller event counts and less stable rates, not all network odds ratios were statistically significant in models of premature and treatable mortality.  Age and sex remain important explanatory variables, with higher odds of mortality among men and at older ages. Mortality gradients by income quintiles were more pronounced than for costs, with odds of all cause mortality 40% greater among the lowest quintile compared to the highest, and odds of premature and treatable mortality more than double. Of the other census variables, lower high school completion and percent recent immigrants were associated with higher odds of premature mortality, and a larger percentage aboriginal population was associated with higher odds of treatable mortality. Still, in in models including age, sex, income, and other census variables, pseudo R2 values were only 0.25 for all-cause mortality, 0.11 for premature mortality, and 0.11 for treatable mortality (Tables 7-5, 7-6, and 7-7).   Model 4 also included individual ADGs. This slighted attenuated odds ratios for other covariates, and improved the explanatory power of the model (pseudo R2 =0.28 for all-cause mortality). However, this is intended as a direct measure of current health status reflecting the combined effect of non-medical determinants, and previous health care. Given the goal of these models was to adjust mortality outcomes for factors outside of the health care system that shape population health outcomes, including ADG would be over-controlling. The following analysis of the relationship between costs and outcomes focuses only on Models 1-3. Relationship between costs and outcomes  Figure 7-2 plots the ratio of observed to expected costs (x-axis), against the ratio of observed to expected morality (y-axis), for Models 1-3 and for each of the three mortality outcomes. Adjusting for age and sex only, there is an apparent linear relationship between ratios of costs and outcomes: networks with higher than expected costs also experienced higher than expected mortality (Figure 7-2). However, more complete adjustment for health status and health service environment (in the case of costs), and non-medical determinants of health (in the case of outcomes) eliminates this linear relationship. Plots of ratios based on fully adjusted  108  models show only small variation in costs (ratios constrained between 0.9 and 1.1), though wider variation in outcomes. Discussion  At study outset, we hypothesized that careful examination of unwarranted variation in health service delivery would highlight places where health services might be provided more efficiently. Our results confirm that there are indeed variations in the cost of health care services across the province. It turns out, however, that when we account for health status using ADGs, and health service environment, expected costs for both networks and HSDAs correspond very closely to observed costs, as is clearly evident in Figure 7-1. In the case of reasonably homogenous provincial health systems, it may be that geography is simply not the right lens through which to identify inefficiencies. Greater gains may be achieved by examining decision-making at the level of provider or organization (17), or in scrutinizing high-use population groups that could be cared for more efficiently (for example, by alternate providers or in alternate settings). In other words, there may still be inefficient or inappropriate service provision, but to the extent it exists, inefficiency does not appear to vary systematically among networks and HSDAs.  In contrast to expectations, unit of analysis (hospital networks or HSDAs), makes no appreciable difference in results. This may be because HSDAs were designed to correspond to provincial geography, and capture road networks connecting major centres, which also shape patterns of travel for health care services (177). This may offer reassurance in situations where data is only accessible at HSDA level, as is commonly the case. Regardless, both of these units explained only a small proportion of individual-level variation.  Variations in outcomes despite relatively invariant costs are apparent and concerning. However, this finding can be interpreted in more than one way. This may suggest that at similar levels of spending, different networks are more or less efficient in producing good health outcomes. In this case, there is the potential for efficiency gains by implementing practices from low-mortality areas in areas with similar spending but poorer-outcomes (5). Differences in outcomes may equally plausibly suggest the need for greater investment in health care or in programs outside of the health care system to attenuate differences in health outcomes. Analyses presented here cannot shed light on which of these interpretations ought to be given more weight.   109   It may also be the case that adjustment based on available measures of health status and non-medical determinants of health was incomplete. No adjustment for risk, health status, or other similar measures using known data sources and methods could ensure that any observed relationship between mortality and expenditures would be free of confounding by unmeasured health status, though patient-level analysis, as is presented here, is the strongest possible approach (7). Importantly, it is not clear a priori in what direction any unmeasured variation associated with underlying population health status would bias results. To explore this idea, in Model 4 we controlled for all variables that might possibly act as proxies for health status in both the cost and outcome models. This meant adding area-level census variables to the cost models, and adding ADG to the outcome models (these models are reported in the rightmost columns of Tables 7-2 and 7-3). While we believe this is over-controlling, this suggests more complete control of health status further attenuates any relationship between costs and outcomes, bringing the plots in Figure 7-2 even closer to a single mass around 1 on both axes (not shown).  The mortality outcomes used here are at best very indirect measures of health system performance. We attempted to adjust for factors outside of the health care system that influence mortality and explored the use of “treatable” mortality as a more direct indicator of health system performance. Though appealing as a feasible, easily understandable indicator of health system performance, previous research has called into question whether treatable mortality is a sufficiently sensitive indicator to pinpoint and capture health system performance (20). The fact that the gradient by income quintile was stronger for treatable mortality than for all-cause morality also suggests that factors outside the health care system have a strong effect on this measure. If misapplied and interpreted as a clear measure of health system performance, results based on treatable mortality have the potential to focus unwarranted attention on places that may in fact be providing high quality care in challenging settings, while distracting from factors outside the health care system that contribute to health inequities. Without outcome measures that capture health system performance more directly (190), it remains impossible to distinguish clearly potential health system inefficiency vs. unmeasured differences between populations.  In contrast to our findings, research from the US has reported threefold variation in unadjusted medical spending, with twofold variation remaining after adjusting form prices and illness burden (18,191). An Institute of Medicine report which used more complete adjustment for health status (hierarchical condition categories) found more modest variation – but still close  110  to twofold overall, with the ratio of the 90th to 10th percentile HRRs in spending exceeding 1.20 (17). We also found no evidence of exceptionally high-spending regions with comparable or even poorer outcomes (1,12).  It may not be surprising that observed variation in a province of Canada is smaller than has been observed elsewhere (1). In the context of a single-payer health system organized at a provincial level, many of the mechanisms driving marked regional variation observed in US studies (7) are not relevant. In British Columbia fees paid to physicians are standardized across the province, as are methods for determining the cost of hospital stays. Reimbursement for medical training is administered separately, so price will not shape any observed variation, as is an issue in US studies (38). All of the services examined here are fully publicly reimbursed, so insurance coverage and degree of cost sharing are also not relevant in BC, though they are drivers of variation in the US (7). There is no difference in legal context for medical malpractice within the province, so the possibility of medical malpractice risk differentially driving “defensive medicine” similarly does not apply (192).  The fact that we calculated hospital costs based on Resource Intensity Weights (RIWs) may underestimate true differences in hospital spending. Up until recently, BC hospitals have been largely paid via global budgets or annual lump-sum transfers (193). This means that while we can measure variation in service volume or use of more resource-intensive approaches to care, we cannot capture differences in price or organizational efficiency. These may still contribute to marked cost differences not observable using our methods.  The availability of individual level data to adjust for health status is a strength of this analysis (7). Risk adjustment based on crude categories of disease status has been shown to be incomplete (194), and analysis using more complete adjustment for health status found “unwarranted” variation to be lower than in other research, and changed which areas were identified as most costly (17). This is consistent with analyses here identifying marked changes in high cost ratio areas between the model with age and sex only and the model including ADG, and may also explain why the magnitude of observed variation was comparatively modest (1).  However, the use of ADGs to control for health status also has some drawbacks. US research has raised the possibility of an “up-diagnosis” bias, wherein residents of high-intensity (and high-cost) regions are assigned more diagnostic codes over the course of medical care, and therefore appear sicker (195). We have no way to determine if networks or HSDAs differ in their intensity of coding. Up until April 2010, all BC hospitals were paid via global budgets, with  111  no accounting for service volume or patient complexity (193). The introduction of some activity-based funding in 2010 (called Patient Focused Funding in BC) in 23 of the largest hospitals in BC had the potential to change this (196), but there would have been no differential incentives among areas to “up-code” patients in the first year of the study when ADGs were determined. Again, the policy context is provincial, and there were no policy changes directed at specific regions of the province during the period of this study.   More broadly, there is an inherent tautology in adjusting for the very health characteristics that we expect health care to modify (30). We used ADG information from only the first year of the three study years pooled in cost data, and so assigned ADG codes do not directly correspond to the full amount of health care received. While some people’s health status might get worse over time, others will recover from acute events. These effects will balance out, and there is no reason to expect any systematic bias by region.   We were unable to account for patient preferences. Recent analysis found that patients’ preferences explained 5% of regional variation in US Medicare spending (197). Similar studies have not been done in Canada. Given the limited unexplained variation at the network or HSDA level, however, unless patient preferences are highly correlated with the variables we controlled for (age, sex, health status, or health service environment), this cannot be a major driver of variation in area-level spending. Available evidence suggests that preferences and values may be more strongly correlated at lower levels of aggregation (families or small neighbourhoods) than at the unit levels examined herein (198), and so differences in beliefs and preferences not correlated with measured characteristics may be expected to wash-out across areas. Certainly patient preference can and should explain some of the greater unexplained variation at an individual level, but it is not possible to capture patient beliefs and preferences using available administrative data.  Lack of information on non-fee-for-service payments to physicians is also an important data limitation, but we do not believe this has the potential to bias results. While on average excluded individuals receiving care from PHCOs were high service users, since they were excluded from both observed and expected cost values their exclusion would not affect observed cost ratios. If included, since some of their service records are missing from our data, we would not only be missing costs, but also potential ICD codes that would alter their ADG assignment. This would make it impossible to adequately adjust for their health status, and could have biased results.  112   While we excluded patients whose primary care providers were not paid fee-for-service, some groups of specialists are also compensated through alternate payment plans (APPs). In cases where services are provided in hospital, costs will still be captured through the RIW. Situations where ambulatory specialist care is compensated through APP are most commonly administered through the Provincial Health Services Authority and will not drive geographic variation, as all provincial residents with relevant conditions would access tertiary care compensated in the same way. To verify this empirically we examined the proportion of physicians in each area identified in hospital data but not captured in fee-for-service payment data, as they are likely paid by salary or other APPs. This proportion was higher in tertiary care centres (notably Vancouver and Prince George). Correcting observed physician costs based on this proportion increased the ratio of observed to expected spending in these two areas, but made little overall difference to the magnitude of variation observed or the concordance between observed and expected values (the R2 for networks, adjusted for age, sex, ADG, and health service environment declined from 0.88 to 0.85). Lack of data on specialist APP compensation is therefore a relatively minor limitation for this analysis.  Finally, few existing studies directly partition individual- and area-level variables using multilevel modeling approaches (66,115,128,129). While this analytic approach warrants further exploration, it was not possible to implement with available data and software.  In conclusion, by far the most important predictor of differences in costs was patient need, measured as age, sex, and health status (ADG). Health service environment was not a strong predictor of individual level costs, but explained a substantial amount of the modest differences among networks and HSDAs. Health service environment showed greater explanatory power than individual measures of resource supply in different areas, consistent with analyses in previous chapters suggesting substitution among service types. The networks developed in Chapter 5 correspond closely to existing health administrative units, and conclusions about the magnitude and causes of variation were not affected by choice of unit. Adjusting for relevant population characteristics, no relationship was observed between costs and outcomes, though there remain substantial variations in outcomes at similar levels of spending. Whether or not this can be attributed to health system successes or failures requires further investigation, and more importantly will depend on more direct measures of health care related outcomes that are not currently captured in data systems.    