UBC Faculty Research and Publications

Development of a population-based microsimulation model of osteoarthritis in Canada Kopec, J. A.; Sayre, E. C.; Flanagan, W. M.; Fines, P.; Cibere, Jolanda; Rahman, Md Mushfiqur; Bansback, N. J.; Anis, A. H.; Jordan, J. M.; Sobolev, Boris; Aghajanian, J.; Kang, W.; Greidanus, Nelson; Garbuz, D. S.; Hawker, G. A.; Badley, Elizabeth M. Oct 9, 2009

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

Item Metadata

Download

Media
52383-Kopec_A._ et_al_OAC_2010.pdf [ 294.84kB ]
Metadata
JSON: 52383-1.0132726.json
JSON-LD: 52383-1.0132726-ld.json
RDF/XML (Pretty): 52383-1.0132726-rdf.xml
RDF/JSON: 52383-1.0132726-rdf.json
Turtle: 52383-1.0132726-turtle.txt
N-Triples: 52383-1.0132726-rdf-ntriples.txt
Original Record: 52383-1.0132726-source.json
Full Text
52383-1.0132726-fulltext.txt
Citation
52383-1.0132726.ris

Full Text

miP. F. Sor# ada, CaxStatistics Canada, Ottawa, ON, CanadaARTICLE IN PRESSOsteoarthritis and Cartilage (2009) jj, jjjejjjª 2009 Osteoarthritis Research Society International. Published by Elsedoi:10.1016/j.joca.2009.10.010ICRSkCentre for Health Economics and Outcome Sciences, St. Paul’s Hospital, Vancouver, BC, Canada{University of North Carolina, Chapel Hill, NC, USA#University of Toronto, Toronto, ON, CanadaSummaryObjectives: The purpose of the study was to develop a population-based simulation model of osteoarthritis (OA) in Canada that can be used toquantify the future health and economic burden of OA under a range of scenarios for changes in the OA risk factors and treatments. In thisarticle we describe the overall structure of the model, sources of data, derivation of key input parameters for the epidemiological component ofthe model, and preliminary validation studies.Design: We used the Population Health Model (POHEM) platform to develop a stochastic continuous-time microsimulation model of physi-cian-diagnosed OA. Incidence rates were calibrated to agree with administrative data for the province of British Columbia, Canada. The effectof obesity on OA incidence and the impact of OA on health-related quality of life (HRQL) were modeled using Canadian national surveys.Results: Incidence rates of OA in the model increase approximately linearly with age in both sexes between the ages of 50 and 80 and plateauin the very old. In those aged 50þ, the rates are substantially higher in women. At baseline, the prevalence of OA is 11.5%, 13.6% in womenand 9.3% in men. The OA hazard ratios for obesity are 2.0 in women and 1.7 in men. The effect of OA diagnosis on HRQL, as measured bythe Health Utilities Index Mark 3 (HUI3), is to reduce it by 0.10 in women and 0.14 in men.Conclusions: We describe the development of the first population-based microsimulation model of OA. Strengths of this model include the useof large population databases to derive the key parameters and the application of modern microsimulation technology. Limitations of the modelreflect the limitations of administrative and survey data and gaps in the epidemiological and HRQL literature.ª 2009 Osteoarthritis Research Society International. Published by Elsevier Ltd. All rights reserved.Key words: Osteoarthritis, Epidemiology, Microsimulation, Modeling, Population, Risk factors, Quality of life, Policy evaluation.IntroductionOsteoarthritis (OA) is the most common form of arthritis anda leading cause of disability1,2. As the population ages, thenumber of persons suffering from OA is expected to increa-se3e6. A number of strategies have been advocated to re-duce the burden of OA7e11. However, quantification of thepotential impact of such strategies on future disease burdenis a complex undertaking. The effects of important risk fac-tors for OA, such as age and overweight/obesity, are non-a population setting is not straightforward as one has toconsider the existing treatment patterns, heterogeneity ofeffects, as well as changes in the demographic structureof the population14. Because of such complexities, standardeconomic and policy analyses of OA burden typically makesimplifying assumptions about the population under studyand the impact of interventions over time15,16.Simulation modeling of disease processes