113  Table 7-1 Blocked modeling approach Model Healthcare costs Mortality outcomes 1 Age, sex Age, sex 2 Age, sex, ADG Age, sex, income 3 Age, sex, ADG, health service environment (metro, non-metro (a), non-metro (b), remote) Age, sex, income, census variables: % no high school, % recent immigrant, % aboriginal, 4 All covariates All covariates    114  Table 7-2 Models of average annual cost with network fixed effects  Networks only Model 1 Model 2 Model 3 Model 4  Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Network (reference=Richmond)            Vancouver General 1.14 (1.10 , 1.18) 1.14 (1.10 , 1.19) 1.06 (1.03 , 1.08) 1.06 (1.03 , 1.08) 1.04 (1.02 , 1.06) Royal Columbian 1.23 (0.98 , 1.53) 1.28 (1.09 , 1.52) 1.14 (1.04 , 1.25) 1.14 (1.04 , 1.25) 1.11 (1.01 , 1.22) Lions Gate 1.31 (1.23 , 1.40) 1.26 (1.22 , 1.31) 1.08 (1.03 , 1.13) 1.07 (1.03 , 1.12) 1.06 (1.02 , 1.10) Langley Memorial 1.18 (1.18 , 1.19) 1.26 (1.26 , 1.27) 1.10 (1.09 , 1.10) 1.08 (1.05 , 1.10) 1.05 (1.03 , 1.08) Surrey Memorial 1.21 (1.16 , 1.26) 1.30 (1.27 , 1.33) 1.12 (1.11 , 1.13) 1.12 (1.11 , 1.13) 1.08 (1.05 , 1.11) Peace Arch District 1.55 (1.54 , 1.55) 1.32 (1.31 , 1.32) 1.12 (1.11 , 1.13) 1.10 (1.07 , 1.13) 1.10 (1.07 , 1.13) Royal Jubilee 1.41 (1.22 , 1.63) 1.26 (1.22 , 1.30) 1.13 (1.12 , 1.14) 1.13 (1.12 , 1.14) 1.05 (1.00 , 1.09) Cowichan District 1.46 (1.46 , 1.46) 1.34 (1.34 , 1.34) 1.15 (1.15 , 1.16) 1.12 (1.09 , 1.16) 1.05 (1.00 , 1.09) Vernon Jubilee 1.51 (1.49 , 1.54) 1.35 (1.31 , 1.40) 1.19 (1.18 , 1.20) 1.16 (1.12 , 1.20) 1.09 (1.04 , 1.14) Kelowna 1.40 (1.40 , 1.40) 1.29 (1.29 , 1.29) 1.17 (1.17 , 1.18) 1.14 (1.11 , 1.18) 1.10 (1.06 , 1.14) Penticton 1.66 (1.62 , 1.70) 1.31 (1.27 , 1.35) 1.17 (1.15 , 1.19) 1.14 (1.09 , 1.18) 1.07 (1.01 , 1.12) Royal Inland 1.37 (1.33 , 1.42) 1.34 (1.30 , 1.38) 1.20 (1.18 , 1.22) 1.16 (1.13 , 1.20) 1.07 (1.03 , 1.12) Nanaimo 1.47 (1.42 , 1.52) 1.28 (1.21 , 1.36) 1.15 (1.09 , 1.21) 1.14 (1.10 , 1.19) 1.04 (0.98 , 1.10) St Josephs 1.44 (1.44 , 1.44) 1.26 (1.25 , 1.26) 1.07 (1.06 , 1.08) 1.04 (1.01 , 1.08) 0.99 (0.95 , 1.03) Campbell River 1.45 (1.43 , 1.47) 1.45 (1.39 , 1.52) 1.20 (1.13 , 1.26) 1.16 (1.10 , 1.23) 1.07 (1.00 , 1.13) Chilliwack 1.45 (1.41 , 1.50) 1.37 (1.36 , 1.39) 1.21 (1.21 , 1.21) 1.18 (1.14 , 1.22) 1.09 (1.04 , 1.15) Ridge Meadows 1.33 (1.33 , 1.33) 1.43 (1.43 , 1.43) 1.19 (1.18 , 1.19) 1.18 (1.18 , 1.19) 1.10 (1.06 , 1.14) Abbotsford  1.38 (1.35 , 1.41) 1.42 (1.41 , 1.43) 1.19 (1.17 , 1.21) 1.16 (1.12 , 1.20) 1.12 (1.06 , 1.18) Fort St. John 1.08 (1.01 , 1.15) 1.27 (1.20 , 1.33) 1.18 (1.12 , 1.24) 1.13 (1.06 , 1.20) 1.07 (1.00 , 1.15) Prince George 1.28 (1.21 , 1.35) 1.39 (1.34 , 1.44) 1.19 (1.16 , 1.22) 1.15 (1.11 , 1.20) 1.06 (1.01 , 1.12) East Kootenay 1.33 (1.24 , 1.42) 1.27 (1.25 , 1.29) 1.23 (1.19 , 1.27) 1.19 (1.12 , 1.25) 1.14 (1.08 , 1.20) Kootenay Boundary 1.43 (1.30 , 1.56) 1.30 (1.20 , 1.40) 1.20 (1.17 , 1.23) 1.16 (1.11 , 1.22) 1.12 (1.07 , 1.17)  115   Networks only Model 1 Model 2 Model 3 Model 4  Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Mills Memorial 1.41 (1.33 , 1.50) 1.53 (1.46 , 1.60) 1.32 (1.26 , 1.37) 1.22 (1.16 , 1.28) 1.08 (1.01 , 1.17) Sex (reference=Male)           Female    1.10 (1.08 , 1.11) 0.97 (0.96 , 0.97) 0.97 (0.96 , 0.97) 0.97 (0.96 , 0.97) Age category (reference=20-24)         0-4    2.09 (1.94 , 2.24) 1.50 (1.44 , 1.55) 1.50 (1.45 , 1.55) 1.49 (1.44 , 1.55) 5-9    0.66 (0.61 , 0.71) 0.76 (0.73 , 0.78) 0.76 (0.73 , 0.78) 0.76 (0.73 , 0.78) 10-14    0.53 (0.50 , 0.56) 0.67 (0.65 , 0.69) 0.67 (0.65 , 0.69) 0.67 (0.65 , 0.69) 15-19    0.74 (0.70 , 0.77) 0.89 (0.85 , 0.92) 0.89 (0.85 , 0.92) 0.89 (0.85 , 0.93) 25-29    1.33 (1.28 , 1.38) 1.15 (1.12 , 1.17) 1.15 (1.12 , 1.17) 1.14 (1.12 , 1.17) 30-34     1.70 (1.62 , 1.78) 1.32 (1.29 , 1.35) 1.32 (1.29 , 1.36) 1.32 (1.28 , 1.35) 35-39     1.70 (1.59 , 1.81) 1.30 (1.25 , 1.35) 1.30 (1.25 , 1.36) 1.30 (1.25 , 1.35) 40-44    1.54 (1.45 , 1.64) 1.24 (1.20 , 1.28) 1.24 (1.20 , 1.28) 1.24 (1.20 , 1.28) 45-49    1.59 (1.52 , 1.67) 1.28 (1.25 , 1.32) 1.28 (1.25 , 1.32) 1.28 (1.25 , 1.32) 50-54    1.86 (1.76 , 1.96) 1.42 (1.38 , 1.46) 1.42 (1.38 , 1.46) 1.42 (1.38 , 1.46) 55-59    2.22 (2.11 , 2.34) 1.58 (1.54 , 1.62) 1.58 (1.54 , 1.62) 1.58 (1.54 , 1.62) 60-64    2.79 (2.63 , 2.96) 1.82 (1.77 , 1.88) 1.82 (1.77 , 1.88) 1.82 (1.77 , 1.88) 65-69    3.59 (3.36 , 3.83) 2.09 (2.03 , 2.16) 2.09 (2.03 , 2.16) 2.10 (2.03 , 2.17) 70-74    4.52 (4.25 , 4.80) 2.33 (2.25 , 2.40) 2.33 (2.25 , 2.40) 2.33 (2.26 , 2.41) 75-79    5.68 (5.37 , 6.01) 2.56 (2.48 , 2.65) 2.57 (2.48 , 2.65) 2.57 (2.48 , 2.65) 80-84    6.91 (6.52 , 7.32) 2.73 (2.63 , 2.84) 2.73 (2.63 , 2.84) 2.73 (2.63 , 2.84) 85-89    7.98 (7.45 , 8.54) 2.87 (2.74 , 3.01) 2.88 (2.74 , 3.01) 2.88 (2.75 , 3.02) 90+    7.66 (7.08 , 8.28) 2.92 (2.79 , 3.05) 2.92 (2.79 , 3.06) 2.92 (2.79 , 3.06) Aggregated Diagnosis Groups             1. Time Limited: Minor       1.03 (1.03 , 1.04) 1.03 (1.03 , 1.04) 1.03 (1.03 , 1.04)  116   Networks only Model 1 Model 2 Model 3 Model 4  Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) 2. Time Limited: Minor-Primary Infections    1.08 (1.07 , 1.09) 1.08 (1.07 , 1.09) 1.08 (1.07 , 1.09) 3. Time Limited: Major       1.48 (1.46 , 1.51) 1.48 (1.46 , 1.51) 1.48 (1.46 , 1.51) 4. Time Limited: Major-Primary Infections    1.28 (1.27 , 1.29) 1.28 (1.26 , 1.29) 1.28 (1.26 , 1.29) 5. Allergies        1.00 (0.99 , 1.01) 1.00 (0.99 , 1.01) 1.00 (0.99 , 1.01) 6. Asthma       1.11 (1.09 , 1.12) 1.11 (1.09 , 1.12) 1.11 (1.09 , 1.12) 7. Likely to Recur: Discrete       1.11 (1.10 , 1.12) 1.11 (1.10 , 1.12) 1.11 (1.10 , 1.12) 8. Likely to Recur: Discrete-Infections    1.08 (1.07 , 1.09) 1.08 (1.07 , 1.09) 1.08 (1.07 , 1.09) 9. Likely to Recur: Progressive       1.32 (1.30 , 1.33) 1.32 (1.30 , 1.33) 1.32 (1.30 , 1.33) 10. Chronic Medical: Stable       1.40 (1.39 , 1.42) 1.40 (1.39 , 1.42) 1.40 (1.38 , 1.42) 11. Chronic Medical: Unstable       1.63 (1.61 , 1.66) 1.63 (1.61 , 1.66) 1.63 (1.61 , 1.65) 12. Chronic Specialty: Stable-Orthopedic    1.24 (1.22 , 1.26) 1.24 (1.22 , 1.26) 1.24 (1.22 , 1.27) 13. Chronic Specialty: Stable-Ear, Nose, Throat    1.08 (1.05 , 1.10) 1.08 (1.05 , 1.10) 1.08 (1.05 , 1.11) 14. Chronic Specialty: Stable-Eye       1.12 (1.11 , 1.13) 1.12 (1.11 , 1.13) 1.12 (1.11 , 1.13) 16. Chronic Specialty: Unstable-Orthopedic    1.21 (1.18 , 1.24) 1.21 (1.18 , 1.24) 1.21 (1.18 , 1.24) 17. Chronic Specialty: Unstable-Ear, Nose, Throat    0.97 (0.97 , 0.95) 0.98 (0.97 , 0.95) 0.99 (0.97 , 0.95) 18. Chronic Specialty: Unstable-Eye       1.15 (1.13 , 1.16) 1.15 (1.13 , 1.16) 1.15 (1.13 , 1.16) 20. Dermatologic       0.99 (0.98 , 1.00) 0.99 (0.98 , 1.00) 0.99 (0.98 , 1.01) 21. Injuries/Adverse Effects: Minor        1.09 (1.08 , 1.10) 1.09 (1.08 , 1.10) 1.09 (1.08 , 1.10) 22. Injuries/Adverse Effects: Major       1.31 (1.30 , 1.32) 1.31 (1.30 , 1.32) 1.31 (1.30 , 1.32) 23. Psychosocial: Time Limited, Minor    1.18 (1.16 , 1.21) 1.18 (1.16 , 1.21) 1.18 (1.16 , 1.21) 24. Psychosocial: Recurrent or Persistent, Stable       1.25 (1.25 , 1.23) 1.27 (1.25 , 1.23) 1.27 (1.25 , 1.23) 25. Psychosocial: Recurrent or Persistent, Unstable    1.70 (1.70 , 1.63) 1.78 (1.70 , 1.63) 1.78 (1.69 , 1.62) 26. Signs/Symptoms: Minor        1.16 (1.15 , 1.17) 1.16 (1.15 , 1.17) 1.16 (1.15 , 1.17)  117   Networks only Model 1 Model 2 Model 3 Model 4  Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) 27. Signs/Symptoms: Uncertain       1.36 (1.34 , 1.38) 1.36 (1.34 , 1.38) 1.36 (1.34 , 1.38) 28. Signs/Symptoms: Major        1.31 (1.29 , 1.33) 1.31 (1.29 , 1.33) 1.31 (1.29 , 1.33) 29. Discretionary       1.16 (1.15 , 1.17) 1.16 (1.15 , 1.17) 1.16 (1.15 , 1.17) 30. See and Reassure       1.05 (1.03 , 1.07) 1.05 (1.03 , 1.07) 1.05 (1.03 , 1.07) 31. Prevention/Administrative        1.19 (1.18 , 1.20) 1.19 (1.18 , 1.20) 1.19 (1.18 , 1.20) 32. Malignancy        1.32 (1.29 , 1.35) 1.32 (1.30 , 1.35) 1.32 (1.30 , 1.35) 33. Pregnancy        2.21 (2.11 , 2.31) 2.21 (2.11 , 2.31) 2.21 (2.11 , 2.31) 34. Dental       1.17 (1.14 , 1.19) 1.17 (1.14 , 1.19) 1.16 (1.14 , 1.19) Health service environment (reference=metropolitan)          Non-metropolitan a         1.03 (0.99 , 1.06) 0.97 (0.94 , 1.00) Non-metropolitan b         1.05 (1.01 , 1.09) 0.95 (0.91 , 0.99) Remote         1.12 (1.04 , 1.20) 0.82 (0.72 , 0.94) Nursing station             Nursing located in LHA of residence          1.09 (1.03 , 1.16) 1.02 (0.96 , 1.09) Dissemination area income quintile (reference=1, highest)         2           1.03 (1.03 , 1.04) 3           1.06 (1.05 , 1.06) 4           1.07 (1.06 , 1.08) 5           1.09 (1.06 , 1.11) Education (LHA-level, per 10% difference)             % age 19-54 with no high school           1.02 (0.97 , 1.08) Ethnic composition (LHA-level, per 10% difference)          % aboriginal           0.91 (0.86 , 0.95)  118   Networks only Model 1 Model 2 Model 3 Model 4  Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) % recent immigrant            1.03 (1.00 , 1.07) Constant 877 (877 , 878) 382 (362 , 403) 258 (250 , 267) 258 (249 , 267) 264 (249 , 280) Pseudo R2 0.006 0.211 0.462 0.463 0.463  119  Table 7-3 Average annual costs with fixed effects by HSDA  HSDAs only Model 1 Model 2 Model 3 Model 4  Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) HDSA (reference=31, Richmond)            11-East Kootenay 1.32 (1.23 , 1.42) 1.26 (1.18 , 1.35) 1.23 (1.18 , 1.28) 1.21 (1.13 , 1.28) 1.10 (1.04 , 1.17) 12-Kootenay Boundary 1.41 (1.29 , 1.55) 1.29 (1.17 , 1.43) 1.20 (1.16 , 1.24) 1.18 (1.12 , 1.24) 1.08 (1.02 , 1.15) 13-Okanagan 1.50 (1.39 , 1.62) 1.31 (1.22 , 1.42) 1.18 (1.15 , 1.20) 1.16 (1.11 , 1.20) 1.05 (0.99 , 1.10) 14-Thompson Cariboo Shuswap 1.37 (1.30 , 1.45) 1.32 (1.23 , 1.42) 1.20 (1.17 , 1.23) 1.18 (1.13 , 1.23) 1.03 (0.98 , 1.09) 21-Fraser East 1.37 (1.31 , 1.44) 1.38 (1.29 , 1.49) 1.19 (1.16 , 1.22) 1.17 (1.13 , 1.22) 1.06 (1.00 , 1.13) 22-Fraser North 1.18 (1.06 , 1.32) 1.23 (1.09 , 1.38) 1.11 (1.05 , 1.18) 1.11 (1.05 , 1.18) 1.07 (1.01 , 1.13) 23-Fraser South 1.24 (1.15 , 1.34) 1.30 (1.21 , 1.39) 1.12 (1.09 , 1.14) 1.11 (1.09 , 1.14) 1.05 (1.02 , 1.09) 32-Vancouver 1.15 (1.09 , 1.21) 1.13 (1.03 , 1.24) 1.06 (1.02 , 1.10) 1.06 (1.02 , 1.10) 1.02 (0.98 , 1.05) 33-North Shore/ Coast Garibaldi 1.29 (1.20 , 1.40) 1.24 (1.14 , 1.35) 1.07 (1.02 , 1.13) 1.07 (1.02 , 1.12) 1.03 (0.97 , 1.08) 41-South Vancouver Island 1.41 (1.21 , 1.64) 1.26 (1.16 , 1.36) 1.13 (1.10 , 1.16) 1.13 (1.10 , 1.16) 1.01 (0.96 , 1.07) 42-Central Vancouver Island 1.45 (1.38 , 1.52) 1.28 (1.17 , 1.40) 1.15 (1.10 , 1.20) 1.14 (1.10 , 1.19) 0.99 (0.94 , 1.05) 43-North Vancouver Island 1.42 (1.35 , 1.49) 1.33 (1.19 , 1.48) 1.12 (1.05 , 1.20) 1.11 (1.03 , 1.19) 0.98 (0.92 , 1.05) 51-Northwest 1.36 (1.28 , 1.45) 1.50 (1.39 , 1.62) 1.29 (1.23 , 1.35) 1.21 (1.14 , 1.28) 1.04 (0.98 , 1.11) 52-Northern Interior 1.27 (1.19 , 1.36) 1.38 (1.28 , 1.49) 1.19 (1.15 , 1.24) 1.17 (1.12 , 1.22) 1.02 (0.95 , 1.09) 53-Northeast 1.08 (1.00 , 1.17) 1.27 (1.16 , 1.38) 1.19 (1.12 , 1.26) 1.15 (1.07 , 1.24) 1.04 (0.96 , 1.13) Sex (reference=Male)           Female    1.10 (1.08 , 1.11) 0.97 (0.96 , 0.97) 0.97 (0.96 , 0.97) 0.97 (0.96 , 0.97) Age category (reference=20-24)         0-4    2.08 (1.94 , 2.24) 1.49 (1.44 , 1.55) 1.50 (1.44 , 1.55) 1.49 (1.44 , 1.55) 5-9    0.66 (0.61 , 0.71) 0.76 (0.73 , 0.78) 0.76 (0.73 , 0.78) 0.76 (0.73 , 0.78) 10-14    0.53 (0.50 , 0.56) 0.67 (0.65 , 0.69) 0.67 (0.65 , 0.69) 0.67 (0.65 , 0.69) 15-19    0.74 (0.70 , 0.78) 0.89 (0.85 , 0.92) 0.89 (0.85 , 0.92) 0.89 (0.85 , 0.93)  120   HSDAs only Model 1 Model 2 Model 3 Model 4  Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) 25-29    1.33 (1.28 , 1.38) 1.15 (1.12 , 1.17) 1.15 (1.12 , 1.17) 1.14 (1.12 , 1.17) 30-34     1.70 (1.62 , 1.78) 1.32 (1.29 , 1.35) 1.32 (1.29 , 1.35) 1.32 (1.28 , 1.35) 35-39     1.70 (1.58 , 1.81) 1.30 (1.25 , 1.35) 1.30 (1.25 , 1.35) 1.30 (1.25 , 1.35) 40-44    1.54 (1.45 , 1.64) 1.24 (1.20 , 1.28) 1.24 (1.20 , 1.28) 1.24 (1.20 , 1.28) 45-49    1.59 (1.52 , 1.67) 1.28 (1.25 , 1.32) 1.28 (1.25 , 1.32) 1.28 (1.25 , 1.31) 50-54    1.86 (1.76 , 1.96) 1.42 (1.38 , 1.46) 1.42 (1.38 , 1.46) 1.42 (1.38 , 1.46) 55-59    2.22 (2.11 , 2.34) 1.58 (1.53 , 1.62) 1.58 (1.53 , 1.62) 1.58 (1.54 , 1.62) 60-64    2.79 (2.63 , 2.96) 1.82 (1.77 , 1.88) 1.82 (1.77 , 1.88) 1.82 (1.77 , 1.88) 65-69    3.59 (3.36 , 3.83) 2.09 (2.02 , 2.16) 2.09 (2.02 , 2.16) 2.10 (2.03 , 2.17) 70-74    4.52 (4.25 , 4.80) 2.33 (2.25 , 2.40) 2.33 (2.25 , 2.40) 2.33 (2.25 , 2.41) 75-79    5.68 (5.37 , 6.02) 2.56 (2.48 , 2.65) 2.56 (2.48 , 2.65) 2.57 (2.48 , 2.65) 80-84    6.92 (6.52 , 7.34) 2.73 (2.62 , 2.84) 2.73 (2.62 , 2.84) 2.73 (2.63 , 2.85) 85-89    7.99 (7.44 , 8.57) 2.87 (2.74 , 3.02) 2.87 (2.74 , 3.02) 2.88 (2.74 , 3.02) 90+    7.67 (7.08 , 8.31) 2.92 (2.79 , 3.06) 2.92 (2.79 , 3.06) 2.92 (2.79 , 3.06) Aggregated Diagnosis Groups             1. Time Limited: Minor       1.03 (1.03 , 1.04) 1.03 (1.03 , 1.04) 1.03 (1.03 , 1.04) 2. Time Limited: Minor-Primary Infections       1.08 (1.07 , 1.09) 1.08 (1.07 , 1.09) 1.08 (1.07 , 1.09) 3. Time Limited: Major       1.48 (1.46 , 1.51) 1.48 (1.46 , 1.51) 1.48 (1.46 , 1.51) 4. Time Limited: Major-Primary Infections       1.28 (1.27 , 1.29) 1.28 (1.27 , 1.29) 1.28 (1.26 , 1.29) 5. Allergies        1.00 (0.99 , 1.01) 1.00 (0.99 , 1.01) 1.00 (0.99 , 1.01) 6. Asthma       1.11 (1.09 , 1.12) 1.11 (1.09 , 1.12) 1.11 (1.09 , 1.12) 7. Likely to Recur: Discrete       1.11 (1.10 , 1.12) 1.11 (1.10 , 1.12) 1.11 (1.10 , 1.12)  121   HSDAs only Model 1 Model 2 Model 3 Model 4  Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) 8. Likely to Recur: Discrete-Infections       1.08 (1.07 , 1.09) 1.08 (1.07 , 1.09) 1.08 (1.07 , 1.09) 9. Likely to Recur: Progressive       1.32 (1.30 , 1.33) 1.32 (1.30 , 1.33) 1.31 (1.30 , 1.33) 10. Chronic Medical: Stable       1.40 (1.39 , 1.42) 1.41 (1.39 , 1.42) 1.40 (1.38 , 1.42) 11. Chronic Medical: Unstable       1.63 (1.61 , 1.66) 1.63 (1.61 , 1.66) 1.63 (1.61 , 1.65) 12. Chronic Specialty: Stable-Orthopedic    1.24 (1.22 , 1.26) 1.24 (1.22 , 1.26) 1.24 (1.22 , 1.26) 13. Chronic Specialty: Stable-Ear, Nose, Throat    1.08 (1.05 , 1.10) 1.08 (1.05 , 1.10) 1.08 (1.05 , 1.11) 14. Chronic Specialty: Stable-Eye       1.12 (1.11 , 1.13) 1.12 (1.11 , 1.13) 1.12 (1.11 , 1.13) 16. Chronic Specialty: Unstable-Orthopedic    1.21 (1.18 , 1.24) 1.21 (1.18 , 1.24) 1.21 (1.18 , 1.24) 17. Chronic Specialty: Unstable-Ear, Nose, Throat    0.97 (0.97 , 0.95) 0.99 (0.97 , 0.95) 0.99 (0.97 , 0.95) 18. Chronic Specialty: Unstable-Eye       1.14 (1.13 , 1.16) 1.14 (1.13 , 1.16) 1.15 (1.13 , 1.16) 20. Dermatologic       0.99 (0.98 , 1.00) 0.99 (0.98 , 1.00) 0.99 (0.98 , 1.01) 21. Injuries/Adverse Effects: Minor        1.09 (1.08 , 1.11) 1.09 (1.08 , 1.11) 1.09 (1.08 , 1.10) 22. Injuries/Adverse Effects: Major       1.31 (1.30 , 1.32) 1.31 (1.30 , 1.32) 1.31 (1.30 , 1.32) 23. Psychosocial: Time Limited, Minor    1.19 (1.16 , 1.21) 1.19 (1.16 , 1.21) 1.18 (1.16 , 1.21) 24. Psychosocial: Recurrent or Persistent, Stable    1.25 (1.25 , 1.23) 1.27 (1.25 , 1.23) 1.27 (1.25 , 1.23) 25. Psychosocial: Recurrent or Persistent, Unstable    1.70 (1.70 , 1.63) 1.78 (1.70 , 1.63) 1.78 (1.69 , 1.62) 26. Signs/Symptoms: Minor        1.16 (1.15 , 1.17) 1.16 (1.15 , 1.17) 1.16 (1.15 , 1.17) 27. Signs/Symptoms: Uncertain       1.36 (1.34 , 1.38) 1.36 (1.34 , 1.38) 1.36 (1.34 , 1.38) 28. Signs/Symptoms: Major        1.31 (1.29 , 1.33) 1.31 (1.29 , 1.33) 1.31 (1.29 , 1.33) 29. Discretionary       1.16 (1.15 , 1.17) 1.16 (1.15 , 1.17) 1.16 (1.15 , 1.17) 30. See and Reassure       1.05 (1.03 , 1.07) 1.05 (1.03 , 1.07) 1.05 (1.03 , 1.07) 31. Prevention/Administrative        1.19 (1.18 , 1.20) 1.19 (1.18 , 1.20) 1.19 (1.18 , 1.20) 32. Malignancy        1.32 (1.29 , 1.35) 1.32 (1.29 , 1.35) 1.32 (1.30 , 1.35)  122   HSDAs only Model 1 Model 2 Model 3 Model 4  Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) Cost ratio (95% CI) 33. Pregnancy        2.21 (2.11 , 2.31) 2.21 (2.11 , 2.31) 2.21 (2.11 , 2.31) 34. Dental       1.17 (1.14 , 1.19) 1.17 (1.14 , 1.19) 1.16 (1.14 , 1.19) Health service environment (reference=metropolitan)          Non-metropolitan a         1.01 (0.98 , 1.05) 0.97 (0.94 , 1.00) Non-metropolitan b         1.03 (0.99 , 1.07) 0.95 (0.91 , 0.99) Remote         1.09 (1.02 , 1.17) 0.82 (0.72 , 0.94) Nursing station             Nursing located in LHA of residence          1.12 (1.05 , 1.18) 1.02 (0.96 , 1.09) Dissemination area income quintile (reference=1, highest)        2           1.03 (1.02 , 1.04) 3           1.05 (1.05 , 1.06) 4           1.07 (1.05 , 1.08) 5           1.09 (1.06 , 1.11) Education (LHA-level, per 10% difference)          % age 19-54 with no high school           1.03 (0.97 , 1.09) Ethnic composition (LHA-level, per 10% difference)          % aboriginal           0.88 (0.84 , 0.93) % recent immigrant            1.04 (1.01 , 1.07) Constant 882 (851 , 915) 383 (352 , 417) 258 (246 , 270) 258 (246 , 270) 274 (254 , 296) Pseudo R2 0.005 0.211 0.462 0.462 0.463  123   Figure 7-1 Observed and expected costs by networks and HSDAs (models 1-3)    124  Table 7-4 Comparing units of analysis  Network Hospital HSDA LHA Number of units 24 76 16 84 Average population 172,293 54,408 258,439 49,226 R2 of empty model 0.0058 0.0105 0.0046 0.0073     125   Table 7-5 Logistic models of all-cause mortality  Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio 95% CI Mortality ratio (95% CI) Network (reference=Richmond)            Vancouver General 1.12 (1.04 , 1.20) 1.13 (1.04 , 1.23) 1.15 (1.06 , 1.24) 1.11 (1.03 , 1.20) 1.02 (0.96 , 1.07) Royal Columbian 1.18 (0.83 , 1.66) 1.34 (1.08 , 1.67) 1.39 (1.13 , 1.70) 1.33 (1.07 , 1.64) 1.17 (1.04 , 1.32) Lions Gate 1.38 (1.20 , 1.60) 1.24 (1.13 , 1.35) 1.37 (1.28 , 1.47) 1.27 (1.12 , 1.45) 1.10 (0.99 , 1.21) Langley Memorial 1.31 (1.30 , 1.31) 1.53 (1.52 , 1.53) 1.59 (1.58 , 1.61) 1.47 (1.34 , 1.61) 1.28 (1.19 , 1.37) Surrey Memorial 1.14 (0.92 , 1.42) 1.37 (1.36 , 1.39) 1.39 (1.31 , 1.48) 1.33 (1.23 , 1.44) 1.19 (1.12 , 1.27) Peace Arch District 2.18 (2.17 , 2.18) 1.38 (1.37 , 1.39) 1.50 (1.46 , 1.54) 1.39 (1.23 , 1.56) 1.23 (1.12 , 1.34) Royal Jubilee 1.93 (1.30 , 2.87) 1.42 (1.25 , 1.62) 1.48 (1.31 , 1.69) 1.33 (1.10 , 1.61) 1.21 (1.07 , 1.36) Cowichan District 1.81 (1.81 , 1.81) 1.48 (1.48 , 1.49) 1.56 (1.54 , 1.58) 1.32 (1.15 , 1.50) 1.17 (1.06 , 1.30) Vernon Jubilee 1.98 (1.86 , 2.11) 1.54 (1.51 , 1.56) 1.54 (1.52 , 1.57) 1.34 (1.17 , 1.53) 1.18 (1.06 , 1.30) Kelowna 1.76 (1.76 , 1.76) 1.41 (1.41 , 1.42) 1.43 (1.42 , 1.44) 1.27 (1.12 , 1.45) 1.21 (1.09 , 1.33) Penticton 2.50 (2.29 , 2.73) 1.42 (1.37 , 1.46) 1.42 (1.38 , 1.47) 1.25 (1.09 , 1.44) 1.14 (1.03 , 1.27) Royal Inland 1.67 (1.56 , 1.80) 1.68 (1.56 , 1.81) 1.69 (1.57 , 1.82) 1.39 (1.21 , 1.60) 1.30 (1.16 , 1.44) Nanaimo 2.14 (2.08 , 2.19) 1.58 (1.46 , 1.71) 1.63 (1.51 , 1.76) 1.41 (1.25 , 1.60) 1.28 (1.16 , 1.41) St Josephs 1.83 (1.83 , 1.83) 1.38 (1.38 , 1.39) 1.43 (1.41 , 1.44) 1.25 (1.10 , 1.43) 1.06 (0.95 , 1.17) Campbell River 1.71 (1.60 , 1.83) 1.92 (1.77 , 2.08) 1.96 (1.80 , 2.12) 1.61 (1.39 , 1.86) 1.33 (1.18 , 1.50) Chilliwack 1.77 (1.55 , 2.02) 1.51 (1.36 , 1.68) 1.51 (1.37 , 1.67) 1.29 (1.11 , 1.50) 1.20 (1.07 , 1.35) Ridge Meadows 1.49 (1.49 , 1.49) 1.78 (1.77 , 1.79) 1.82 (1.81 , 1.84) 1.65 (1.48 , 1.84) 1.44 (1.33 , 1.57) Abbotsford  1.46 (1.45 , 1.46) 1.48 (1.43 , 1.53) 1.48 (1.44 , 1.53) 1.35 (1.19 , 1.53) 1.20 (1.09 , 1.33) Fort St. John 1.15 (0.82 , 1.64) 1.93 (1.71 , 2.18) 2.05 (1.77 , 2.38) 1.69 (1.39 , 2.06) 1.51 (1.23 , 1.85) Prince George 1.28 (1.14 , 1.44) 1.72 (1.62 , 1.81) 1.76 (1.67 , 1.85) 1.43 (1.23 , 1.67) 1.26 (1.12 , 1.42)  126   Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio 95% CI Mortality ratio (95% CI) East Kootenay 1.64 (1.36 , 1.98) 1.51 (1.43 , 1.60) 1.54 (1.44 , 1.65) 1.34 (1.16 , 1.56) 1.26 (1.10 , 1.45) Kootenay Boundary 2.02 (1.57 , 2.60) 1.68 (1.47 , 1.92) 1.69 (1.45 , 1.98) 1.49 (1.22 , 1.83) 1.35 (1.14 , 1.61) Mills Memorial 1.33 (1.26 , 1.41) 1.73 (1.64 , 1.83) 1.74 (1.63 , 1.86) 1.21 (0.97 , 1.51) 1.17 (0.98 , 1.39) Sex (reference=Female)           Male    1.43 (1.39 , 1.46) 1.44 (1.40 , 1.47) 1.44 (1.40 , 1.47) 1.37 (1.33 , 1.40) Age category (reference=20-24)         0-4    0.52 (0.32 , 0.84) 0.51 (0.32 , 0.83) 0.51 (0.32 , 0.83) 0.44 (0.27 , 0.71) 5-9    0.20 (0.12 , 0.32) 0.20 (0.12 , 0.32) 0.20 (0.12 , 0.32) 0.19 (0.12 , 0.31) 10-14    0.17 (0.11 , 0.27) 0.17 (0.11 , 0.27) 0.17 (0.11 , 0.27) 0.19 (0.12 , 0.29) 15-19    0.51 (0.37 , 0.70) 0.51 (0.37 , 0.70) 0.51 (0.37 , 0.70) 0.54 (0.40 , 0.74) 25-29    1.02 (0.77 , 1.35) 1.00 (0.75 , 1.33) 1.00 (0.76 , 1.33) 0.97 (0.73 , 1.28) 30-34     1.47 (1.22 , 1.78) 1.44 (1.19 , 1.74) 1.44 (1.20 , 1.74) 1.36 (1.13 , 1.64) 35-39     1.45 (1.15 , 1.83) 1.43 (1.13 , 1.80) 1.43 (1.13 , 1.80) 1.33 (1.05 , 1.67) 40-44    2.26 (1.85 , 2.75) 2.23 (1.83 , 2.72) 2.23 (1.83 , 2.72) 2.03 (1.67 , 2.46) 45-49    3.36 (2.81 , 4.01) 3.33 (2.79 , 3.98) 3.34 (2.79 , 3.99) 2.96 (2.49 , 3.52) 50-54    4.93 (4.14 , 5.87) 4.92 (4.13 , 5.85) 4.92 (4.13 , 5.85) 4.29 (3.63 , 5.07) 55-59    7.89 (6.62 , 9.40) 7.89 (6.62 , 9.39) 7.89 (6.62 , 9.40) 6.69 (5.63 , 7.94) 60-64    11.4 (9.5 , 13.8) 11.4 (9.5 , 13.8) 11.4 (9.50 , 13.8) 9.4 (7.8 , 11.3) 65-69    17.9 (14.9 , 21.4) 17.9 (14.9 , 21.5) 17.9 (14.9 , 21.5) 13.8 (11.6 , 16.4) 70-74    29.8 (25.0 , 35.5) 29.7 (24.9 , 35.4) 29.7 (24.9 , 35.4) 21.3 (18.1 , 25.0) 75-79    49.3 (41.0 , 59.4) 48.9 (40.5 , 58.9) 48.9 (40.6 , 59.0) 31.9 (26.7 , 38.1) 80-84    86.0 (72.2 , 102) 84.8 (71.2 , 101) 85.0 (71.3 , 101) 49.8 (42.0 , 59.1) 85-89    164 (135 , 199) 161 (132 , 195) 161 (133 , 196) 87 (72 , 105)  127   Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio 95% CI Mortality ratio (95% CI) 90+    326 (265 , 401) 318 (258 , 393) 319 (259 , 394) 176 (145 , 214) Dissemination area income quintile (reference=1, highest)         2       1.14 (1.09 , 1.2) 1.14 (1.10 , 1.19) 1.11 (1.07 , 1.15) 3       1.24 (1.15 , 1.33) 1.23 (1.17 , 1.3) 1.18 (1.13 , 1.24) 4       1.30 (1.19 , 1.41) 1.29 (1.22 , 1.38) 1.23 (1.16 , 1.30) 5       1.44 (1.32 , 1.58) 1.44 (1.35 , 1.55) 1.37 (1.29 , 1.45) Education (LHA-level, per 10% difference)          % age 19-54 with no high school         1.00 (0.84 , 1.19) 1.08 (0.96 , 1.20) Ethnic composition (LHA-level, per 10% difference          % aboriginal         0.90 (0.77 , 1.05) 0.94 (0.83 , 1.06) % recent immigrants         1.09 (1.00 , 1.18) 1.06 (0.99 , 1.13) Aggregated Diagnosis Groups             1. Time Limited: Minor           0.97 (0.94 , 0.99) 2. Time Limited: Minor-Primary Infections        1.05 (1.02 , 1.09) 3. Time Limited: Major           1.40 (1.32 , 1.48) 4. Time Limited: Major-Primary Infections        1.52 (1.44 , 1.59) 5. Allergies            0.80 (0.73 , 0.87) 6. Asthma           1.05 (1.00 , 1.10) 7. Likely to Recur: Discrete           1.02 (0.99 , 1.05) 8. Likely to Recur: Discrete-Infections        1.16 (1.12 , 1.21) 9. Likely to Recur: Progressive           1.20 (1.15 , 1.25) 10. Chronic Medical: Stable           0.94 (0.90 , 0.97)  128   Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio 95% CI Mortality ratio (95% CI) 11. Chronic Medical: Unstable           1.88 (1.81 , 1.96) 12. Chronic Specialty: Stable-Orthopedic        0.78 (0.72 , 0.84) 13. Chronic Specialty: Stable-Ear, Nose, Throat        0.89 (0.80 , 0.98) 14. Chronic Specialty: Stable-Eye           0.85 (0.82 , 0.89) 16. Chronic Specialty: Unstable-Orthopedic        0.84 (0.75 , 0.94) 17. Chronic Specialty: Unstable-Ear, Nose, Throat        0.99 (0.67 , 0.58) 18. Chronic Specialty: Unstable-Eye           0.90 (0.86 , 0.95) 20. Dermatologic           0.78 (0.75 , 0.80) 21. Injuries/Adverse Effects: Minor            1.01 (0.98 , 1.05) 22. Injuries/Adverse Effects: Major           1.15 (1.10 , 1.20) 23. Psychosocial: Time Limited, Minor        1.03 (0.97 , 1.11) 24. Psychosocial: Recurrent or Persistent, Stable        1.27 (1.12 , 1.09) 25. Psychosocial: Recurrent or Persistent, Unstable        1.78 (2.53 , 2.41) 26. Signs/Symptoms: Minor            1.10 (1.07 , 1.13) 27. Signs/Symptoms: Uncertain           1.03 (0.99 , 1.06) 28. Signs/Symptoms: Major            1.27 (1.23 , 1.31) 29. Discretionary           0.80 (0.78 , 0.83) 30. See and Reassure           0.83 (0.74 , 0.93) 31. Prevention/Administrative            1.00 (0.96 , 1.04) 32. Malignancy            1.92 (1.81 , 2.04) 33. Pregnancy            0.70 (0.58 , 0.84)  129   Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio 95% CI Mortality ratio (95% CI) 34. Dental           1.22 (1.10 , 1.35) Constant 5.11 x10-3 (5.11 x10-3 , 5.12 x10-3) 3.17 x10-4 (2.64 x10-4 , 3.81 x10-4) 2.54 x10-4 (2.10 x10-4 , 3.06 x10-4) 2.80 x10-4 (2.08 x10-4 , 3.78 x10-4) 2.51 x10-4 (1.95 x10-4 , 3.23 , x10-4) Pseudo R2 0.006 0.245 0.246 0.246 0.282     130  Table 7-6 Logistic models of premature mortality  Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Network (reference=Richmond)            Vancouver General 1.11 (0.94 , 1.31) 1.15 (0.96 , 1.37) 1.17 (0.98 , 1.41) 1.10 (0.93 , 1.30) 0.99 (0.87 , 1.12) Royal Columbian 1.21 (0.97 , 1.51) 1.27 (1.05 , 1.54) 1.39 (1.18 , 1.63) 1.27 (1.05 , 1.54) 1.11 (0.99 , 1.25) Lions Gate 1.21 (1.02 , 1.45) 1.16 (1.02 , 1.32) 1.42 (1.30 , 1.56) 1.27 (1.13 , 1.42) 1.08 (0.98 , 1.20) Langley Memorial 1.31 (1.30 , 1.32) 1.46 (1.45 , 1.47) 1.67 (1.64 , 1.71) 1.39 (1.24 , 1.56) 1.25 (1.14 , 1.36) Surrey Memorial 1.29 (1.23 , 1.36) 1.43 (1.38 , 1.48) 1.46 (1.39 , 1.54) 1.29 (1.21 , 1.38) 1.17 (1.11 , 1.24) Peace Arch District 1.45 (1.45 , 1.45) 1.29 (1.28 , 1.30) 1.59 (1.54 , 1.64) 1.37 (1.21 , 1.55) 1.22 (1.10 , 1.35) Royal Jubilee 1.47 (1.20 , 1.81) 1.36 (1.26 , 1.47) 1.49 (1.41 , 1.56) 1.20 (1.00 , 1.43) 1.14 (0.99 , 1.31) Cowichan District 1.85 (1.84 , 1.85) 1.58 (1.57 , 1.59) 1.78 (1.74 , 1.81) 1.27 (1.06 , 1.52) 1.17 (1.02 , 1.34) Vernon Jubilee 1.80 (1.45 , 2.24) 1.51 (1.29 , 1.76) 1.56 (1.34 , 1.83) 1.14 (0.93 , 1.41) 1.05 (0.85 , 1.30) Kelowna 1.50 (1.49 , 1.50) 1.39 (1.38 , 1.40) 1.49 (1.47 , 1.50) 1.16 (0.98 , 1.38) 1.13 (0.98 , 1.30) Penticton 1.99 (1.81 , 2.19) 1.53 (1.44 , 1.62) 1.55 (1.45 , 1.66) 1.13 (0.95 , 1.36) 1.06 (0.91 , 1.22) Royal Inland 2.17 (1.93 , 2.44) 1.93 (1.79 , 2.07) 1.99 (1.89 , 2.10) 1.38 (1.15 , 1.67) 1.36 (1.18 , 1.57) Nanaimo 2.15 (2.00 , 2.31) 1.74 (1.58 , 1.91) 1.87 (1.69 , 2.06) 1.38 (1.16 , 1.64) 1.31 (1.14 , 1.51) St Josephs 2.02 (2.02 , 2.02) 1.61 (1.59 , 1.62) 1.72 (1.70 , 1.75) 1.30 (1.08 , 1.55) 1.15 (0.99 , 1.33) Campbell River 2.37 (2.17 , 2.58) 2.07 (1.81 , 2.37) 2.17 (1.88 , 2.51) 1.50 (1.22 , 1.84) 1.27 (1.08 , 1.49) Chilliwack 1.69 (1.24 , 2.32) 1.63 (1.26 , 2.11) 1.68 (1.33 , 2.12) 1.20 (0.93 , 1.56) 1.12 (0.91 , 1.40) Ridge Meadows 1.80 (1.80 , 1.80) 2.00 (1.99 , 2.00) 2.19 (2.15 , 2.23) 1.75 (1.52 , 2.03) 1.58 (1.41 , 1.78) Abbotsford  1.37 (1.33 , 1.41) 1.52 (1.48 , 1.56) 1.60 (1.56 , 1.64) 1.25 (1.10 , 1.42) 1.11 (0.98 , 1.25) Fort St. John 1.67 (1.34 , 2.09) 2.07 (1.80 , 2.38) 2.40 (1.98 , 2.91) 1.61 (1.26 , 2.07) 1.48 (1.16 , 1.89) Prince George 1.81 (1.60 , 2.06) 1.88 (1.70 , 2.07) 2.02 (1.86 , 2.19) 1.38 (1.12 , 1.68) 1.22 (1.04 , 1.44) East Kootenay 1.59 (1.37 , 1.84) 1.41 (1.29 , 1.54) 1.52 (1.39 , 1.66) 1.11 (0.90 , 1.37) 1.11 (0.92 , 1.33)  131   Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Kootenay Boundary 2.00 (1.48 , 2.70) 1.68 (1.35 , 2.10) 1.69 (1.29 , 2.22) 1.27 (0.91 , 1.77) 1.24 (0.91 , 1.68) Mills Memorial 1.95 (1.74 , 2.19) 1.99 (1.81 , 2.20) 1.99 (1.80 , 2.20) 1.21 (0.91 , 1.62) 1.16 (0.93 , 1.44) Sex (reference=Female)           Male    1.55 (1.49 , 1.61) 1.56 (1.50 , 1.61) 1.55 (1.50 , 1.61) 1.51 (1.46 , 1.56) Age category (reference=20-24)         0-4    0.52 (0.32 , 0.84) 0.50 (0.31 , 0.81) 0.50 (0.31 , 0.81) 0.43 (0.26 , 0.70) 5-9    0.20 (0.12 , 0.32) 0.19 (0.12 , 0.32) 0.19 (0.12 , 0.32) 0.20 (0.12 , 0.32) 10-14    0.17 (0.11 , 0.27) 0.17 (0.11 , 0.27) 0.17 (0.11 , 0.27) 0.19 (0.12 , 0.30) 15-19    0.51 (0.37 , 0.69) 0.51 (0.37 , 0.70) 0.51 (0.37 , 0.70) 0.55 (0.40 , 0.75) 25-29    1.02 (0.77 , 1.35) 0.99 (0.74 , 1.32) 0.99 (0.75 , 1.32) 0.95 (0.72 , 1.26) 30-34     1.48 (1.23 , 1.78) 1.42 (1.17 , 1.71) 1.42 (1.18 , 1.72) 1.32 (1.10 , 1.59) 35-39     1.46 (1.16 , 1.84) 1.41 (1.12 , 1.77) 1.41 (1.12 , 1.78) 1.27 (1.01 , 1.59) 40-44    2.26 (1.86 , 2.76) 2.21 (1.82 , 2.70) 2.22 (1.82 , 2.70) 1.90 (1.57 , 2.31) 45-49    3.37 (2.82 , 4.02) 3.31 (2.78 , 3.95) 3.32 (2.78 , 3.96) 2.72 (2.29 , 3.23) 50-54    4.95 (4.15 , 5.89) 4.90 (4.12 , 5.83) 4.91 (4.12 , 5.84) 3.86 (3.27 , 4.56) 55-59    7.91 (6.63 , 9.42) 7.89 (6.62 , 9.40) 7.89 (6.62 , 9.41) 5.84 (4.92 , 6.92) 60-64    11.4 (9.50 , 13.8) 11.4 (9.50 , 13.9) 11.5 (9.40 , 13.9) 7.9 (6.6 , 9.5) 65-69    17.9 (14.9 , 21.4) 17.9 (14.9 , 21.5) 17.9 (14.9 , 21.5 11.0 (9.2 , 13.1) 70+    29.8 (25.0 , 35.5) 29.4 (24.7 , 35.1) 29.4 (24.6 , 35.2) 16.1 (13.6 , 19.1) Dissemination area income quintile (reference=1, highest)        2       1.25 (1.16 , 1.36) 1.24 (1.15 , 1.34) 1.20 (1.11 , 1.30) 3       1.42 (1.32 , 1.54) 1.40 (1.30 , 1.52) 1.32 (1.23 , 1.42) 4       1.69 (1.54 , 1.86) 1.67 (1.52 , 1.83) 1.51 (1.38 , 1.65)  132   Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) 5       2.20 (1.98 , 2.45) 2.17 (1.95 , 2.42) 1.87 (1.7 , 2.04) Education (LHA-level, per 10% difference)          % age 19-54 with no high school         0.78 (0.63 , 0.96) 0.85 (0.72 , 1.01) Ethnic composition (LHA-level, per 10% difference)          % aboriginal         1.11 (1.02 , 1.21) 1.14 (1.06 , 1.23) % recent immigrant          1.04 (0.98 , 1.11) 1.02 (0.96 , 1.09) Aggregated Diagnosis Groups             1. Time Limited: Minor           0.88 (0.84 , 0.92) 2. Time Limited: Minor-Primary Infections        1.03 (0.99 , 1.07) 3. Time Limited: Major           1.83 (1.65 , 2.03) 4. Time Limited: Major-Primary Infections        1.71 (1.60 , 1.82) 5. Allergies            0.78 (0.68 , 0.88) 6. Asthma           1.15 (1.06 , 1.25) 7. Likely to Recur: Discrete           1.05 (1.00 , 1.11) 8. Likely to Recur: Discrete-Infections        1.07 (1.01 , 1.12) 9. Likely to Recur: Progressive           1.38 (1.26 , 1.51) 10. Chronic Medical: Stable           0.99 (0.95 , 1.04) 11. Chronic Medical: Unstable           1.99 (1.89 , 2.09) 12. Chronic Specialty: Stable-Orthopedic        0.77 (0.69 , 0.87) 13. Chronic Specialty: Stable-Ear, Nose, Throat        1.02 (0.81 , 1.27) 14. Chronic Specialty: Stable-Eye           1.02 (0.93 , 1.11) 16. Chronic Specialty: Unstable-Orthopedic        0.95 (0.81 , 1.12)  133   Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) 17. Chronic Specialty: Unstable-Ear, Nose, Throat        0.99 (0.58 , 0.47) 18. Chronic Specialty: Unstable-Eye           0.99 (0.92 , 1.07) 20. Dermatologic           0.76 (0.72 , 0.80) 21. Injuries/Adverse Effects: Minor            1.02 (0.98 , 1.06) 22. Injuries/Adverse Effects: Major           1.25 (1.17 , 1.35) 23. Psychosocial: Time Limited, Minor        1.13 (1.05 , 1.22) 24. Psychosocial: Recurrent or Persistent, Stable        1.27 (1.17 , 1.12) 25. Psychosocial: Recurrent or Persistent, Unstable        1.78 (2.83 , 2.63) 26. Signs/Symptoms: Minor            1.16 (1.11 , 1.20) 27. Signs/Symptoms: Uncertain           0.98 (0.94 , 1.02) 28. Signs/Symptoms: Major            1.27 (1.23 , 1.31) 29. Discretionary           0.78 (0.74 , 0.83) 30. See and Reassure           0.71 (0.63 , 0.81) 31. Prevention/Administrative            0.99 (0.93 , 1.05) 32. Malignancy            3.53 (3.31 , 3.77) 33. Pregnancy            0.56 (0.41 , 0.77) 34. Dental           1.25 (1.13 , 1.40) Constant 1.82 x10-3 (1.82 x10-3 , 1.82 x10-3) 2.95 x10-4 (2.48 x10-4 , 3.51 x10-4) 1.87 x10-4 (1.54 x10-4 , 2.27 x10-4) 2.21 x10-4 (1.62 x10-4 , 3.00 x10-4) 2.00 x10-4 (1.50 x10-4 , 2.68 x10-4) Pseudo R2 0.004 0.102 0.106 0.107 0.161     134  Table 7-7 Logistic models of treatable mortality  Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Network (reference=Richmond)            Vancouver General 0.90 (0.81 , 1.01) 0.93 (0.81 , 1.07) 0.96 (0.84 , 1.10) 0.93 (0.80 , 1.08) 0.85 (0.75 , 0.96) Royal Columbian 0.98 (0.71 , 1.35) 1.03 (0.78 , 1.36) 1.12 (0.87 , 1.44) 1.07 (0.82 , 1.40) 0.94 (0.77 , 1.15) Lions Gate 1.02 (0.94 , 1.10) 0.96 (0.91 , 1.02) 1.18 (1.12 , 1.24) 1.09 (0.96 , 1.25) 0.95 (0.84 , 1.08) Langley Memorial 1.05 (1.05 , 1.06) 1.19 (1.18 , 1.20) 1.35 (1.32 , 1.38) 1.24 (1.07 , 1.44) 1.09 (0.93 , 1.27) Surrey Memorial 1.08 (1.00 , 1.17) 1.21 (1.18 , 1.23) 1.23 (1.15 , 1.32) 1.17 (1.06 , 1.31) 1.07 (0.96 , 1.21) Peace Arch District 1.12 (1.12 , 1.13) 0.97 (0.96 , 0.97) 1.19 (1.16 , 1.22) 1.11 (0.95 , 1.28) 0.99 (0.85 , 1.14) Royal Jubilee 1.15 (1.00 , 1.33) 1.04 (0.97 , 1.12) 1.14 (1.08 , 1.21) 1.02 (0.83 , 1.25) 1.00 (0.80 , 1.24) Cowichan District 1.13 (1.13 , 1.13) 0.94 (0.94 , 0.95) 1.05 (1.04 , 1.07) 0.88 (0.70 , 1.10) 0.80 (0.64 , 1.01) Vernon Jubilee 1.48 (1.22 , 1.79) 1.20 (1.06 , 1.35) 1.24 (1.10 , 1.39) 1.06 (0.83 , 1.36) 0.98 (0.75 , 1.29) Kelowna 0.98 (0.98 , 0.98) 0.89 (0.88 , 0.89) 0.94 (0.93 , 0.96) 0.84 (0.68 , 1.03) 0.84 (0.68 , 1.03) Penticton 1.70 (1.45 , 2.00) 1.23 (1.18 , 1.29) 1.25 (1.21 , 1.29) 1.09 (0.87 , 1.37) 1.03 (0.81 , 1.32) Royal Inland 1.77 (1.56 , 2.00) 1.54 (1.44 , 1.64) 1.58 (1.50 , 1.66) 1.26 (1.00 , 1.59) 1.24 (0.98 , 1.57) Nanaimo 1.78 (1.60 , 1.99) 1.39 (1.22 , 1.58) 1.49 (1.31 , 1.69) 1.27 (1.02 , 1.58) 1.22 (0.97 , 1.54) St Josephs 1.41 (1.41 , 1.41) 1.07 (1.07 , 1.08) 1.15 (1.13 , 1.16) 1.00 (0.80 , 1.25) 0.87 (0.69 , 1.10) Campbell River 1.68 (1.59 , 1.77) 1.44 (1.32 , 1.58) 1.51 (1.37 , 1.67) 1.21 (0.96 , 1.54) 1.02 (0.80 , 1.31) Chilliwack 1.40 (1.02 , 1.92) 1.32 (1.02 , 1.70) 1.35 (1.07 , 1.69) 1.13 (0.85 , 1.50) 1.05 (0.79 , 1.39) Ridge Meadows 1.46 (1.46 , 1.46) 1.64 (1.63 , 1.64) 1.78 (1.74 , 1.81) 1.60 (1.33 , 1.92) 1.47 (1.22 , 1.76) Abbotsford  1.24 (1.16 , 1.33) 1.38 (1.37 , 1.40) 1.45 (1.43 , 1.47) 1.30 (1.10 , 1.54) 1.18 (0.98 , 1.42) Fort St. John 1.27 (0.74 , 2.18) 1.62 (1.07 , 2.47) 1.88 (1.17 , 3.02) 1.52 (0.92 , 2.51) 1.38 (0.79 , 2.44) Prince George 1.31 (1.22 , 1.40) 1.36 (1.30 , 1.42) 1.46 (1.37 , 1.55) 1.16 (0.92 , 1.47) 1.02 (0.78 , 1.33) East Kootenay 0.92 (0.58 , 1.47) 0.80 (0.49 , 1.29) 0.85 (0.50 , 1.46) 0.73 (0.41 , 1.29) 0.73 (0.41 , 1.32)  135   Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Kootenay Boundary 1.44 (1.01 , 2.04) 1.17 (0.90 , 1.53) 1.18 (0.85 , 1.62) 1.03 (0.70 , 1.53) 1.00 (0.67 , 1.48) Mills Memorial 1.44 (1.06 , 1.96) 1.48 (1.10 , 1.99) 1.47 (1.09 , 2.00) 0.98 (0.63 , 1.55) 0.95 (0.64 , 1.41) Sex (reference=Female)           Male    1.43 (1.37 , 1.50) 1.44 (1.38 , 1.51) 1.44 (1.38 , 1.51) 1.34 (1.28 , 1.40) Age category (reference=20-24)         0-4    2.00 (0.65 , 6.12) 1.94 (0.63 , 5.94) 1.94 (0.63 , 5.95) 1.53 (0.49 , 4.76) 5-9    0.59 (0.17 , 2.03) 0.58 (0.17 , 2.00) 0.58 (0.17 , 2.00) 0.58 (0.17 , 2.00) 10-14    0.66 (0.24 , 1.83) 0.66 (0.24 , 1.82) 0.66 (0.24 , 1.83) 0.70 (0.25 , 1.95) 15-19    0.37 (0.13 , 1.10) 0.37 (0.13 , 1.10) 0.37 (0.13 , 1.10) 0.40 (0.13 , 1.17) 25-29    1.21 (0.68 , 2.15) 1.17 (0.66 , 2.09) 1.17 (0.66 , 2.09) 1.14 (0.64 , 2.03) 30-34     2.21 (1.15 , 4.27) 2.12 (1.10 , 4.09) 2.12 (1.10 , 4.11) 2.00 (1.04 , 3.87) 35-39     3.57 (1.96 , 6.48) 3.45 (1.90 , 6.25) 3.45 (1.90 , 6.27) 3.15 (1.74 , 5.70) 40-44    6.47 (3.71 , 11.2) 6.33 (3.63 , 11.0) 6.34 (3.63 , 11.0) 5.50 (3.17 , 9.54) 45-49    10.8 (6.3 , 18.5) 10.6 (6.2 , 18.2) 10.6 (6.2 , 18.3) 8.7 (5.1 , 14.9) 50-54    20.3 (12.0 , 34.4) 20.1 (11.8 , 34.2) 20.1 (11.9 , 34.2) 15.6 (9.3 , 26.4) 55-59    34.1 (20.4 , 57.0) 34.0 (20.3 , 57.0) 34.1 (20.3 , 57.0) 24.4 (14.5 , 40.8) 60-64    47.5 (27.9 , 80.7) 47.5 (27.9 , 80.9) 47.6 (28.0 , 80.9) 30.8 (18.2 , 52.0) 65-69    82.0 (48.8 , 138) 82.1 (48.8 , 138) 82.1 (48.9 , 138) 46.2 (27.5 , 77.4) 70+    142 (84 , 239) 140 (83 , 237) 140 (83 , 237) 68 (40 , 115) Dissemination area income quintile (reference=1, highest)        2       1.31 (1.17 , 1.47) 1.31 (1.17 , 1.46) 1.27 (1.13 , 1.42) 3       1.49 (1.34 , 1.64) 1.48 (1.35 , 1.63) 1.4 (1.27 , 1.53) 4       1.75 (1.57 , 1.96) 1.74 (1.57 , 1.94) 1.59 (1.44 , 1.77)  136   Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) 5       2.21 (1.97 , 2.48) 2.20 (1.96 , 2.47) 1.93 (1.73 , 2.16) Education (LHA-level, per 10% difference)          % age 19-54 with no high school         0.90 (0.69 , 1.16) 0.97 (0.75 , 1.00) Ethnic composition (LHA-level, per 10% difference)          % aboriginal         1.01 (0.90 , 1.13) 1.03 (0.91 , 1.17) % recent immigrant          1.10 (1.01 , 1.19) 1.08 (0.99 , 1.17) Aggregated Diagnosis Groups             1. Time Limited: Minor           0.90 (0.85 , 0.96) 2. Time Limited: Minor-Primary Infections        0.98 (0.91 , 1.05) 3. Time Limited: Major           2.05 (1.73 , 2.43) 4. Time Limited: Major-Primary Infections        1.72 (1.56 , 1.90) 5. Allergies            0.75 (0.63 , 0.91) 6. Asthma           0.93 (0.79 , 1.10) 7. Likely to Recur: Discrete           1.06 (0.99 , 1.14) 8. Likely to Recur: Discrete-Infections        1.13 (1.03 , 1.24) 9. Likely to Recur: Progressive           1.95 (1.78 , 2.13) 10. Chronic Medical: Stable           1.10 (1.01 , 1.21) 11. Chronic Medical: Unstable           1.95 (1.82 , 2.10) 12. Chronic Specialty: Stable-Orthopedic        0.83 (0.68 , 1.00) 13. Chronic Specialty: Stable-Ear, Nose, Throat        1.14 (0.82 , 1.60) 14. Chronic Specialty: Stable-Eye           0.94 (0.83 , 1.07) 16. Chronic Specialty: Unstable-Orthopedic        0.90 (0.63 , 1.28)  137   Networks only Model 1 Model 2 Model 3 Model 4  Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) Mortality ratio (95% CI) 17. Chronic Specialty: Unstable-Ear, Nose, Throat        0.99 (0.58 , 1.36) 18. Chronic Specialty: Unstable-Eye           1.24 (1.09 , 1.41) 20. Dermatologic           0.65 (0.58 , 0.73) 21. Injuries/Adverse Effects: Minor            0.93 (0.86 , 1.00) 22. Injuries/Adverse Effects: Major           1.25 (1.14 , 1.38) 23. Psychosocial: Time Limited, Minor        0.94 (0.85 , 1.05) 24. Psychosocial: Recurrent or Persistent, Stable        1.27 (1.07 , 0.98) 25. Psychosocial: Recurrent or Persistent, Unstable        1.78 (2.20 , 1.98) 26. Signs/Symptoms: Minor            1.04 (0.96 , 1.12) 27. Signs/Symptoms: Uncertain           0.92 (0.87 , 0.98) 28. Signs/Symptoms: Major            1.25 (1.19 , 1.31) 29. Discretionary           0.77 (0.70 , 0.84) 30. See and Reassure           0.65 (0.53 , 0.80) 31. Prevention/Administrative            1.07 (0.98 , 1.170 32. Malignancy            3.55 (3.23 , 3.89) 33. Pregnancy            0.51 (0.28 , 0.95) 34. Dental           1.18 (0.94 , 1.48) Constant 7.54 x10-4 (7.54 x10-4 , 7.55 x10-4) 3.09 x10-5 (1.83 x10-5 , 5.20 x10-5) 1.91 x10-5 (1.12 x10-5 , 3.24 x10-5) 2.10 x10-5 (1.13 x10-5 , 3.88 x10-5) 2.12 x10-5 (1.16 x10-5 , 3.87 ,x10-5) Pseudo R2 0.003 0.112 0.115 0.116 0.166     138    Figure 7-2 Plots of all-cause, premature, and treatable mortality ratios and cost ratios (models 1-3) 139  Chapter 8 - Conclusion  The overarching goal of this thesis was to produce new knowledge about how and why health care costs vary by geography, and the extent to which we see variation in the province of British Columbia (BC). Policy makers in Canada face concerns about escalating health spending, and cost pressures are likely to continue given health care innovation and demographic pressures (199). Research in the US has found substantial unexplained geographic variation in health care costs, with estimates that up to 30% of health care spending is waste, based on the fact that high spending areas show neither improved quality nor better outcomes. Given this, identifying and reducing unwarranted variation has received widespread attention as a strategy to respond to cost pressures and improve