in a populationcan inform health policy and has consequently become in-creasingly accepted and adopted17. Recent developments inDevelopment of a population-basedosteoarthritis in CanadaJ. A. Kopecyz*, E. C. Sayreyz, W. M. Flanaganx,N. J. Bansbackk, A. H. Anisyzk, J. M. Jordan{, BN. V. Greidanusyz, D. S. Garbuzyz, G. A. HawkeyUniversity of British Columbia, Vancouver, BC, CanazArthritis Research Centre of Canada, Vancouver, BClinear and the distribution of both factors in the populationis changing2,12,13. Similarly, extrapolating the benefits andside-effects of interventions from randomized trials to*Address correspondence and reprint requests to: J.A. Kopec,School of Population and Public Health, University ofBritish Columbia, Research Scientist, Arthritis ResearchCentre of Canada, 895 West 10th Avenue, Vancouver, BCV5Z 1L7, Canada. Tel: 1-604-871-4588; Fax: 1-604-879-3791;E-mail: jkopec@arthritisresearch.caReceived 4 April 2009; revision accepted 15 October 2009.1Please cite this article in press as: Kopec JA et al., Development of a poOsteoarthritis Cartilage (2009), doi:10.1016/j.joca.2009.10.010crosimulation model ofinesx, J. Cibereyz, Md. M. Rahmanz,bolevy, J. Aghajanianz, W. Kangz,nd E. M. Badley#nadavier Ltd. All rights reserved.nternationalartilageepairocietymodeling technology include microsimulation, continuous-time models, and advances in model calibration18. An impor-tant distinction is between macrolevel (group-based) modelsand microsimulation. In macrolevel models, groups of individ-uals move through several stages based on transition proba-bilities and the summary measures of interest are calculated.In contrast,microsimulationmodels simulate individual life his-tories, with the timing of events determined stochastically19.The main advantages of microsimulation include a potentiallyinfinite number of events and stages in disease process thatpulation-based microsimulation model of osteoarthritis in Canada,can be modeled and flexibility in defining input populations,merging multiple data sources, taking into account heteroge-neity of effects, and implementing changes in the model18e21.Thanks toadramaticprogress incomputer technology it isnowpossible to perform the calculations needed to simulate mil-lions of individual life histories18. Complex population-levelprocesses can be described by a series of relatively simpleequations and probabilistic parameters at the level of the indi-vidual. The model integrates such individual-level data andtranslates them into projections of disease burden in the pop-ulation. By changing the input parameters, such as the distri-bution of disease risk factors and comparing the likelyoutcomes, it is possible to evaluate various interventions interms of their expected impact on disease incidence, preva-lence, mortality, quality-adjusted life years (QALYs), andcosts17e19. Although simulationmodels have been used in ar-thritis research for cost-effectiveness and decision analys-es22e27, none of these models has been based ongenerating individual life histories in a population or includeda primary prevention component.The purpose of the study was to develop a simulationmodel of OA in a Canadian population that can be usedto quantify the future health and economic burden of OA un-der a range of plausible scenarios for changes in the riskfactors and treatments. In this article we provide an over-view of the model development process, with a focus onOA incidence and its impact on quality of life. Modeling ofthe impact of OA treatment (both surgical and non-surgical)and economic aspects of OA will be presented in separatepublications. The following stages in model developmentare described: (1) conceptualization of the disease; (2) im-plementation of a computer simulation program; (3) deriva-tion of the model parameters; and (4) model calibration andvalidation. The key assumptions and limitations of themodel are discussed.MethodsCONCEPTUAL FRAMEWORKPOHEM-OA is a population-based simulation model. The disease is de-fined as physician-diagnosed OA of any site, as opposed to radiographicor clinical OA. Population-level data on radiographically or clinically definedOA of specific joints