system efficiency (3–6). However, whether similar unwarranted geographic variation in costs exists in a Canadian province was unknown. This thesis aimed to fill this gap. It first explored conceptually how various potential drivers of variation might be patterned geographically, and what is known from existing Canadian research. Subsequent analysis of patterns of service use, and resulting variation in costs helps determine the potential for geographically based planning and policies to improve health care delivery. In short, I found that in contrast to the United States, from which a large proportion of variations research derives, we see very little unwarranted variation in total cost by geography, regardless of the unit examined. There are, at the same time, substantial variations in the sub-components of service use that contribute to those overall totals. In this sense it appears that services are often substitutes for one another; as one example, people with lower supply of health care resources in their communities make greater use of inpatient services, and use a relatively greater proportion of generalist than specialist physician services. The question for policy-makers may then be to what extent such arrangements represent the realities of the vast geography of Canada and the inability to have metropolitan-style health care systems everywhere, or an opportunity to make structural change and improve care. Contributions Conceptual framework and literature review Some preparatory conceptual and methodological work was needed in support of my overarching objective. I began by considering existing conceptual frameworks useful for 140  understanding geographic variation in health care, and developed my own to address their limitations. Most importantly, rather than simply identifying potential causes of variation, I considered how these might be patterned geographically, and at what spatial scale. I also considered both population and health system factors in detail, and made the connection between patterns of service use, and resulting population and health system outcomes. This framework helped to structure a review of the Canadian literature (Chapter 3). I completed this review in order to identify gaps and methodological limitations of existing studies, as well as to help place my findings about differences in costs in the context of existing literature examining specific services and procedures (Chapter 3). In the process I provide a current picture of Canadian research in this area. Analysis in Chapters 5-7 was designed to address methodological limitations identified in existing research. I also used the conceptual framework in designing analyses in Chapter 5-7. It helped structure analysis aimed at refining units of analysis, and understanding the geographic patterning of health care use (Chapter 5 and 6). It was especially useful in selecting appropriate adjustment variables in analysis aimed at identifying unwarranted variation in health care costs, and in connecting variation in costs to health outcomes, while controlling for factors external to the health care system that shape population health (Chapter 7). The framework may be similarly useful in future research aimed at describing and understanding variation in health care. Importantly, it outlines the multiple related population and health care system factors that must be considered in interpreting any observed differences in service use and costs. Multispecialty networks and Health Service Delivery Areas as units of analysis  Based on patterns of physician and hospital service use, I identified 23 multispecialty physician networks in BC, each including at least 50,000 individuals and representing a system of care that encompasses a broad spectrum of services. There are some higher-level services (e.g. very specialized cardiac care) that are provincial in nature, but for most services, the networks provide a more local and focused lens for analysis, and were remarkably self-contained. This demonstrates that an approach first applied in Ontario has broad relevance (21).  In contrast to expectations, identified networks were almost completely nested within the 16 existing health regions (Health Service Delivery Areas (HSDAs)), and the magnitude of variation in costs differed little between these units. In other words, whether the perspective used is networks or HSDAs, the conclusion that there is little unwarranted variation in total 141  system cost holds. There are, of course, differences in these costs, but most of those differences can be explained by age, sex, and need for health care services.  This provides useful validation that HSDAs do correspond to patterns of service use. This is important as researchers can often only access data at the level of HSDA. This finding also supports the ongoing use of these units in national reporting (22). At the same time, as there are cases where more than one network is nested within an HSDA, networks may offer a more refined unit for some purposes. For example, networks may be used to structure accountability around existing patterns of care, and may have more resonance with providers than administrative boundaries. Distinct health service environments  Cluster analysis described in Chapter 6 revealed four distinct areas of the province (which we called health service environments), based only on the distribution of health care costs across categories of care. These corresponded to metropolitan centres, non-metropolitan areas that contain large community hospitals and greater physician supply, non-metropolitan areas with smaller hospitals and fewer physicians, and remote communities. Our analysis revealed marked differences in the mix of providers and settings of care, though differences in total cost were subtle. For example, non-metropolitan areas had lower use of specialist services, and higher use of hospital services, especially medical admissions. Non-metropolitan residents were also more likely to see a GP in hospital than metropolitan residents. There is evidence of substitution between care providers and settings based on resources available, since costs within categories vary more than total spending.  A high proportion of medical decisions fall into the “gray areas” of medicine, what Wennberg and others refer to as preference or supply-sensitive care (16). In these cases, variations may arise either because of differences in treatment preferences of patients or because of the lack of consensus in the best available evidence about appropriate treatment patterns (200). There are often no clear guidelines, for example, for how often a patient should receive care from a generalist vs. a specialist, or optimal length of stay in hospital. When resources available differ, so may the calculus of potential benefit or risk from not receiving a given service weighed against the time and effort required in accessing it. Researchers have described the effect of supplier-induced demand, where local resource supply, and not patient need, drives use of particular types of services (25,53,88). At its most extreme, this is a plausible cause of inefficiency in health care. However, there may also be more subtle cases 142  where practice patterns and culture are very reasonably shaped by resources that are locally available (both supply of physicians, but also material resources) (201).  For this reason, and in contrast to previous work that focused only on patient need and preferences as the only “warranted” drivers of variation (16), I suggest that some variation based on service delivery environment may also be warranted, provided outcomes are comparable and differences in costs are acceptable. As is described in Chapter 3, this is sometimes tacitly acknowledged by the inclusion of a variable capturing urban vs. rural locations as a control variable in analysis of variation in use, costs, or outcomes. However, such measures may offer only a rough proxy for the health service environment that more directly shapes use. The approach to categorizing areas based on cluster analysis described herein may therefore be useful to other researchers. Alternatively, researchers may choose to control for health resources such as the supply of acute care facilities and specialist physicians directly, though analyses here suggest this approach has lower explanatory power. Limited unwarranted variation  Influential US research has largely concluded that health care spending and service use vary substantially from place to place, and that much of this variation cannot be explained by differences in the health status of the underlying populations (17,172,202,203). Perhaps most importantly, places that spend more do not appear to have better health outcomes. In fact, some research suggests worse outcomes in the highest-spending regions (12).  At study outset, we hoped careful examination of unwarranted variation in health service delivery might also highlight places where health services could be provided more efficiently. We attempted to control as completely as possible for warranted sources of variation (16), namely population need and health service environment. Reassuringly, but somewhat surprisingly, our analysis revealed very limited unwarranted variation in health services use (or at least in the use of the types of services included in our analysis) in BC.  We should perhaps have expected that variation in cost across areas within a provincial health system would be modest. As is described in Chapter 7, many factors known to drive variation in the US (7,204), do not exist in Canada. In BC’s fee-for-service payment environment there is incentive for physicians to earn more by providing additional services or seeing patients more often, and price may drive a supply-driven preference for high cost procedures. However, our analysis suggests that the extent to which this occurs does not vary systematically between places. It may simply be that in the case of a reasonably homogenous provincial health system, 143  geography is simply not the right lens through which to identify inefficiencies. Greater gains may be achieved by examining decision-making at the level of provider or organization (17), or by scrutinizing high-use population groups that could be cared for more efficiently. In other words, there could still be considerable amounts of inefficient, or inappropriate, service provision nested within geographies in ways that are masked at the levels of aggregation used here.  No relationship between adjusted cost and health outcomes  Careful examination of the outcomes of health care is necessary in order to interpret our findings. Even though variation in costs is modest, if outcomes differ, questions about the equity, quality, and appropriateness of health care emerge. When analyses were limited to only age and sex adjusted values, it was tempting to conclude that a linear relationship exists between high cost and poor outcome areas. That is, there is higher health care spending in areas that have higher mortality rates. Pushing things further, however, this relationship was clearly confounded by unmeasured health status. When health status and non-medical determinants of health were also controlled for, there was no relationship between cost and outcomes.   This lack of relationship arises in part because of the absence of variation in spending once appropriate adjustments are made to observed spending. The implication is that at similar levels of spending, we did still observe differences in outcomes, even in fully adjusted models. Outlying hospital networks or regions, with much better than expected outcomes at similar levels of cost, may serve as case studies and offer lessons that can be applied elsewhere (5). One such example is Richmond, which is a clear outlier in analysis for this thesis, as well as previous research (5). What these analyses cannot tell us is whether its status as an outlier is explained by an exceptionally high performing health care system, or factors outside the health care system we could not adequately measure and adjust for.  This analysis makes clear that while there is only modest regional variation in health care costs in BC, there are large differences in mortality outcomes among hospital networks and health regions. Our conclusions are limited by the fact that our outcome measures were relatively blunt. This analysis uses the concepts of premature and treatable mortality as a best available option for an initial look at he relationship between variations and outcomes relevant to health system performance (169). Despite intended improvement over all-cause mortality, even differences in treatable mortality may be explained by incompletely-measured factors outside the health care system (20), suggested for example by a stronger income gradient for this 144  outcome. More direct measures of service quality, outcomes (especially changes in morbidity), and patient experience are needed, but are not routinely collected. At the same time, as addressing variation in health is the ultimate goal, continued attention to factors outside the health care system is still needed. Strengths and limitations  Strengths and limitations specific to the methods and data used in each chapter have been discussed. Here I wish to highlight the strengths and limitations of the thesis as a whole.   Findings are based on our best possible use of population-based data for the entire province of British Columbia, which has a population of ~4.4 million. We took care to ensure that observation of variation was not obscured by units of analysis that do not reflect patterns of service use, and that examination of the role of health service environment was not blunted by reliance on existing urban/rural classifications not relevant to service delivery. We also had access to individual-level data for most pieces of the analyses which allowed more complete adjustment for population characteristics than is the case for most analyses of regional variation (5). Still, there are limits to what we could accomplish with administrative data collected for other primary purposes.  Examining health system costs alone does not reveal the degree of inconvenience, out-of-pocket expense, or hardship involved in accessing and using health care services, nor whether services received were acceptable to patients. While self-reported unmet health need is collected on the Canadian Community Health Survey and has been reported in several studies (148,205), there is no systematic collection of data on ease or difficulty in accessing health services in Canada. More direct measures of health system performance and accessibility are clearly needed. Ideally outcomes would take into account morbidity and prognosis (7) and could include measures of effectiveness for individual services.  We had no geographic information below the level of LHA, though we could attach individuals to hospitals based on patterns of use. LHAs represent large areas, especially in the sparsely populated north and interior. However, this also means relatively small population sizes even for these large areas, so it would not have been possible to calculate reliable cost and outcome measures for smaller geographic units. Ideally we would have identified the precise location of individuals assigned to each hospital, in order to calculate travel time and confirm plausibility of assignment. However, given high concordance in network assignment based on LHA of residence, we are confident in the validity of network assignment. 145   We were also somewhat limited by the fact that measures of health status available in health administrative data are by definition themselves derived from encounters with the health care system. US research has raised the possibility of an “up-diagnosis” bias, wherein residents of high-intensity (and high-cost) regions are assigned a larger number of diagnostic codes, and therefore appear sicker (195). In the study period there were no differences in payment mechanisms that might incentivize some areas to “up-code” patients. However, we have no way to determine if networks or HSDAs differ in their intensity of coding for other reasons.  While we excluded patients whose primary care providers were not paid fee-for-service, some groups of specialists are also compensated through alternate payment plans. This is more common for specific types of tertiary care (cancer care or psychiatric services, for example), and so all provincial residents with relevant conditions would access tertiary care compensated in the same way. Correcting for the proportion of physicians in each area identified in hospital data but not captured in fee-for-services did not change conclusions. Lack of data on specialist alternate payment plan compensation is therefor a relatively minor limitation for this analysis.  Of potentially greater concern is the fact that hospital costs based on Resource Intensity Weights may underestimate true differences in hospital costs. In the absence of data on global budgets and other direct payments to hospitals, we cannot assess differences in price or organizational efficiency. However, we can still be certain that there are no systematic differences among regions or hospital networks in volume or intensity of service provision. Implications for research and policy  These analyses show clearly that population health varies by geography, but there was no compelling story to tell about the relationship between these outcomes and health care costs. This is largely because of a lack of variation in (adjusted) spending. Outliers with relatively good or poor outcomes with no higher costs may provide case studies warranting further analysis, but this is beyond the scope of this research. Moreover, it remains uncertain if treatable mortality more directly reflects health system performance than all-cause mortality (20). To move forward, the collection of outcomes more directly relevant to health system performance, for example, Patient Reported Experience and Outcome Measures (PREMs and PROMs), remains a pressing need (190,206).  One of the analytic choices made for these analyses was to convert hospital use to costs using provincial-average rather than hospital or patient-specific costing (207). This approach is the standard, but does obscure the fact that hospitals of different size and location have 146  different costs, largely driven by the influence of fixed costs. Future research may aim at assessing whether hospital-specific conversion of stays to costs results in a different conclusion about variations. While our approach with RIW-derived costs was aimed at uncovering variation in service volume or use of more resource-intensive approaches to care, a different approach is needed to capture differences in price or organizational efficiency.  The results of the present analysis are not inconsistent with substantial variation in procedures and classes of care described in the scoping review (Chapter 2). Results simply suggest that health regions or hospital networks do not differ systematically in overall intensity of service provision, as would be reflected in costs. This implies that geography may not be the correct lens through which to identify opportunities for improved efficiency. Rather, targeting decision making at the level of individual providers or health care teams, and ensuring that all care is based on the best available evidence may be more promising future directions (17).   At the same time, there are clearly differences across the province in the relative mix of services used under the calm surface of relatively invariant total costs. The identification of similar health service use areas may be a useful tool for future policy development, in which it can be explicitly acknowledged where variations in approaches to care are (or are not) warranted. Conclusions  Fundamentally, all research involves the systematic investigation of variation at some level. The study of variation in health care may be helpful to identify opportunities for performance improvement. In their text on systems design, G. and D. Weinberg describe “The Fundamental Regulator Paradox” as follows:  “The task of a regulator is to eliminate variation, but this variation is the ultimate source of information about the quality of its work. Therefore, the better the job a regulator does the less information it gets about how to improve.”3 There may therefore be some comfort in the finding of modest variation among health regions and hospital networks, largely explained by population need and health service environment. Huge disparities in spending would be unacceptable within a centrally administered and collectively funded system. At the same time, this means that examination of variation in costs alone offers no clear path forward for improving health system performance.                                                3 Gerald M. Weinberg and Daniela Weinberg, The Design of Stable Systems (1979), 250 147   Results make clear that there are no exceptionally high- or low-cost areas in BC, suggesting systematically higher volume or more intensive service provision for populations with similar health needs. However, this does not mean that important variation does not exist and cannot be uncovered and addressed. It only means that there is no low-hanging fruit identifiable by area-level cost alone. It may be that a different lens (variation at the level of provider, among population groups, or focused on how specific diseases or conditions are treated) will prove more promising.  At the same time, the degree of variation in all-cause, premature, and treatable mortality is concerning. This might suggest that spending should in fact be higher in some areas. Though we sought to make best possible use of existing data, their limitations became abundantly clear. Outcome measures that more directly reflect system performance are needed. Otherwise, it remains ambiguous to what extent findings point to failures of hospital and physician care, or factors outside the health care system (narrowly defined), that would not be modified through increased investment in categories of care tracked herein.  While geographic variations research draws attention to the issues of inconsistent quality and uneven costs, ultimately targeting specific regions for cost savings is not supported by this research. In the end, broader policy reforms aimed at quality and efficiencyfor example, fixing distortions created by outdated payment models, and strengthening accountability for outcomes may be more effective at reducing whatever unwarranted variations in practice patterns are out there. 148  References 1.  Corallo AN, Croxford R, Goodman DC, et al. A systematic review of medical practice variation in OECD countries. Health Policy. 2014;114(1):5–14.  2.  Paul-Shaheen P, Clark JD, Williams D. Small area analysis: a review and analysis of the North American literature. Journal of Health Politics, Policy and Law. 1987;12(4):741–809.  3.  Roos N, Stukel TA. Too Much Health Care Can Be Harmful. Huffington Post. June 20, 2013. 4.  Hundert M, Heylar C, Closson T. The hidden waste in Ontario health care. Toronto Star. July 21, 2014. 5.  Canadian Institute for Health Information. Measuring the Level and Determinants of Health System Efficiency in Canada. Ottawa: CIHI; 2014. 6.  Health Council of Canada. Value for Money: Making Canadian Health Care Stronger. Toronto: Health Council of Canada; 2009. 7.  Manning WG, Norton EC, Wilk AS. Explaining Geographic Variation in Health Care Spending, Use and Quality, and Associated Methodological Challenges. Washington, D.C.: Institute of Medicine; 2012. 8.  Wennberg JE, Gittelsohn. Small area variations in health care delivery. Science. 1973;182(4117):1102–1108.  9.  Dartmouth Atlas of Health Care Working Group. The Dartmouth Atlas of Health care. Hanover, NH: Center for the Evaluative Clinical Sciences, Dartmouth Medical School; 1996. 10.  Wennberg JE. Geography And The Debate Over Medicare Reform. Health Affairs. 2002;Suppl Web :W96–114.  11.  Fisher E, Wennberg DE. The implications of regional variations in Medicare spending. Part 1: the content, quality, and accessibility of care. Annals of Internal Medicine. 2003;138(4):273–287.  12.  Fisher E, Wennberg DE, Stukel TA. The implications of regional variations in Medicare spending. Part 2: health outcomes and satisfaction with care. Annals of Internal Medicine. 2003;138(4):288–98.  13.  Sutherland JM, Fisher ES, Skinner JS. Getting past denial—the high cost of health care in the United States. New England Journal of Medicine. 2009;361(13):1227–1230.  14.  Fisher E, Bynum JPW, Skinner JS. Slowing the growth of health care costs-lessons from regional variation. New England Journal of Medicine. 2009;360(9):849–852.  149  15.  Bernstein J, Reschovsky J, White C. Geographic Variation in Health Care: Changing Policy Directions. Washington, D.C.: National Institute for Health Care Reform; 2011. 16.  Wennberg JE. Unwarranted variations in healthcare delivery: implications for academic medical centres. British Medical Journal. 2002;325(7370):961–964.  17.  Institute of Medicine (IOM). Variation in Health Care Spending: Target Decision Making, Not Geography. Washington, D.C.: The National Academies Press; 2013. 18.  Skinner J. Causes and consequences of regional variations in health care. In: Handbook of Health Economics, Volume 2. Oxford: Elsevier B.V.; 2012:45–93. 19.  Diehr P, Cain K. UW Biostatistics Working Paper Series. An Empirical Study of Small-Area Variation for ICD-9 Surgical Procedures. Seattle: UW Biostatistics Working Paper Series; 1994. 20.  Lavergne MR, McGrail K. What, if anything, does amenable mortality tell us about regional health system performance? Healthcare Policy. 2013;8(3):79–90.  21.  Stukel TA, Glazier R, Schultz SE. Multispecialty physician networks in Ontario. Open Medicine. 2013;7(2):40–55.  22.  Canadian Institute for Health Information. Health Indicators 2013. Ottawa: CIHI; 2013. 23.  Chassin M. Explaining Geographic Variations: The Enthusiasm Hypothesis. Medical Care. 1993;31(5):YS37–YS44.  24.  Wright JG, Hawker GA, Bombardier C, et al. Physician enthusiasm as an explanation for area variation in the utilization of knee replacement surgery. Medical Care. 1999;37(9):946–956.  25.  Yasaitis LC, Bynum JPW, Skinner JS. Association between physician supply, local practice norms, and outpatient visit rates. Medical Care. 2013;51(6):524–531.  26.  De Jong JD, Groenewegen PP, Westert GP. Mutual influences of general practitioners in partnerships. Social Science & Medicine. 2003;57(8):1515–1524.  27.  Anderson JG. Demographic factors affecting health services utilization: a causal model. Medical Care. 1973;11(2):104–120.  28.  Andersen RM. Revisiting the behavioral model and access to medical care: does it matter? Journal of Health and Social Behavior. 1995;36(1):1–10.  29.  Evans R, Stoddart GL. Producing health, consuming health care. Social Science & Medicine. 1990;31(12):1347–1363.  30.  Evans RG, Stoddart GL. Consuming research, producing policy? American Journal of Public Health. 2003;93(3):371–379.  150  31.  Macintyre S, Ellaway A, Cummins S. Place effects on health: how can we conceptualise, operationalise and measure them? Social Science & Medicine. 2002;55(1):125–139.  32.  Cummins S, Curtis S, Diez-Roux A V, et al. Understanding and representing “place” in health research: a relational approach. Social Science & Medicine. 2007;65(9):1825–1838.  33.  Diez Roux A V. Conceptual approaches to the study of health disparities. Annual Review of Public Health. 2012;33:41–58.  34.  Anthony DL, Herndon MB, Gallagher PM, et al. How much do patients’ preferences contribute to resource use? Health Affairs. 2009;28(3):864–873.  35.  Cutler D, Skinner J, Stern AD, et al. Physician beliefs and patient preferences: A new look at regional variation in health care spending. Cambridge: National Bureau of Economic Research Working Paper; 2013. 36.  Hawker GA, Wright JG, Coyte PC, et al. Determining the need for hip and knee arthroplasty: the role of clinical severity and patients’ preferences. Medical Care. 2001;39(3):206–216.  37.  De Jong J. Explaining medical practice variation: Social organization and institutional mechanisms. Utrecht: Nivel; 2009. 38.  Gottlieb DJ, Zhou W, Song Y, et al. Prices don’t drive regional Medicare spending variations. Health Affairs. 2010;29(3):537–543.  39.  Wennberg JE, Fisher ES, Stukel TA, et al. Use of Medicare claims data to monitor provider-specific performance among patients with severe chronic illness. Health Affairs. 2004;Suppl Vari:VAR5–VAR18.  40.  Bach P. A Map to Bad Policy - Hospital Efficiency Measures in the Dartmouth Atlas. New England Journal of Medicine. 2010;362(7):569–573.  41.  Feasby TE, Quan H, Ghali WA. Geographic variation in the rate of carotid endarterectomy in Canada. Stroke. 2001;32(10):2417–2422.  42.  Tonelli M, Klarenbach S, Manns B, et al. Residence location and likelihood of kidney transplantation. Canadian Medical Association Journal. 2006;175(5):478–482.  43.  Jimenez-Rubio D, Smith PC, Van Doorslaer E. Equity in health and health care in a decentralised context: evidence from Canada. Health Economics. 2008;17(3):377–392.  44.  McNiven C, Puderer H, Janes D. Geography Working Paper Series Census Metropolitan Area and Census Agglomeration Influenced Zones (MIZ): A Description of the Methodology. Ottawa: Statistics Canada, Geography Division; 2000. 151  45.  Ibáñez B, Librero J, Bernal-Delgado E, et al. Is there much variation in variation? Revisiting statistics of small area variation in health services research. BMC Health Services Research. 2009;9(60).  46.  Cain K, Diehr P. The Multiple Admission Factor (MAF) in Small Area Variation Analysis. Seattle: UW Biostatistics Working Paper Series; 1992. 47.  Diehr P, Cain K, Ye Z, et al. Small area variation analysis. Methods for comparing several diagnosis-related groups. Medical Care. 1993;31(5 Suppl):YS45–53.  48.  Carriere KC, Roos LL. A method of comparison for standardized rates of low-incidence events. Medical Care. 1997;35(1):57–69.  49.  Carriere KC, Roos LL. Comparing standardized rates of events. American journal of Epidemiology. 1994;140(5):472–482.  50.  Larsen K, Merlo J. Appropriate assessment of neighborhood effects on individual health: integrating random and fixed effects in multilevel logistic regression. American Journal of Epidemiology. 2005;161(1):81–88.  51.  Coyte PC, Croxford R, Asche C V, et al. Physician and population determinants of rates of middle-ear surgery in Ontario. JAMA. 2001;286(17):2128–2135.  52.  Lopushinsky SR, Austin PC, Rabeneck L, et al. Regional variation in surgery for gastroesophageal reflux disease in Ontario. Surgical Innovation. 2007;14(1):35–40.  53.  Alter D, Stukel TA, Newman A. The relationship between physician supply, cardiovascular health service use and cardiac disease burden in Ontario: supply-need mismatch. Canadian Journal of Cardiology. 2008;24(3):187–193.  54.  Bederman SS, Coyte PC, Kreder HJ, et al. Who’s in the driver's seat? The influence of patient and physician enthusiasm on regional variation in degenerative lumbar spinal surgery: a population-based study. Spine. 2011;36(6):481–489.  55.  Goel V, Williams JI, Anderson GM, et al. The ICES Practice Atlas 2nd Edition: Patterns of Health Care in Ontario. Toronto: Institute for Clinical Evaluative Sciences; 1996. 56.  Johansen H, Bernier J, Fines P, et al. Variations by health region in treatment and survival after heart attack. Health Reports. 2009;20(2):29–34.  57.  Brownell M. Tonsillectomy rates for Manitoba children: temporal and spatial variations. Healthcare Management Forum. 2002;Suppl:21–26.  58.  Vanasse A, Niyonsenga T, Courteau J, et al. Spatial variation in the management and outcomes of acute coronary syndrome. BMC Cardiovascular Disorders. 2005;5(21).  59.  Hanley GE, Janssen PA, Greyson D. Regional variation in the cesarean delivery and assisted vaginal delivery rates. Obstetrics and Gynecology. 2010;115(6):1201–1208.  152  60.  Cree M, Yang Q, Scharfenberger A, et al. Variations in treatment of femoral neck fractures in Alberta. Canadian Journal of Surgery. 2002;45(4):248–254.  61.  Blais R. Variations in surgical rates in Quebec: does access to teaching hospitals make a difference? CMAJ. 1993;148(10):1729–1736.  62.  Chan B, Anderson G, Dales RE. Spirometry utilization in Ontario: practice patterns and policy implications. CMAJ. 1997;156(2):169–176.  63.  Goel V, Iron K, Williams JI. Enthusiasm or uncertainty: small area variations in the use of mammography services in Ontario, Canada. Journal of Epidemiology & Community Health. 1997;51(4):378–382.  64.  Hilsden RJ. Patterns of use of flexible sigmoidoscopy, colonoscopy and gastroscopy: a population-based study in a Canadian province. Canadian Journal of Gastroenterology. 2004;18(4):213–219.  65.  Hilsden RJ, Romagnuolo J, May GR. Patterns of use of endoscopic retrograde cholangiopancreatography in a Canadian province. Canadian Journal of Gastroenterology. 2004;18(10):619–624.  66.  Pefoyo AJK, Laurier C, Rivard M. [Determinants of geographic variations in asthma medication in Quebec]. Analyse des variations geographiques de l’usage des medicaments de l'asthme au Quebec. Revue des Maladies Respiratoires. 2010;27(1):49–62.  67.  Morgan SG, Cunningham CM, Hanley GE. Individual and contextual determinants of regional variation in prescription drug use: an analysis of administrative data from British Columbia. PloS one. 2010;5(12):e15883.  68.  Tamblyn R, McLeod PJ, Abrahamowicz M, et al. Questionable prescribing for elderly patients in Quebec. CMAJ. 1994;150(11):1801–1809.  69.  Gaudette LA, Gao R-N, Spence A, et al. Declining use of mastectomy for invasive breast cancer in Canada, 1981-2000. Canadian Journal of Public Health. 2004;95(5):336–340.  70.  Pilote L, Merrett P, Karp I, et al. Cardiac procedures after an acute myocardial infarction across nine Canadian provinces. The Canadian Journal of Cardiology. 2004;20(5):491–500.  71.  Hogan DB, Maxwell CJ, Fung TS, et al. Regional variation in the use of medications by older Canadians--a persistent and incompletely understood phenomena. Pharmacoepidemiology and Drug Safety. 2003;12(7):575–582.  72.  Myers T, Godin G, Lambert J, et al. Sexual risk and HIV-testing behaviour by gay and bisexual men in Canada. AIDS Care. 1996;8(3):297–309.  73.  Menzies D, Lewis M, Oxlade O. Costs for tuberculosis care in Canada. Canadian Journal of Public Health. 2008;99(5):391–396.  153  74.  Vanasse A, Courteau J, Cohen AA, et al. Rural-urban disparities in the management and health issues of chronic diseases in Quebec (Canada) in the early 2000s. Rural and Remote Health. 2010;10(4):1548.  75.  Dagenais P, Vanasse A, Courteau J, et al. Disparities between rural and urban areas for osteoporosis management in the province of Quebec following the Canadian 2002 guidelines publication. Journal of Evaluation in Clinical Practice. 2010;16(3):438–444.  76.  Brownell M, Kozyrkyj A, Roos N, et al. Health service utilization by Manitoba children. Canadian Journal of Public Health. 2002;93 Suppl 2:S57–62.  77.  Veugelers PJ, Yip AM, Elliott DC. Geographic variation in health services use in Nova Scotia. Chronic Diseases in Canada. 2003;24(4):116–123.  78.  Veugelers PJ, Yip AM, Burge FI. Inequalities in health and health services delivery: a multilevel study of primary care and hypertension control. Chronic Diseases in Canada. 2004;25(3-4):101–107.  79.  Boyle E, Badley EM, Glazier RH. The relationship between local availability and first-time use of specialists in an arthritis population. Canadian Journal of Public Health. 2006;97(3):210–213.  80.  Locker D, Clarke M. Geographic variations in dental services provided to older adults in Ontario, Canada. Community Dentistry and Oral Epidemiology. 1999;27(4):275–282.  81.  Taylor VM, Anderson GM, McNeney B, et al. Hospitalizations for back and neck problems: a comparison between the Province of Ontario and Washington State. Health services research. 1998;33(4 Pt 1):929–945.  82.  Mustard CA, Kozyrskyj AL, Barer ML, et al. Emergency department use as a component of total ambulatory care: a population perspective. Canadian Medical Association Journal. 1998;158(1):49–55.  83.  Coyte PC, Young W. Regional variations in the use of home care services in Ontario, 1993/95. CMAJ. 1999;161(4):376–380.  84.  Peterson S, Shapiro E, Roos NP. Regional variation in home care use in Manitoba. Canadian Journal on Aging. 2005;24 Suppl 1:69–80.  85.  Ng C, Chatwood S, Young TK. Arthritis in the Canadian Aboriginal population: north-south differences in prevalence and correlates. Preventing Chronic Disease. 2011;8(1):A04.  86.  Williams R, Kopala L, Malla A, et al. Medication decisions and clinical outcomes in the Canadian National Outcomes Measurement Study in Schizophrenia. Acta Psychiatrica Scandinavica. 2006;(430):12–21.  87.  Wen SW, Liu S, Marcoux S, et al. Trends and variations in length of hospital stay for childbirth in Canada. Canadian Medical Association Journal. 