are not available in Canada. Furthermore, this is an in-cidence-driven model e the key parameter is OA incidence (hazard) rate.The incidence rate is modeled as a function of established risk factors forOA, allowing for non-linear dose-response relationships and effect modifica-tion. Trends in OA incidence are determined by changes in the distribution ofrisk factors in the population over time. The burden of OA can be expressedin terms of the number of cases and impact on health-related quality of life(HRQL). The key components of the model and relationships betweenthem are presented in Fig. 1. Main model parameters, methods of derivation,and data sources are listed in Table I. Baseline population data and technicaldetails are provided in the Appendix.SIMULATION PLATFORMSimulations are performed using the Population Health Model (POHEM)platform28. POHEM is a generic, continuous-time microsimulation environ-ment developed at Statistics Canada29. The unit of simulation is the individ-ual and the events can occur at arbitrarily small time intervals. This modelingapproach differs from group-based methods of modeling proportions andBMIOA gnosRQLRegionIncomeEducationLinear regression el (NP regreel (CC in thedepicta. Ban Cd in t OsteCLHalth SARTICLE IN PRESS2 J. A. Kopec et al.: A microsimulation model of osteoarthritisAge & SexDiaHIncidence rates(BCLHD)modTobitmodIn the diagram, bubbles show the variablesrelationships between the variables, boxes variables and (in parentheses) sources of daobtained from the 2001 cycle of the CanadiTreatment effects and costs are not includePOHEM = Population Health Model; OA =Quality of Life; BMI = Body Mass Index; BDatabase; NPHS = National Population HeHealth Survey Fig. 1. POHEM-OA: Schematic representation of the key relationships anPlease cite this article in press as: Kopec JA et al., Development of a poOsteoarthritis Cartilage (2009), doi:10.1016/j.joca.2009.10.010 model, arrow connectors indicate the t the statistical models of changes in the aseline distributions of the variables were ommunity Health Survey (CCHS). he diagram.  oarthritis; HRQL = Health-Related D = British Columbia Linked Health urvey; CCHS = Canadian Community isssionHS)HS)Survival regressionmodel (NPHS)d sources of data pertaining to OA incidence and impact on HRQL.pulation-based microsimulation model of osteoarthritis in Canada,ableta sourvedined alationD-bad onrtalityined fssionationent caa 12ined ne rateistribuined fdjust.5, 18fromObtained fOA status-sectined fed bytudinaARTICLE IN PRESScrossEffects of prior BMI, sex, education,income, and region on change in BMIObtadefinlongiHUI3¼Health Utilities Index Mark 3.Effects of OA, age, sex, and BMI on HUI3TKey POHEM-OA parameters, daParameterDistribution of baseline population by age, sex,province, education, income, BMI and HUI3ObseBaseline prevalence of OA by age and sex ObtapopuBCLHMortality rates by age and sex over time Baseof moBaseline incidence rates of OA by age and sex ObtaprofeInternIncidusingReference incidence rates of OA(for persons with normal BMI) by age and sexObtaThesthe dEffect of BMI on incident OA, by sex Obtaand a(<18dataOsteoarthritis and Cartilage Vol. jj, No. jjclassic Markov models which assume constant transition rates between pre-specified states30,31. POHEM generates individual life trajectories withina large population representative of Canada, one individual at a time, untildeath. The stochastic nature of POHEM means that replicated lives withidentical initial conditions have different life trajectories, because they havea different stream of random numbers used in their decision path. POHEMhas been validated and used to assess the impact of prevention strategies,treatments, and cost of care in breast, lung, and colorectal cancer32e38.PARAMETERS AND SOURCES OF DATAAge and genderPOHEM-OA uses the 2001 Canadian Community Health Survey (CCHS)sample as the baseline population39. The CCHS is a cross-sectional surveyconducted every 2 years, with a total national sample of 130,000. The CCHStarget population includes persons age 12þ, living in households in the 10 Ca-nadian provinces, with the exception of Indian reserves, military bases, andsome