1998;158(7):875–880.  154  88.  Cohen A a, Vanasse A, Courteau J. Specialized cardiological care may be overutilized in urban areas of Quebec. European Journal of Preventive Cardiology. 2012;19(4):731–737.  89.  Finlayson M, Lix L, Finlayson G, et al. Trends in the utilization of specific health care services among older Manitobans: 1985 to 2000. Canadian Journal on Aging. 2005;24 Suppl 1:15–27.  90.  Tepper J, Pollett W, Jin Y, et al. Utilization rates for surgical procedures in rural and urban Canada. Canadian Journal of Rural Medicine. 2006;11(3):195–203.  91.  To T, Langer JC. Does access to care affect outcomes of appendicitis in children? - a population-based cohort study. BMC Health Services Research. 2010;10(250).  92.  Karunanayake CP, Pahwa P. Statistical modelling of mental distress among rural and urban seniors. Chronic diseases in Canada. 2009;29(3):118–27.  93.  Hardy CL, Kelly KD, Voaklander DC. Does rural residence limit access to mental health services? Rural Remote Health. 2011;11(4):1766.  94.  Pong RW, DesMeules M, Heng D, et al. Patterns of health services utilization in rural Canada. Chronic Diseases and Injuries in Canada. 2011;31 Suppl 1:1–36.  95.  Martens PJ, Brownell M, Kozyrskyj A. The virtual classroom: a summary of child health indicators. Canadian Journal of Public Health. 2002;93 Suppl 2:S77–80.  96.  Shapiro E, Roos LL. Using health care: rural/urban differences among the Manitoba elderly. The Gerontologist. 1984;24(3):270–274.  97.  Haggerty JL, Roberge D, Pineault R, et al. Features of primary healthcare clinics associated with patients’ utilization of emergency rooms: urban-rural differences. Healthc Policy. 2007;3(2):72–85.  98.  Gamble J-M, Eurich DT, Ezekowitz JA, et al. Patterns of care and outcomes differ for urban versus rural patients with newly diagnosed heart failure, even in a universal healthcare system. Circulation. Heart failure. 2011;4(3):317–323.  99.  Lum EY, Sharpe HM, Nilsson C, et al. Urban and rural differences in the management of asthma amongst primary care physicians in Alberta. The Canadian Journal of Clinical Pharmacology. 2007;14(3):e275–282.  100.  Booth GL, Hux JE, Fang J, et al. Time trends and geographic disparities in acute complications of diabetes in Ontario, Canada. Diabetes Care. 2005;28(5):1045–1050.  101.  Olatunde S, Leduc ER, Berkowitz J. Different practice patterns of rural and urban general practitioners are predicted by the General Practice Rurality Index. Canadian Journal of Rural Medicine. 2007;12(2):73–80.  155  102.  Barbera L, Elit L, Krzyzanowska M, et al. End of life care for women with gynecologic cancers. Gynecologic Oncology. 2010;118(2):196–201.  103.  Paszat LF, Mackillop WJ, Groome PA, et al. Radiotherapy for breast cancer in Ontario: rate variation associated with region, age and income. Clinical and Investigative Medicine. 1998;21(3):125–134.  104.  Menec VH, Nowicki S, Kalischuk A. Transfers to acute care hospitals at the end of life: do rural/remote regions differ from urban regions? Rural and Remote Health. 2010;10(1):1281.  105.  Heaman MI, Green CG, Newburn-Cook C V, et al. Social inequalities in use of prenatal care in Manitoba. Journal of Obstetrics and Gynaecology Canada. 2007;29(10):806–816.  106.  Dodek PM, Keenan SP, Norena M, et al. Structure, process, and outcome of all intensive care units within the province of British Columbia, Canada. Journal of intensive care medicine. 2010;25(3):149–155.  107.  Olivotto IA, Kan L, King S. Waiting for a diagnosis after an abnormal screening mammogram. SMPBC diagnostic process workgroup. Screening Mammography Program of British Columbia. Canadian Journal of Public Health. 2000;91(2):113–117.  108.  Wijeysundera HC, Stukel TA, Chong A, et al. Impact of clinical urgency, physician supply and procedural capacity on regional variations in wait times for coronary angiography. BMC Health Services Research. 2010;10(5).  109.  Mohan S, Chen G, Campbell NRC, et al. Regional variations in not treating diagnosed hypertension in Canada. The Canadian Journal of Cardiology. 2010;26(8):409–413.  110.  Lee DS, Johansen H, Gong Y, et al. Regional outcomes of heart failure in Canada. The Canadian Journal of Cardiology. 2004;20(6):599–607.  111.  Goodridge D, Lawson J, Rennie D, et al. Rural/urban differences in health care utilization and place of death for persons with respiratory illness in the last year of life. Rural and Remote Health. 2010;10(2):1349.  112.  Rowe BH, Voaklander DC, Marrie TJ, et al. Outcomes following chronic obstructive pulmonary disease presentations to emergency departments in Alberta: a population-based study. Canadian Respiratory Journal. 2010;17(6):295–300.  113.  Mitchell LA, Strain LA, Blandford AA. Indicators of home care use in urban and rural settings. Canadian Journal on Aging. 2007;26(3):275–280.  114.  Wilson DM, Truman CD, Huang J, et al. The possibilities and the realities of home care. Canadian Journal of Public Health. 2005;96(5):385–389.  115.  Ngamini Ngui A, Perreault M, Fleury M-J, et al. A multi-level study of the determinants of mental health service utilization. Revue d’epidemiologie et de sante publique. 2012;60(2):85–93.  156  116.  Mustard CA, Frohlich N. Socioeconomic status and the health of the population. Medical Care. 1995;33(12 Suppl):DS43–54.  117.  Stuart H. Access to physician treatment for a mental disorder: a regional analysis. Social Psychiatry and Psychiatric Epidemiology. 2000;35(2):61–70.  118.  Offord DR, Boyle MH, Szatmari P, et al. Ontario Child Health Study. II. Six-month prevalence of disorder and rates of service utilization. Archives of General Psychiatry. 1987;44(9):832–836.  119.  Van Walraven C V, Paterson JM, Kapral M, et al. Appropriateness of primary total hip and knee replacements in regions of Ontario with high and low utilization rates. CMAJ. 1996;155(6):697–706.  120.  Harvard SS, Hill WD, Buxton JA. Harm reduction product distribution in British Columbia. Canadian Journal of Public Health. 2008;99(6):446–450.  121.  Lin E, Chan B, Goering P. Variations in mental health needs and fee-for-service reimbursement for physicians in Ontario. Psychiatric Services. 1998;49(11):1445–1451.  122.  Roos NP, Roos LL. High and low surgical rates: risk factors for area residents. American Journal of Public Health. 1981;71(6):591–600.  123.  Frohlich N, Fransoo R, Roos N. Health service use in the Winnipeg Regional Health Authority: variations across areas in relation to health and socioeconomic status. Healthcare Management Forum. 2002;Suppl:9–14.  124.  Alter D, Austin PC, Tu J V. Community factors, hospital characteristics and inter-regional outcome variations following acute myocardial infarction in Canada. The Canadian Journal of Cardiology. 2005;21(3):247–255.  125.  Tataryn DJ, Roos NP, Black C. Utilization of physician resources for ambulatory care. Medical Care. 1995;33(12 Suppl):DS84–99.  126.  Badley EM, Canizares M, Mahomed N, et al. Provision of orthopaedic workforce and implications for access to orthopaedic services in Ontario. The Journal of Bone and Joint Surgery. 2011;93(9):863–870.  127.  Blais R, Breton J-J, Fournier M, et al. Are mental health services for children distributed according to needs? Canadian Journal of Psychiatry. 2003;48(3):176–186.  128.  Woltman KJ, Newbold KB. Immigrant women and cervical cancer screening uptake: a multilevel analysis. Canadian Journal of Public Health. 2007;98(6):470–475.  129.  Diaz-Granados N, Georgiades K, Boyle MH. Regional and individual influences on use of mental health services in Canada. Canadian Journal of Psychiatry. 2010;55(1):9–20.  157  130.  Kelly A, Jones W. Small area variation in the utilization of mental health services: implications for health planning and allocation of resources. Canadian Journal of Psychiatry. 1995;40(9):527–532.  131.  Ronen GM, Meaney BF. Pediatric neurology services in Canada: demand versus supply. Journal of Child Neurology. 2003;18(3):180–184.  132.  Vanasse A, Dagenais P, Niyonsenga T, et al. Bone mineral density measurement and osteoporosis treatment after a fragility fracture in older adults: regional variation and determinants of use in Quebec. BMC Musculoskeletal Disorders. 2005;6(33).  133.  To T, Feldman W, Young W, et al. Hospitalization rates of children with gastroenteritis in Ontario. Canadian Journal of Public Health. 1996;87(1):62–65.  134.  Rosenberg MW, Hanlon NT. Access and utilization: a continuum of health service environments. Social Science & Medicine. 1996;43(6):975–983.  135.  Klein-Geltink JE, Pogany LM, Barr RD, et al. Waiting times for cancer care in Canadian children: impact of distance, clinical, and demographic factors. Pediatric Blood & Cancer. 2005;44(4):318–327.  136.  Sirois M-J, Dionne CE, Lavoie A. Regional differences in rehabilitation needs, rehabilitation access, and physical outcomes among multiple trauma survivors. American Journal of Physical Medicine & Rehabilitation. 2009;88(5):387–398.  137.  Hartford K, Ross LL, Walld R. Regional variation in angiography, coronary artery bypass surgery, and percutaneous transluminal coronary angioplasty in Manitoba, 1987 to 1992: the funnel effect. Medical Care. 1998;36(7):1022–1032.  138.  Roos NP. Hysterectomy: variations in rates across small areas and across physicians’ practices. American Journal of Public Health. 1984;74(4):327–335.  139.  Langley GR, Minkin S, Till JE. Regional variation in nonmedical factors affecting family physicians’ decisions about referral for consultation. CMAJ. 1997;157(3):265–272.  140.  Rathwell T, Lee P, Sturtevant D. Does It Matter? Decision-maker Perceptions on the Impact of the Regional Training Centres. Healthcare Policy. 2008;3:106–117.  141.  Lemstra M, Neudorf C, Opondo J. Health disparity by neighbourhood income. Canadian Journal of Public Health. 2006;97(6):435–439.  142.  Kwon JS, Carey MS, Cook EF, et al. Are there regional differences in gynecologic cancer outcomes in the context of a single-payer, publicly-funded health care system? A population-based study. Canadian Journal of Public Health. 2008;99(3):221–226.  143.  Gomez D, Berube M, Xiong W, et al. Identifying targets for potential interventions to reduce rural trauma deaths: a population-based analysis. The Journal of Trauma. 2010;69(3):633–639.  158  144.  Lamarche PA, Pineault R, Haggerty J, et al. The experience of primary health care users: a rural-urban paradox. Canadian Journal of Rural Medicine. 2010;15(2):61–66.  145.  Keenan SP, Dodek P, Chan K, et al. Length of ICU stay for chronic obstructive pulmonary disease varies among large community hospitals. Intensive Care Medicine. 2003;29(4):590–595.  146.  Gawande A. The cost conundrum. What a Texas town can teach us about health care. New York Times. 2009;82(6):36–44.  147.  Goodman DC, Fisher E, Stukel TA, et al. The distance to community medical care and the likelihood of hospitalization: is closer always better? American Journal of Public Health. 1997;87(7):1144–1150.  148.  Sibley LM, Weiner JP. An evaluation of access to health care services along the rural-urban continuum in Canada. BMC Health Services Research. 2011;11(20).  149.  McDonald JT, Conde H. Does geography matter? The health service use and unmet health care needs of older Canadians. Canadian Journal on Aging. 2010;29(1):23–37.  150.  Pohar SL, Majumdar SR, Johnson JA. Health care costs and mortality for Canadian urban and rural patients with diabetes: population-based trends from 1993-2001. Clinical Therapeutics. 2007;29:1316–1324.  151.  Pong R, Pitblado JR. Geographic Distribution of Physicians in Canada: Beyond How Many and Where. Ottawa: CIHI; 2005. 152.  McGrail KM, Schaub P, Black C. The British Columbia Health Atlas. Vancouver: Centre for Health Services and Policy Research; 2004. 153.  Morgan S, Cunningham C, Hanley G, et al. The British Columbia Medical and Hospital Atlas. Vancouver: Centre for Health Services and Policy Research; 2009. 154.  Badley EM, Glazier R. Arthritis and Related Conditions in Ontario: ICES Research Atlas. Toronto: Institute for Clinical Evaluative Sciences; 2004. 155.  Hux J, Booth GL, Slaughter P, et al. Diabetes in Ontario: An ICES Practice Atlas. Toronto: Institute for Clinical Evaluative Sciences; 2003. 156.  Jaakkimainen L, Upshur REG, Klein-Geltink JE, et al. Primary care in Ontario: ICES atlas. Toronto: Institute for Clinical Evaluative Sciences; 2006. 157.  British Columbia Ministry of Health [Creator]. Consolidation File (MSP Registration & Premium Billing). Population Data BC [publisher] Data extract. MOH (2011).  158.  British Columbia Ministry of Health [Creator]. Medical Services Plan (MSP) Payment Information File. Population Data BC [publisher] Data extract. MOH (2011).  159  159.  British Columbia Ministry of Health [Creator]. Discharge Abstract Database (Hospital Separations). Population Data BC [publisher]. Data Extract. MOH (2011).  160.  Buntin MB, Zaslavsky AM. Too much ado about two-part models and transformation? Comparing methods of modeling Medicare expenditures. Journal of Health Economics. 2004;23(3):525–542.  161.  Zheng B, Agresti A. Summarizing the predictive power of a generalized linear model. Statistics in Medicine. 2000;19:1771–1781.  162.  John’s Hopkins Bloomberg School of Public Health. The Johns Hopkins ACG Case-Mix System Reference Manual Version 7.0. John’s Hopkins Bloomberg School of Public Health; 2005. 163.  Reid RJ, MacWilliam L, Verhulst L, et al. Performance of the ACG case-mix system in two Canadian provinces. Medical Care. 2001;39(1):86–99.  164.  Hanley GE, Morgan S, Reid RJ. Explaining prescription drug use and expenditures using the adjusted clinical groups case-mix system in the population of British Columbia, Canada. Medical Care. 2010;48(5):402–408.  165.  Nolte E, McKee CM. Measuring the health of nations: Updating an earlier analysis. Health Affairs. 2008;27(1):58–71.  166.  Tobias M, Yeh L-C. How much does health care contribute to health inequality in New Zealand? Australian and New Zealand Journal of Public Health. 2007;31(3):207–210.  167.  Rutstein DD, Berenberg W, Chalmers TC, et al. Measuring the quality of medical care. A clinical method. The New England Journal of Medicine. 1976;294(11):582–588.  168.  Charlton J, Silver R, Hartley R, et al. Geographical variation in mortality from conditions amenable to medical intervention in England and Wales. The Lancet. 1983;1(8326 Pt 1):691–696.  169.  Canadian Institute for Health Information. Health Indicators 2012. Ottawa, Ontario: CIHI; 2012. 170.  Allin S, Grignon M. Examining the Role of Amenable Mortality as an Indicator of Health System Effectiveness. Healthcare Policy. 2014;9(3):12–19.  171.  Bynum JPW, Bernal-Delgado E, Gottlieb D, et al. Assigning ambulatory patients and their physicians to hospitals: a method for obtaining population-based provider performance measurements. Health Services Research. 2007;42(1 Pt 1):45–62.  172.  Fisher ES, Staiger DO, Bynum JPW, et al. Creating accountable care organizations: the extended hospital medical staff. Health Affairs. 2007;26(1):w44–57.  173.  Landon B, Onnela J, Keating N. Using Administrative Data to Identify Naturally Occurring Networks of Physicians Bruce. Medical Care. 2013;51(8):715–721.  160  174.  Landon B, Keating N, Barnett M. Variation in Patient-Sharing Networks of Physicians Across the United States. JAMA. 2012;308(3):265–273.  175.  Katz A, Coster C De, Bogdanovic B, et al. Using administrative data to develop indicators of quality in family practice. Winnipeg: Manitoba Centre for Health Policy; 2004. 176.  Lavergne M, McGrail K, Peterson S, et al. Defining and measuring “full-service” family practice in British Columbia. Vancouver: Centre for Health Services and Policy Research; 2013. 177.  British Columbia Ministry of Health Planning. A New Era for Patient-Centred Health Care: Building a Sustainable, Accountable Structure for Delivery of High-Quality Patient Services. Victoria: British Columbia Ministry of Health Planning; 2001. 178.  Kralj B. Measuring Rurality - RIO2008 BASIC: Methodology and Results. Toronto: Ontario Medical Association; 2008. 179.  Leduc E. Defining rurality: a General Practice Rurality Index for Canada. Canadian Journal of Rural Medicine. 2012;2(2):125.  180.  CIHI. Canadian Hospital Reporting Project Technical Notes — Clinical Indicators. Toronto: CIHI; 2013. 181.  StataCorp. Stata Statistical Software: Release 13. College Station, TX: StataCorp LP.; 2013. 182.  Du Plessis V, Beshiri R, Bollman RD. Definitions of rural. Rural and Small Town Canada Analysis Bulletin. 2001;3(3).  183.  Canadian Institute for Health Information. Developing a Model for Measuring the Efficiency of the Health System in Canada — Data Availability. Ottawa: CIHI; 2012. 184.  Nolte E, McKee M. Measuring the health of nations: analysis of mortality amenable to health care. British Medical Journal (Clinical Research Edition). 2003;327(7424):1129.  185.  Nolte E, McKee CM. Amenable mortality: The authors respond. Health Affairs. 2008;27(4):1197.  186.  Wilkins R. PCCF+ Version 5E User’s Guide. Automated Geographic Coding Based on the Statistics Canada Postal Code Conversion Files, Including Postal Codes through March 2009. Catalogue 82F0086-XDB. Ottawa: Health Analysis Division, Statistics Canada; 2009. 187.  Ng E. Article The healthy immigrant effect and mortality rates. Health Reports. 2011;22(4).  188.  Tjepkema M, Wilkins R, Senécal S. Mortality of Métis and Registered Indian adults in Canada: An 11-year follow-up study. Health Reports. 2009;20(4).  161  189.  Health Canada. A Statistical Profile on the Health of First Nations in Canada: Health Services Utilization in Western Canada, 2000. Ottawa: Health Canada; 2013. 190.  McGrail K, Bryan S, Davis J. Let’s All Go to the PROM: The Case for Routine Patient-Reported Outcome Measurement in Canadian Healthcare. Healthcare Papers. 2012;11(4):8–18.  191.  Wennberg JE, Fisher E, Goodman D, et al. Tracking the Care of Patients with Severe Chronic Illness-The Dartmouth Atlas of Health Care 2008. The Dartmouth Institute for Health Policy & Clinical Practice; 2008. 192.  Chandra A, Skinner JS. Technology Growth and Expenditure Growth in Health Care. Cambridge: National Bureau of Economic Research; 2011. 193.  Sutherland JM, Crump RT, Repin N, et al. Paying for Hospital Services : A Hard Look.  194.  Rosenthal T. Geographic variation in health care. Annual Review of Medicine. 2012;63:493–509.  195.  Song Y, Skinner J, Bynum JPW, et al. Regional variations in diagnostic practices. The New England Journal of Medicine. 2010;363(1):45–53.  196.  