remote northern areas. POHEM-OA includes subjects 18 years of ageor older. Each subject from the CCHS sample is replicated at time of initializa-tion to reflect its survey sampleweight. Consequently, approximately 25millionindividual lives, reflecting the non-institutionalized adult Canadian population,are simulated in a single run. Since POHEM-OA simulates a dynamic popula-tion, subjects are removed from the population as a result of death and newsubjects are added to the population by aging into it (on their 18th birthday).To simulate these events, death and birth rates by age and sex were obtainedfrom the national vital statistics and census databases routinely used byStatis-tics Canada for demographic projections40.Body mass index (BMI)At the population level, the most important modifiable risk factor for OA isBMI2,12,41. The distribution of BMI in the baseline population was obtained di-rectly from the CCHS, based on the formula BMI¼weight/height2 (Table I).Individual BMI trajectories were simulated using a regression model basedon longitudinal data from the National Population Health Survey (NPHS) inCanada42. The NPHS started in 1994 with a random household populationsample of about 20,000, with a sampling frame similar to that of theCCHS. These subjects are followed every 2 years43. The model to simulateBMI used data from 1996 to 2004 and included age, sex, province of resi-dence, education, income, and prior BMI.Please cite this article in press as: Kopec JA et al., Development of a poOsteoarthritis Cartilage (2009), doi:10.1016/j.joca.2009.10.010Irces, and methods of derivationSource and method of derivationin CCHS (2001)s the final stable prevalence from a simulation of the Canadianover a 50-year time horizon, under constant age/sex-specific,sed incidence rates.mortality data and using Statistics Canada projectionsfor Canada.rom BCLHD. OA was defined as at least two visits to a healthal within 2 years or one hospitalization with theal Classification of Diseases, Ninth Revision (ICD-9) code 715.ses were identified after excluding prevalent cases,-year run-in period.umerically using an iterative algorithm (calibration).s are based on the incidence rates from BCLHD andtion of BMI levels in the CCHS population, within age/sex groups.rom a survival regression model, separately for men and women,ed for age. BMI was treated as a categorical variable.5e25, 25e30, 30þ). The model is based on longitudinalthe NPHS (2000e2002).rom a tobit regression model including age, sex, BMI, and(with interactions as required). The model is based onional data from the CCHS (2001).rom a series of linear regression models for groupsBMI and age categories. The models are based onl data from the NPHS (1996e2004).3OA incidenceIn POHEM-OA, OA incidence at the start of simulation is adjusted to re-flect the incidence of physician-diagnosed OA in Canada. This adjustment,referred to as calibration, uses age/sex-specific incidence rates for 2003/4(the last year for which data were available), derived from the British Colum-bia Linked Health Database (BCLHD) (Table I). The BCLHD is a well-estab-lished administrative database in the province of British Columbia, Canadathat has been used extensively for research purposes44. It contains dataon physician billings in the publicly funded health care system covering>95% of the population. The methodology for estimating incidence ratesof OA from administrative data has been described elsewhere45. The defini-tion of OA required either two physician visits within 2 years of each other orone hospitalization with the ICD-9 code 715 (Osteoathrosis and allied disor-ders) or the corresponding ICD-10 codes. To obtain incidence rates for Can-ada, age/sex-specific BC rates were applied to the age/sex distribution of theCanadian population.OA prevalenceAs not all individuals with diagnosed OA seek regular medical care fortheir condition46, age/sex-specific prevalence proportions of OA from an ad-ministrative database may underestimate the true prevalence, particularly ifthe run-in time is too short45. To ensure consistency between incidenceand prevalence data, we simulated OA prevalence in the Canadian