BC Health Services Purchasing Organization. Annual Report for 2011 / 12. Victoria: BC Ministry of Health; 2013. 197.  Baker LC, Bundorf MK, Kessler DP. Patients’ preferences explain a small but significant share of regional variation in medicare spending. Health Affairs. 2014;33(6):957–963.  198.  Roos LL, Walld R. Neighbourhood, family and health care. Canadian Journal of Public Health. 2007;98 Suppl 1:S54–61.  199.  Canadian Institute for Health Information. Health Care Cost Drivers: The Facts. 2011. 200.  Djulbegovic B, Guyatt G. Evidence-Based Practice Is Not Synonymous With Delivery of Uniform Health Care. JAMA. 2014;312(13):1293–1294.  201.  Langley GR, Tritchler DL, Llewellyn-Thomas HA, et al. Use of written cases to study factors associated with regional variations in referral rates. J Clin Epidemiol. 1991;44(4-5):391–402.  202.  Wennberg JE. Forty years of unwarranted variation--and still counting. Health Policy. 2014;114(1):1–2.  203.  Wennberg JE. Time to tackle unwarranted variations in practice. British Medical Journal. 2011;342:d1513.  204.  Emanuel E, Tanden N, Altman S, et al. A Systemic Approach to Containing Health Care Spending. New England Journal of Medicine. 2012;367(10):949–954.  162  205.  Nelson CH, Park J. The nature and correlates of unmet health care needs in Ontario, Canada. Social Science & Medicine. 2006;62(9):2291–2300.  206.  Bryan S, Broesch J, Dalzell K, et al. What are the most effective ways to measure patient health outcomes of primary health care integration through PROM (Patient Reported Outcome Measurement) instruments? Vancouver: Centre for Clinical Epidemiology and Evaluation; 2013. 207.  CIHI. Canadian Patient Cost Database Technical Document: MIS Patient Costing Methodology, November 2011. Ottawa: CIHI; 2011. 208.  Anderson GM, Pulcins IR. High variation medical conditions as an explanation of regional and temporal differences in hospital utilization. Medical Care. 1992;30(5):461–465.  209.  Balogh RS, Ouellette-Kuntz H, Hunter DJW. Regional variation in dental procedures among people with an intellectual disability, Ontario, 1995-2001. Journal of the Canadian Dental Association. 2004;70(10):681–681f.  210.  Billings J, Anderson GM, Newman LS. Recent findings on preventable hospitalizations. Health Aff. 1996;15(3):239–249.  211.  Feehan LM, Sheps SS. Treating hand fractures: Population-based study of acute health care use in British Columbia. Canadian Family Physician. 2008;54(7):1001–1007.  212.  Gillis AM. Referring patients for consideration of device treatment of sudden cardiac death and heart failure: incorporating evidence-based therapies within the Canadian health care system. The Canadian Journal of Cardiology. 2005;21(Suppl A):25A–30A.  213.  Hilsden RJ, Verhoef MJ, Best A, et al. A national survey on the patterns of treatment of inflammatory bowel disease in Canada. BMC Gastroenterology. 2003;3(10).  214.  Hislop TG, Inrig KM, Bajdik CD, et al. Health care services and Pap testing behavior for Chinese women in British Columbia. Journal of Immigrant Health. 2003;5(4):143–152.  215.  Joly MF, Rannou A. [A geographic analysis of regional disparities in temporal accessibility of emergency services to traffic accident victims in Quebec]. Can J Public Heal. 1994;85(1):41–46.  216.  Pampalon R, Lebel A, Hamel D. Exploring rural health inequalities at a local scale: the case of Portneuf, Québec. Canadian journal of public health. Revue canadienne de santé publique. 2007;98(2):143–148.  217.  Parikh S V, Wasylenki D, Goering P, et al. Mood disorders: rural/urban differences in prevalence, health care utilization, and disability in Ontario. Journal of Affective Disorders. 1996;38:57–65.  218.  Platt GH, Svenson LW, Woodhead SE. Coronary artery bypass grafting in Alberta from 1984 to 1989. The Canadian Journal of Cardiology. 1993;9(7):621–624.  163  219.  Ryan BL, Stewart M, Campbell MK, et al. Understanding adolescent and young adult use of family physician services: a cross-sectional analysis of the Canadian Community Health Survey. BMC Family Practice. 2011;12(118).     164  Appendices A Search strategy to identify Canadian geographic variations research 1. (Canad* or "British Columbia*" or Alberta* or Saskatchewan or Manitoba* or Ontari* or Quebec* or "New Brunswick" or "Nova Scotia*" or Prince Edward Island or Yukon or Northwest Territories or Nunavut).mp. [mp=title, abstract, original title, name of substance word, subject heading word, protocol supplementary concept, rare disease supplementary concept, unique identifier] 2. ((rural* adj6 variation*) or (rural* adj6 differen*) or (rural* adj6 disparit*)).mp. [mp=title, abstract, original title, name of substance word, subject heading word, protocol supplementary concept, rare disease supplementary concept, unique identifier] 3. ((geograph* adj3 variation*) or (geograph* adj3 differen*) or (geograph* adj3 disparit*)).mp. [mp=title, abstract, original title, name of substance word, subject heading word, protocol supplementary concept, rare disease supplementary concept, unique identifier] 4. ((region* adj3 variation*) or (region* adj3 differen*) or (region* adj3 disparit*)).mp. [mp=title, abstract, original title, name of substance word, subject heading word, protocol supplementary concept, rare disease supplementary concept, unique identifier] 5. ((area* adj3 variation*) or (area* adj3 differen*) or (area* adj3 disparit*)).mp. [mp=title, abstract, original title, name of substance word, subject heading word, protocol supplementary concept, rare disease supplementary concept, unique identifier] 6. ((neighbo*rhood* adj3 variation*) or (neighbo*rhood* adj3 differen*) or (neighbo*rhood* adj3 disparit*)).mp. [mp=title, abstract, original title, name of substance word, subject heading word, protocol supplementary concept, rare disease supplementary concept, unique identifier] 7. 2 or 3 or 4 or 5 or 6 8. (health care or "health care" or "health service*").mp. [mp=title, abstract, original title, name of substance word, subject heading word, protocol supplementary concept, rare disease supplementary concept, unique identifier] 9. 1 and 7 and 8165  B List of references consulted in scoping review Table B-1 Citation information and description of references consulted in scoping review Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Alter, Austin, & Tu, 2005 (124) Admin Acute MI patients (condition-defined) CAN Province or sub-national regions X   Age-sex In-hospital mortality, cardiac readmissions, 30-day revascularization   X   X   X X  Postal code Mortality, hospital readmission, revascularization Alter, Stukel, & Newman, 2008 (53) Admin All people (population-based) ON County/regional municipality   X Age-sex Cardiac interventions, noninvasive testing, hospitalization, statin use   X  X   X    n/a n/a Anderson & Pulcins, 1992 (208) Admin All people (population-based) ON County/regional municipality X   Crude  Hospital discharges for high- and low-variation medical conditions   X         n/a n/a Badley, Canizares, Mahomed, Veinot, & Davis, 2011 (126) Survey & Admin All people (population-based) ON Health X   Crude  Physician rates of office visits, rates of surgery, supply X    X X      n/a n/a Balogh, Ouellette-Kuntz, & Hunter, 2004 (209) Admin People with intellectual disabilities (condition-defined) ON Health X  X Age-sex In-hospital dental procedures   X         n/a n/a 166  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Barbera, Elit, Krzyzanowska, Saskin, & Bierman, 2010 (102) Admin Women with gynecologic cancers (condition-defined) ON Health X   Age-sex Chemotherapy in last 2 weeks of life, death in an acute care bed, emergency department visits in last 2 weeks of life, home care nursing visits in last 6 months of life, physician house calls in last 2 weeks of life X  X      X   n/a n/a Bederman et al., 2011 (54) Survey & Admin All people (population-based) ON County/regional municipality X  X Multivariate  Degenerative lumbar spinal surgery   X  X X X X X X X n/a n/a Billings, Anderson, & Newman, 1996 (210) Admin Urban residents (population-based) ON Census X   Crude  Ambulatory-care sensitive hospitalizations X        X   n/a n/a Blais, 1993 (61) Admin All people (population-based) QC Health X  X Age-sex Surgical rates   X   X      n/a n/a 167  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Blais, Breton, Fournier, St-Georges, & Berthiaume, 2003 (127) Admin All people (population-based) QC Sub-provincial regions researcher-defined X   Crude  Rates of physician visits for mental health problems X    X   X    Dichotomy (defined cities) n/a Booth, Hux, Fang, & Chan, 2005 (100) Admin Diabetes patients (condition-defined) ON Urban/ rural other X   Multivariate  Diabetes hospitalizations X   X     X X  Dichotomy (population size) n/a Boyle, Badley, & Glazier, 2006 (79) Survey & Admin Patients with arthritis and rheumatism (condition-defined) ON Health X  X Multivariate  Specialist encounter rates X    X X   X X  n/a n/a Brownell, 2002 (57) Admin Children (population-based) MB Health X   Age-sex Tonsillectomy rates   X X        Dichotomy (defined cities) n/a Brownell et al., 2002 (76) Admin Children ages 0-19 (population-based) MB Health X   Age-sex Hospitalizations, physician visits, use of nurses X        X   Dichotomy (defined cities) Premature mortality 168  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Chan, Anderson, & Dales, 1997 (62) Admin All people (population-based) ON Health X  X Age-sex Spirometry use   X         n/a n/a Cohen, Vanasse, & Courteau, 2012 (88) Admin Patients with cardiovascular events (condition-defined) QC Urban/ rural census X   Multivariate  Revascularizations, specialist use X X   X   X    CMA/CA vs. elsewhere CVD deaths, all-cause mortality, hospital readmission Coyte & Young, 1999 (83) Admin Patients with inpatient or same-day surgery separations (condition-defined) ON Health X  X Age-sex Home care use X           n/a n/a Coyte et al., 2001 (51) Admin Children ages 0-14 (population-based) ON County/regional municipality X  X Multivariate  Insertion of tympanostomy tubes   X  X  X  X  X Census n/a Cree, Yang, Scharfenberger, Johnson, & Carriere, 2002 (60) Admin Patients age 65+ with femoral neck fractures (condition-defined) AB Hospital/clinic catchment X   Multivariate  Arthroplasty rates, hospital stay   X         n/a n/a 169  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Dagenais, Vanasse, Courteau, Orzanco, & Asghari, 2010 (75) Admin Patients age 65+ with fragility fractures (condition-defined) QC Urban/ rural census X   Age-sex Bone mineral density testing, bisphosphonates prescribing, hormone replacement therapy prescribing   X X        MIZ n/a Diaz-Granados, Georgiades, & Boyle, 2010 (129) Survey All people (population-based) CAN Health  X  Multivariate  Use of health services for mental health reasons X    X   X X   Census n/a Dodek, Keenan, Norena, Martin, & Wong, 2010 (106) Admin Patients of intensive care units (population-based) BC Health X   Crude  ICU capacity, ICU use X X          Dichotomy (details not stated) Hospital mortality Feasby, Quan, & Ghali, 2001 (41) Admin All people (population-based) CAN Census X  X Age-sex Carotid endarterectomy   X         n/a n/a Feehan & Sheps, 2008 (211) Admin Patients with hand fractures (condition-defined) BC Health X   Crude  Admission rates for hand fractures X           n/a n/a 170  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Finlayson, Lix, Finlayson, & Fong, 2005 (89) Admin People age 75+ (population-based) MB Urban/ rural cities (vs. elsewhere) X   Crude  Hospital separations, cataract surgeries, hip/knee replacements, physician services X X X         Dichotomy (defined cities) n/a Frohlich, Fransoo, & Roos, 2002 (123) Admin Winnipeg residents (population-based) MB Census X   Age-sex Visits to GPs, specialist visits, hospitalization days, diagnostic imaging, high profile procedures, screening and preventative services X       X    n/a Premature mortality Gamble et al., 2011 (98) Admin Patients with incident heart failure (condition-defined) AB Urban/ rural postal X   Multivariate  Mortality, rehospitalizations, office-based physician visits, emergency department visits, prescribing X  X     X    Postal code Mortality, hospital readmission Gaudette et al., 2004 (69) Admin Patients with invasive breast cancer (condition-defined) CAN Province or sub-national regions X   Age-sex Rates of mastectomy, breast-conservative surgery   X X        n/a n/a 171  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Gillis, 2005 (212) Admin Patients at high risk of cardiac death (condition-defined) CAN Sub-provincial regions researcher-defined X   Crude  Rates of implantable cardioverter difibrillators   X         n/a n/a Goel, Iron, & Williams, 1997 (63) Admin Women ages 30+ (population-based) ON Health   X Crude  Mammography screening or probable diagnostic   X         n/a n/a Gomez et al., 2010 (143) Admin Trauma decedents (condition-defined) ON Urban/ rural census X   Crude  Emergency department accessibility X     X      Census Prehospital mortality Goodridge, Lawson, Rennie, & Marciniuk, 2010 (111) Admin COPD or lung cancer patients at end of life (condition-defined) SK Urban/ rural population X   Crude  Physician visits, length of stay, hospitalizations, institutional care, home care, transition between care settings, location of death X X          Census category for CMA/CA n/a Haggerty, Roberge, Pineault, Larouche, & Touati, 2007 (97) Survey Primary care patients (population-based) QC Hospital/clinic catchment X   Multivariate  Emergency department use X     X      Dichotomy (defined cities) n/a 172  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Hanley, Janssen, & Greyson, 2010 (59) Admin Deliveries with no previous cesarean (condition-defined) BC Health X   Multivariate  Caeserian use   X        X n/a n/a Hardy, Kelly, & Voaklander, 2011 (93) Survey All people (population-based) CAN Urban/ rural census X   Multivariate  Service use or use of specialized mental health services X       X X X  Census n/a Hartford, Ross, & Walld, 1998 (137) Admin All people (population-based) MB Sub-provincial regions researcher-defined X   Age-sex Angiography and CABG rates   X     X X   Health system distance n/a Harvard, Hill, & Buxton, 2008 (120) Admin All people (population-based) BC Health X   Crude  Distribution of harm reduction products X       X    n/a n/a Hawker et al., 2001 (36) Survey Persons ages 55+ in two study areas high and low surgical rates (population-based) ON County/regional municipality X   Age-sex Hip and knee arthroplasty   X     X   X n/a n/a 173  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Heaman, Green, Newburn-Cook, Elliott, & Helewa, 2007 (105) Admin Women who gave birth (condition-defined) MB Health X   Multivariate  Adequate prenatal care X       X X X  n/a n/a Hilsden, 2004 (64) Admin All people (population-based) AB Health X   Age-sex Gastrointetinal endoscopy flexible sigmoidoscopy, colonoscopy, and gastroscopy   X         Dichotomy (defined cities) n/a Hilsden, Romagnuolo, & May, 2004 (65) Admin All people (population-based) AB Health X   Age-sex Endoscopic retrograde cholangiopancreatography, sphincterotomy, stone extraction, stent placement    X         Dichotomy (defined cities) n/a Hilsden, Verhoef, Best, & Pocobelli, 2003 (213) Survey Patients with Crohn's and Colitis (condition-defined) CAN Province or sub-national regions X   Multivariate  Surgical rates, drug use, primary physician X       X X   n/a n/a Hislop et al., 2003 (214) Survey Chinese women (population-based) BC Census X   Crude  Pap testing   X  X X   X X X n/a n/a 174  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Hogan, Maxwell, Fung, & Ebly, 2003 (71) Survey People ages 65+ (population-based) CAN Province or sub-national regions X   Crude  Number of medications consumed and prevalence of use of various drugs X  X     X    n/a n/a Jimenez-Rubio, Smith, & Van Doorslaer, 2008 (43) Survey All people (population-based) CAN Province or sub-national regions X   Multivariate  General practitioner visits, hospital nights, specialist visits X       X X   n/a n/a Johansen et al., 2009 (56) Admin Heart attack patients (condition-defined) CAN Health X   Age-sex Revascularization rates post acute MI   X         n/a Mortality Joly & Rannou, 1994 (215) Admin Traffic accident victims (condition-defined) QC County/regional municipality X   Crude  Time between accident and arrival in hospital X     X      n/a n/a Karunanayake & Pahwa, 2009 (92) Survey Canadian seniors (population-based) CAN Urban/ rural census X   Multivariate  Unmet health needs, consultations with family doctor X       X X   Census n/a Keenan et al., 2003 (145) Admin COPD patients (condition-defined) BC Hospital/clinic catchment X   Multivariate  Length of ICU stay  X      X    n/a Hospital mortality 175  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Kelly & Jones, 1995 (130) Admin Psychiatric patients (condition-defined) BC Census X   Multivariate  Inpatient psychiatric service use X    X X   X   n/a n/a Klein-Geltink, Pogany, Barr, Greenberg, & Mery, 2005 (135) Admin Canadian children with cancer (condition-defined) CAN Province or sub-national regions X   Multivariate  Days between symptom onset and anti-cancer treatment  X   X X  X    Health system distance n/a Kwon, Carey, Cook, Qiu, & Paszat, 2008 (142) Admin Women with endometrial cancer (condition-defined) ON Health X   Multivariate  Surgical staging, adjuvant pelvic radiotherapy, hospital type, physician type X  X   X  X X   n/a Mortality Lamarche et al., 2010 (144) Survey Primary care patients (population-based) QC Urban/ rural other X   Multivariate  Family physician use, medical specialist use, hospital emergency department uses, assessed accessibility, continuity and responsiveness of primary care X   X        Census n/a Langley, Minkin, & Till, 1997 (139) Survey Patients of NS family physicians (population-based) NS Hospital/clinic catchment X   Crude  Decisions about referral X    X X X X   X n/a n/a 176  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Lee et al., 2004 (110) Admin Patients hospitalized for heart failure (condition-defined) CAN Health X   Age-sex Readmission rates 30-day, 90-day, one-year X           n/a Mortality Lin, Chan, & Goering, 1998 (121) Survey & Admin All people (population-based) ON Health X   Crude  Fee-for-service expenditures for mental health care X       X    n/a n/a Locker & Clarke, 1999 (80) Survey People ages 50+ (population-based) ON County/regional municipality X   Crude  Regular dental visits, use of specialist services X       X X X  n/a