popula-tion over a long period of time (up to 50 years) while keeping the incidencerates constant at the most recent level. The age/sex-specific prevalence pro-portions obtained from this simulation were applied to the baseline CCHSpopulation to obtain the distribution of OA at baseline47.Effect of BMIWhen simulating the occurrence of OA in individuals, the age/sex-spe-cific hazard rates are modified according to the person’s BMI. The effectof BMI on OA risk has been estimated using longitudinal data from thetwo cycles of the NPHS (2000 and 2002) that asked the question on anyself-reported physician-diagnosed OA and were available for analysis. Tothis end, we have fit a multivariable survival regression model. The effectof BMI is expressed as hazard ratios for underweight (BMI< 18.5), over-weight (BMI 25.0e29.9) and obesity (BMI 30þ), compared to normal BMI(18.5e24.9). The effects were estimated separately for males and femalesand adjusted for age (Table I).pulation-based microsimulation model of osteoarthritis in Canada,HRQLPOHEM-OA uses the Health Utilities Index Mark 3 (HUI3) as a measureof HRQL. HUI3 is based on eight attributes (vision, hearing, speech, ambu-lation, dexterity, emotional function, cognitive function, and pain)48. The attri-butes are combined into an overall index, ranging from 0.36 to 1, usinga multi-attribute utility theory model based on societal preferences. There-fore, HUI3 can be employed to estimate QALY. The initial distribution ofHUI3 in POHEM-OA is obtained directly from the CCHS. Individual HUI3 tra-jectories were simulated with a tobit regression model (accounting for theskewed distribution of HUI3 scores) that included age, gender, BMI, andOA status49. The model was derived from the 2001 cycle of the CCHS(Table I). Model fit was evaluated by the likelihood ratio test and a com-parison of the scale parameter, which estimates the standard deviation ofthe normal error term, between the reduced and saturated model.MODEL CALIBRATION AND VALIDATIONThe population modeled by POHEM-OA reflects the Canadian adulthousehold population surveyed by the CCHS. As the model simulates indi-vidual life histories, the risk of OA is influenced by individual BMI values ac-prevalence in the CCHS and BCLHD, and clinical and radiographic OA inci-dence and prevalence from the literature45,47,50,51. We also assessed the ef-0.7 in most age groups. The baseline prevalence of OAgenerated from the POHEM-OA simulation is 11.5%,13.6% in women and 9.3% in men.In Table III we provide the age-adjusted OA hazard ratiosfor the different BMI categories, separately for men andwomen. Since the main purpose of this analysis was to pro-vide the input parameters for the simulation model, confi-dence intervals are not reported. The effect of BMI isstronger in women, with the hazard ratios of 1.76 and2.03 for BMI 25e29.9 and 30þ, respectively, compared tonormal weight (BMI between 18.5 and 24.9). The corre-sponding hazard ratios for men are 1.07 and 1.69. Ratesare decreased for men and women with BMI< 18.5 (inour sample there were no cases of OA among men withBMI< 18.5).n CTable IIAge/sex-specific incidence rates of physician-diagnosed OA per1000 person-years in British Columbia, CanadaAge range Men Women20e24 0.54 0.6325e29 0.91 0.8230e34 1.54 1.5535e39 2.55 2.5940e44 4.05 4.0945e49 6.22 7.32Table IIIEffect of BMI on OA incidence in the NPHS in Canada(2000e2002)UnderweightBMI< 18.5Normal weight18.5BMI< 25Overweight25BMI< 30ObeseBMI 30Females 0.33 1.0 1.76 2.03Males 0.00 1.0 1.07 1.69The estimates are age-adjusted hazard ratios from a survival re-ARTICLE IN PRESS4 J. A. Kopec et al.: A microsimulation model of osteoarthritis05000001000000150000020-2425-2930-3435-3940-4445-4950-5455-5960-6465-6970-7475-7980-8485-89Age GroupPopulatioCensusPOHEMPOHEM-based counts use the 2001 Canadian Community HealthSurvey as baseline population. Census-based counts are estimatedusing the 2006 Canadian Census20000002500000ounfects of OA definition and run-in time on the incidence and prevalence of OAin administrative data45. In addition, we have estimated the sensitivity andspecificity of the administrative definition of OA by linking the BCLHD toa