Patient-reported improvement Lopushinsky, Austin, Rabeneck, Kulkarni, & Urbach, 2007 (52) Admin Patients who underwent a primary antireflux procedure (condition-defined) ON County/regional municipality X  X Age-sex Antireflux surgery   X         n/a n/a 177  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Lum et al., 2007 (99) Admin Asthma patients (condition-defined) AB Urban/ rural large cities vs. other X   Crude  Pulmonary tests, referrals, asthma education, medication use X  X         Dichotomy (defined cities) n/a Martens, Brownell, & Kozyrskyj, 2002 (95) Survey & Admin Children and adolescents (population-based) MB Urban/ rural large cities vs. other X   Crude  Hospital utilization, pharmaceutical use X           Dichotomy (defined cities) n/a Mcdonald & Conde, 2010 (149) Survey People ages 55+ (population-based) CAN Urban/ rural census X   Crude  Has a general practitioner, visited a general practitioner, visited a specialist, visited a dentist X       X X   Census category for CMA/CA n/a Menec, Nowicki, & Kalischuk, 2010 (104) Admin Decedents ages 19+ (condition-defined) MB Health X   Crude  Transfers to hospital at end of life X    X X  X    Dichotomy (defined cities) n/a 178  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Menzies, Lewis, & Oxlade, 2008 (73) Survey & Admin Tuberculosis patients (condition-defined) CAN Province or sub-national regions X   Crude  TB costs per active case X           n/a n/a Mitchell, Strain, & Blandford, 2007 (113) Survey Adults ages 65+ (population-based) MB Urban/ rural Beale X   Multivariate  Home care use X       X    Dichotomy Beale n/a Mohan, Chen, Campbell, & Hemmelgarn, 2010 (109) Survey Patients with hypertension (condition-defined) CAN Province or sub-national regions X   Multivariate  Treatment for hypertension X       X X X  n/a n/a Morgan, Cunningham, & Hanley, 2010 (67) Admin All people (population-based) BC Health   X Multivariate  Antihypertensives, statins, acid reducing drugs, opioids   X  X   X X X  n/a n/a Mustard & Frohlich, 1995 (116) Admin All people (population-based) MB Health X   Age-sex Hospital separations X        X   n/a Health Status Index, mortality Mustard, Kozyrskyj, Barer, & Sheps, 1998 (82) Admin Urban residents (population-based) MB Sub-provincial regions researcher-X   Multivariate  Proportion of ambulatory care provided in emergency departments X       X X X  Health system distance n/a 179  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) defined Myers, Godin, Lambert, Calzavara, & Locker, 1996 (72) Survey Gay men (population-based) CAN Province or sub-national regions X   Multivariate  HIV testing   X         Census category for CMA/CA n/a Ng, Chatwood, & Young, 2011 (85) Survey Aboriginal people with and without arthritis (population-based) CAN Province or sub-national regions X   Crude  Consulting a MD, nurse, or traditional healer X           n/a n/a Ngamini Ngui, Perreault, Fleury, & Caron, 2012 (115) Survey People with a mental health disorder (condition-defined) QC Census  X  Multivariate  Mental health services X    X X  X X   n/a n/a Offord et al., 1987 (118) Survey Children with and without psychiatric disorders (population-based) ON Health X   Crude  Mental health services, social services, ambulatory medical care X           Dichotomy (defined cities) n/a 180  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Olatunde, Leduc, & Berkowitz, 2007 (101) Admin Patients of general practitioners (population-based) BC Health X   Crude  Fee-for-service billings for emergency visits, inpatient care, orthopedics, trauma, surgery, anaesthesia, obstetrics, mental health, and chemotherapy,  X     X      Health system n/a Olivotto, Kan, & King, 2000 (107) Admin Women with abnormal screening mamograms (condition-defined) BC Health X   Crude  Time from abnormal screen to diagnosis  X          n/a n/a Pampalon, Lebel, & Hamel, 2007 (216) Admin Rural residents (population-based) QC Sub-provincial regions researcher-defined X   Age-sex Hospitalization X           n/a n/a Parikh, Wasylenki, Goering, & Wong, 1996 (217) Survey All people (population-based) ON Urban/ rural CMA X   Crude  Use of informal services, non-psychiatrist md, other health professionals, psychiatrist, outpatient, ED, inpatient X           Census category for CMA/CA n/a 181  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Paszat et al., 1998 (103) Admin Women with invasive breast cancer (condition-defined) ON Hospital/clinic catchment X   Crude  Radiotherapy use   X   X   X   Health system distance n/a Pefoyo, Laurier, & Rivard, 2010 (66) Admin All people (population-based) QC Health  X  Multivariate  Prevalence of use of asthma medication   X  X X   X   n/a n/a Peterson, Shapiro, & Roos, 2005 (84) Admin All people (population-based) MB Health X   Crude  Home care use X           n/a n/a Pilote et al., 2004 (70) Admin Acute MI patients (condition-defined) CAN Health X   Age-sex Rates and waiting times for PCI, CABG, overall revascularization  X X         n/a n/a Platt, Svenson, & Woodhead, 1993 (218) Admin All people (population-based) AB Census X   Crude  CABG rates   X         n/a n/a Pong et al., 2011 (94) Survey & Admin All people (population-based) CAN Urban/ rural census X   Age-sex Wide range from CCHS and admin X           MIZ Multiple CCHS-derived 182  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Ronen & Meaney, 2003 (131) Survey All people (population-based) CAN Hospital/clinic catchment X   Crude  Clinic hours, waiting time X X   X       n/a n/a Roos & Walld, 2007 (198) Admin Same-sex siblings (population-based) MB Urban/ rural census X   Crude  Hospital stays, physician visits, costs X           Census category for CMA/CA n/a Roos, 1984 (138) Admin Female population (population-based) MB Hospital/clinic catchment X   Age-sex Hysterectomy rates   X  X X X X    n/a n/a Roos & Roos, 1981 (122) Admin All people (population-based) MB Hospital/clinic catchment X   Age-sex Surgical rates X  X  X   X X X  n/a n/a Rosenberg & Hanlon, 1996 (134) Survey & Admin All people (population-based) ON Health X   Multivariate  General practitioner visits, specialist visits, emergency services, hospital admissions X    X X   X   Health system n/a 183  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Rowe et al., 2010 (112) Admin COPD patients presenting to the ED (condition-defined) AB Urban/ rural large cities vs. other X   Crude  Time to non-emergency follow-up visits X X       X X  Dichotomy (defined cities) n/a Ryan, Stewart, Campbell, Koval, & Thind, 2011 (219) Survey Adolescents and young adults (population-based) CAN Province or sub-national regions X   Multivariate  Utilization and intensity of family physician services X       X X X  Census n/a Shapiro & Roos, 1984 (96) Admin Elderly people (population-based) MB Urban/ rural cities vs. elsewhere X   Crude  Physician visits, hospital usage X       X    Dichotomy (defined cities) n/a 184  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Sirois, Dionne, & Lavoie, 2009 (136) Survey Patients admitted to level I or level II trauma centres who required rehab services (condition-defined) QC Health X   Multivariate  Barriers to services X     X  X X   n/a n/a Stuart, 2000 (117) Admin All people AB Health X   Age-sex Treated prevalence of mental health services  X    X    X X  n/a n/a Tamblyn et al., 1994 (68) Admin Elderly medicare registrants who made a visit to a physician (population-based) QC Health X   Age-sex High-risk prescribing of psychotropic drugs   X         n/a n/a Tataryn, Roos, & Black, 1995 (125) Admin All people (population-based) MB Urban/ rural large cities vs. X   Age-sex Contact with physician, overall and by specialty X    X       Dichotomy (defined cities) n/a 185  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) other Taylor et al., 1998 (81) Admin All people (population-based) ON County/regional municipality X  X Age-sex Surgical and medical back and neck hospitalizations, length of stay X X X         n/a n/a Tepper et al., 2006 (90) Admin Residents of Calgary, Edmonton, London, and rural areas (population-based) AB, ON Urban/ rural other X   Age-sex Appendectomy, carpal tunnel release, closed hip fracture repair, rectal cancer surgery, joint replacement, thyroidectomy, unilateral of bilateral inguinal herniorrhaphy, cholecystectomy   X   X      Health system distance n/a 186  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) To, Feldman, Young, & Maloney, 1996 (133) Admin Children (population-based) ON Health X  X Age-sex Hospitalization for gastroenteritis X     X      n/a n/a To & Langer, 2010 (91) Admin Children ages 0-19 (population-based) ON Health X  X Age-sex Paediatric appendectomy   X  X X   X X  CMA/CA vs. Elsewhere n/a Tonelli et al., 2006 (42) Admin Dialysis patients (condition-defined) CAN Province or sub-national regions X   Multivariate  Kidney transplantation from a deceased donor   X   X  X X   n/a n/a Van Walraven et al., 1996 (119) Admin People ages 60+ with primary hip and knee replacements (condition-defined) ON County/regional municipality X   Age-sex Proportion of joint replacements deemed inappropriate   X X        n/a n/a 187  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Vanasse, Courteau, Cohen, Orzanco, & Drouin, 2010 (74) Admin Atherosclerotic patients newly hospitalized for a MI, osteoporotic patients who have suffered a fragility fracture, incident cases of diabetes (condition-defined) QC Urban/ rural MIZ X   Age-sex Material resource use, physician consultation, drug treatment X  X         MIZ Mortality, hospital readmission Vanasse, Dagenais, et al., 2005 (132) Admin All people ages 65+ for whom a physician claimed a consultation for a low velocity vertebral, hip, wrist, or humerus fracture (condition-defined) QC Health X   Multivariate  Osteoporosis treatment and bone mineral density testing   X   X  X X   Health system distance n/a 188  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Vanasse, Niyonsenga, et al., 2005 (58) Admin Patients hospitalized for acute coronary syndrome (condition-defined) QC Health X   Multivariate  Rates of invasive cardiac procedure, hospital death, length of stay   X   X      Health system distance n/a   Survey All people (population-based) NS Census X   Multivariate  Family physician use X    X       Dichotomy (defined cities) n/a Veugelers, Yip, & Elliott, 2003 (77) Admin All people (population-based) NS Census X   Multivariate  Family physician use, specialist use, hospital days X       X X   Dichotomy (defined cities) n/a Wen, Liu, Marcoux, & Fowler, 1998 (87) Admin Women who gave birth (population-based) CAN Province or sub-national regions X   Multivariate  Length of hospital stay for all births  X      X    n/a n/a Wijeysundera, Stukel, Chong, Natarajan, & Alter, 2010 (108) Admin Patients referred for coronary angiography (condition-defined) ON Health X   Multivariate  Wait times for coronary angiography, % above recommended maximum waiting time  X X  X X  X    n/a n/a 189  Citation Data source & study population Description of variation Primary variable examined Health system causes Population causes Rurality Outcome Year Reference Data source Population Study location Geographic unit Strata Multilevel model CV, SCV Adjustment Variables examined Volume/mix Wait times/LOS Procedures/drugs Quality/appropriateness Health human resources Material resources Professional culture Health status SES Race/ ethnicity Beliefs /values Rural definition Outcome examined  (if applicable) Williams et al., 2006 (86) Survey Patients with schizophrenia (condition-defined) CAN Province or sub-national regions X   Multivariate  Hospitalization, type of antipsychotic X  X     X X   n/a n/a Wilson et al., 2005 (114) Admin All people (population-based) AB Urban/ rural large cities vs. other X   Crude  Home care use X           Dichotomy n/a Woltman & Newbold, 2007 (128) Survey Immigrant and native-born female CMA residents (population-based) CAN County/regional municipality  X  Multivariate  Pap tests   X  X   X X X  n/a n/a Wright et al., 1999 (24) Survey & Admin All people (population-based) ON County/regional municipality X  X Multivariate  Knee replacement surgery   X  X X X    X n/a n/a    190  C Network characteristics Table C-1 Linked hospitals, patients, and physicians by network Name of hub hospital Size of hub hospital (# patients assigned) # hospitals  # patients # GPs # medical specialists # surgical specialists # lab specialists # imaging specialists Vancouver ≥150,001 7 823,312 1133 1031 717 114 30 Surrey Memorial ≥150,001 2 353,802 373 141 109 28 23 Lions Gate ≥150,001 3 226,889 352 114 112 21 8 Royal Jubilee ≥150,001 4 331,469 619 283 190 33 9 Richmond ≥150,001 1 167,276 166 73 62 3 1 Kelowna ≥150,001 1 167,193 226 102 105 20 11 Royal Columbian ≥150,001 2 267,536 312 189 117 27 12 Nanaimo 50,001-150,000 3 172,559 267 87 71 15 7 Langley Memorial 50,001-150,000 1 138,729 146 27 45 7 2 Abbotsford  50,001-150,000 3 158,544 184 57 52 27 4 Royal Inland 50,001-150,000 6 162,281 359 115 122 20 9 Prince George 50,001-150,000 8 148,422 252 59 116 10 6 Peace Arch District 50,001-150,000 1 84,719 115 34 35 6 1 Vernon Jubilee 50,001-150,000 3 118,817 183 43 55 9 8 Chilliwack 50,001-150,000 2 85,368 112 23 30 0 1 Ridge Meadows 50,001-150,000 1 73,492 88 31 24 6 1 Cowichan District 50,001-150,000 1 63,396 97 23 25 5 0 Penticton 50,001-150,000 3 77,342 122 51 39 9 2 St Josephs 50,001-150,000 1 56,631 104 24 40 6 3 Campbell River 15,001-50,000 4 51,623 93 20 20 6 1 East Kootenay C 15,001-50,000 5 68,329 145 26 28 9 3 Kootenay Boundary 15,001-50,000 4 68,837 150 25 32 10 2 Fort St. John 15,001-50,000 4 59,443 70 8 10 10 0   191   Table C-2 Loyalty measures, observed, and expected costs by network Name of hub hospital % of primary care visits within network  % of specialty physician visits within network % of acute medical hospitalizations within network Observed costs Predicted costs (based on age, sex and ADG) Difference between observed and expected costs Vancouver 91.1 77.7 79.3 1,020 1,084 -63.7 Surrey Memorial 89.6 55.0 67.3 1,074 1,081 -7.3 Lions Gate 90.7 54.4 80.5 1,167 1,222 -55.9 Royal Jubilee 95.8 68.6 92.7 1,253 1,251 2.5 Richmond 87.8 36.3 69.9 889 1,000 -111.6 Kelowna 95.8 88.5 90.0 1,247 1,203 44.7 Royal Columbian 87.2 47.4 66.9 1,094 1,078 16.1 Nanaimo 94.8 59.8 83.4 1,308 1,287 20.7 Langley Memorial 84.2 41.6 66.6 1,050 1,083 -32.3 Abbotsford  92.2 51.8 81.1 1,233 1,168 64.8 Royal Inland 93.2 75.0 84.9 1,220 1,148 71.7 Prince George 94.8 72.2 87.8 1,138 1,081 57.2 Peace Arch District 87.5 42.8 72.5 1,380 1,389 -8.4 Vernon Jubilee 94.1 79.3 87.0 1,346 1,282 63.9 Chilliwack 92.0 42.0 79.2 1,295 1,207 87.8 Ridge Meadows 91.2 42.7 70.0 1,190 1,131 59.1 Cowichan District 92.6 41.8 80.4 1,299 1,276 22.7 Penticton 93.8 76.3 89.6 1,477 1,436 41.0 St Josephs 92.3 74.9 80.7 1,278 1,353 -74.7 Campbell River 91.1 46.6 78.0 1,285 1,217 67.8 East Kootenay 96.5 84.7 84.9 1,182 1,089 93.4 Kootenay Boundary 94.0 84.9 86.1 1,270 1,196 74.3 Fort St. John 95.8 47.8 81.3 959 920 38.7    192  Table C-3 Breakdown of total costs by network Name of hub hospital % costs from GP services % costs from medical specialist services % costs from surgical specialist services % costs from imaging specialist services % costs from laboratory specialist services % costs from acute care services % costs from day surgery % hospital spending overall % physician services from specialists Vancouver 16.3 11.3 10.1 4.0 5.2 47.5 5.6 53.1 65.2 Surrey Memorial 17.9 9.0 10.3 3.8 4.8 47.7 6.6 54.3 60.9 Lions Gate 17.3 10.1 10.8 4.3 4.9 45.5 7.0 52.5 63.5 Royal Jubilee 17.2 9.2 9.7 3.8 4.9 49.4 5.6 55.0 61.6 Richmond 18.5 10.6 10.8 4.1 5.1 44.3 6.5 50.8 62.3 Kelowna 17.6 8.3 9.7 3.2 4.4 51.1 5.8 56.9 59.3 Royal Columbian 15.8 8.8 10.9 4.2 4.9 48.8 6.5 55.3 64.6 Nanaimo 17.1 6.7 10.9 4.2 4.7 50.1 6.3 56.4 60.8 Langley Memorial 16.6 7.7 11.1 4.1 4.7 48.1 7.7 55.8 62.4 Abbotsford  17.8 6.6 9.9 3.6 4.2 51.1 6.8 57.9 57.7 Royal Inland 16.0 6.3 10.2 3.5 4.5 51.9 7.6 59.5 60.5 Prince George 16.0 5.2 9.2 3.5 4.5 54.7 6.9 61.6 58.3 Peace Arch District 16.6 7.4 10.5 3.9 4.3 50.4 7.0 57.4 61.1 Vernon Jubilee 18.6 5.7 10.2 4.0 4.6 49.9 7.0 56.9 56.8 Chilliwack 17.3 5.4 10.6 3.4 3.8 52.1 7.4 59.5 57.3 Ridge Meadows 17.3 7.5 9.8 3.5 4.2 51.0 6.7 57.7 59.1 Cowichan District 18.1 6.6 9.9 3.6 4.7 51.1 6.1 57.2 57.8 Penticton 18.0 6.6 9.9 2.9 4.8 51.0 6.7 57.7 57.3 St Josephs 17.4 6.1 11.3 4.2 5.1 48.4 7.5 55.9 60.5 Campbell River 19.4 5.8 9.5 3.8 5.2 49.4 6.9 56.3 55.6 East Kootenay 17.2 2.8 9.3 4.0 4.7 55.5 6.5 62.0 54.7 Kootenay Boundary 17.8 5.7 9.5 4.0 4.8 50.9 7.4 58.3 57.4 Fort St. John 20.6 3.1 6.1 3.4 4.4 56.3 6.1 62.4 45.2    193  Table C-4 Linked hospitals, patients, and physicians by satellite network Name of satellite network Size of hub hospital (# patients assigned) # hospitals  # patients # GPs # medical specialists # surgical specialists # lab specialists # imaging specialists Mills Memorial 15,001-50,000 4 45,993 79 16 22 4 2 Powell River 15,001-50,000 1 18,091 46 8 4 1 0 R.W. Large ≤15,000 1 375 9 0 0 0 0 Northern Haida Gwaii ≤15,000 1 270 12 1 0 0 0 Queen Charlotte Islands ≤15,000 1 118 5 0 0 0 0 Bella Coola ≤15,000 1 117 7 0 0 0 0  Table C-5 Loyalty measures, observed, and expected costs by satellite network Name of satellite network % of primary care visits within network  % of specialty physician visits within network % of acute medical hospitalizations within network Observed costs Predicted costs (based on age, sex and ADG) Difference between observed and expected costs Mills Memorial 94.5 58.6 81.5 1,247 1,073 173.3 Powell River 94.5 36.9 83.5 1,473 1,420 52.9 R.W. Large 52.9 6.2 86.4 640 737 -97.0 Northern Haida Gwaii 35.6 3.6 76.7 664 645 19.2 Queen Charlotte Islands 62.2 12.3 77.8 854 765 88.5 Bella Coola 57.6 8.8 80.0 340 589 -249.5    194  Table C-6 Breakdown of total costs by satellite network Name of hub hospital % costs from GP services % costs from medical specialist services % costs from surgical specialist services % costs from imaging specialist services % costs from laboratory specialist services % costs from acute care services % costs from day surgery % hospital spending overall % physician services from specialists Vancouver 15.5 3.8 9.6 3.5 4.3 55.5 7.8 63.3 57.8 Surrey Memorial 20.0 5.4 9.0 3.3 4.8 49.8 7.8 57.6 52.9 Lions Gate 5.3 5.8 9.0 4.1 5.7 63.1 6.9 70.0 82.3 Royal Jubilee 15.1 4.4 13.5 3.8 4.4 50.7 8.1 58.8 63.3 Richmond 4.8 3.7 10.5 3.4 3.2 67.0 7.4 74.4 81.3 Kelowna 4.9 6.3 9.2 2.8 4.4 65.7 6.7 72.4 82.2     

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0167685/manifest

Comment

Related Items