population-based cohort study of 171 subjects with extensive clinical andradiographic data52. Finally, our model for predicting HUI3 in the generalpopulation has been validated by comparing the predicted and observedHUI3 scores in a different cycle of the CCHS.ResultsPOHEM-OA has a large number of parameters, includingdescriptive parameters, such as age and sex distribution atbaseline, birth and death rates by age, sex, and year, edu-cation and income distribution, age/sex-specific OA preva-lence proportions at baseline, age/sex-specific OAincidence rates, and other parameters. In this article we dis-cuss the key input parameters related to the frequency andhealth impact of OA (Table I).3000000tcording to the relative risk derived from the NPHS. In order that theaggregated age/sex-specific incidence rates in the simulated population cor-respond to the observed population rates in administrative data, given thepopulation distribution of BMI, the age/sex-specific rates for the referenceBMI category need to be estimated. This was achieved through model cali-bration, whereby the reference rates were iteratively adjusted using numer-ical methods until the simulated and observed incidence rates (within age/sex categories) agreed18,19.The population distribution by age from 2001 to 2007 in POHEM-OA hasbeen compared to the actual Canadian population-based on estimates fromthe 2001 and 2006 censuses. To validate the simulated OA frequency, wecompared the incidence and prevalence of OA in POHEM-OA with self-re-ported incidence of arthritis and rheumatism in the NPHS, self-reported OAFig. 2. Comparisons of POHEM-based and Census-based popula-tion counts for Canada in 2007, by age group.Please cite this article in press as: Kopec JA et al., Development of a poOsteoarthritis Cartilage (2009), doi:10.1016/j.joca.2009.10.010In Fig. 2, we compare the simulated age distribution inPOHEM-OA in 2007 with the distribution estimated fromthe 2006 census. The data show a nearly perfect corre-spondence between the two datasets. Minor discrepanciesare mainly due to the exclusion of the institutionalized pop-ulations in the CCHS and differences between the weightsdeveloped for the CCHS in 2001 and the revised censusestimates.Age/sex-specific incidence rates of physician-diagnosedOA in POHEM-OA are presented in Table II. The rates in-crease approximately linearly with age in both sexes be-tween the ages of 50 and 80 and plateau in the very old.In persons 50þ, women have substantially higher ratesthan men, with the male-to-female ratio between 0.6 and50e54 8.10 12.0455e59 11.36 18.2160e64 14.66 22.4865e69 17.59 26.9370e74 20.30 31.0075e79 23.26 34.4780e84 24.22 33.7785e89 25.77 33.4290þ 25.54 31.55Incident OA was defined as at least two visits to a physician withICD-9 code 715 within 1e730 days or one hospital diagnosis witha diagnostic code 715 in a person not diagnosed with OA in the pre-vious 12 years. Rates are calculated for the period from April 1st2003 to March 31st, 2004, based on data from the BC LinkedHealth Database.gression model. OA is defined as self-reported physician-diag-nosed OA. BMI¼weight/height2.pulation-based microsimulation model of osteoarthritis in Canada,ARTICLE IN PRESSIn POHEM-OA, HRQL (measured by HUI3) is predictedby OA, age, sex and BMI, using a tobit model49 (Table IV).The effects of OA and BMI on HUI3 were different in menand women. This difference in effects is captured by the in-teractions between BMI and sex and between OA and sexthat were highly significant. The average impact of OA diag-Table IVEffects of age, sex, OA diagnosis and BMI category on HUI3 scoresin the CCHS (2001)Regression coefficientIntercept 0.62Age 12e19 0.30Age 20e29 0.29Age 30e39 0.29Age 40e49 0.24Age 50e59 0.20Age 60e69 0.19Age 70e79 0.15Age 80þ (ref) 0.00Sex e male 0.06Sex e female (ref) 0.00OA diagnosis e yes 0.10OA diagnosis e no (ref) 0.00BMI 0e18.4 0.05BMI 18.5e24.9 0.08BMI 25.0e29.9 0.06BMI 30þ (ref) 0.00BMI (0e18.4)Sex (M) 0.07BMI (18.5e24.9)Sex (M) 0.06BMI (25.0e29.9)Sex (M) 0.03BMI (30þ)Sex (ref) 0.00OA (yes)Sex (M) 0.04OA (no)Sex (ref) 0.00HUI3¼Health Utilities Index Mark 3.Osteoarthritis and Cartilage Vol. jj, No. jjnosis is to decrease HUI3 score by 0.10 in women and 0.14in men. Age has a negative effect on HUI3 and womenhave, on average, lower HUI3 scores than men. Model fitwas adequate, as suggested by the estimated error termwhich was very similar in the reduced and saturated models(data not shown). The likelihood ratio test was significant,indicating that additional interaction terms might improvethe fit of the model, but this test is less informative whenthe sample size is very large.DiscussionThis article describes the methodology and key parame-ters for a population-based simulation model of OA in Can-ada, with a focus on OA incidence and its impact on qualityof life. Most of the input parameters have been derived fromthe Canadian Census and vital statistics databases, nationalsurveys, administrative data in British Columbia, and thehealth literature. Advanced statistical methods have beenused to derive time-to-event distributions and the relation-ships between the variables. Model calibration ensures thatthe simulated incidence rates agree with the observed rates.The model uses the POHEM platform, a state-of-the-art mi-crosimulation tool that can simulate a dynamic population as itchanges over time, as opposed to modeling a fixed cohort ora stationary population. Events are modeled in continuoustime and there is no limit on the number of events that can besimulated. An important feature of POHEM-OA is the transpar-ency of the model. The structure of the model, parametervalues, and sourcesof data are publicly available. Assumptionsand limitations of the model are acknowledged and discussed.Please cite this article in press as: Kopec JA et al., Development of a poOsteoarthritis Cartilage (2009), doi:10.1016/j.joca.2009.10.010OA incidence rates in POHEM-OA are higher than thoseestimated by Oliveria et al. for radiographic and symptomaticOA of the knee, hip and hand, derived from an administrativedatabase in the US50. This is epidemiologically plausiblesince our definition included all OA sites and did not requireradiographic confirmation. At the same time, our rates arelower than published incidence rates of self-reported ‘‘arthri-tis or rheumatism’’ in Canada51. Direct comparisons betweenthe clinical and administrative diagnosis of OA are scarce.Harrold et al.53 found a positive predictive value of 62% inthe US using medical records as a gold standard. In our pre-vious validation study, the administrative definition of OA hadsensitivity around 30% and specificity over 90% against a di-agnosis based on clinical and radiographic criteria52.The simulation-derived estimates of baselineOA prevalencein POHEM-OA are higher than those directly obtained from ad-ministrative data45. The most likely reason is the tendency foradministrative data to underestimate OA prevalence due to in-sufficient run-in time. This tendency has been previously dem-onstrated in our database45. Our estimates of baselineprevalence are consistent with the most recent incidence ratesobserved in the data. Although our previous study using the BCdatabasesuggestedan increase inage-standardized incidencerates of OA among women between 1996 and 2003, data onlong-term trends inOA incidencearenotavailable3.For this rea-son, and to simplify the analysis, we assumed that the age/sex-specific incidence rates in the past were constant. Baselineprevalence in our model is overestimated if the incidence rateswere lower prior to baseline (i.e., if there was an increasingtrend), and underestimated if the incidence rates were higher,but the difference is very unlikely to be substantial.The risk of developing OA in POHEM-OA is determined byage, gender and BMI. The effect of BMI on the OA incidencerate was estimated from a longitudinal population survey inCanada using self-reported physician-diagnosed OA. Com-parative data for a similar definition of OA are not available.In a studyof radiographic kneeOAbyFelsonet al.41 in theFra-mingham cohort, the age-adjusted odds ratios in the highestand second highest quintiles of height-adjusted weight were2.07 and 1.44, respectively, for women, and 1.51 and 1.00, re-spectively, for men. Given the differences in definitions andmethodology between the two studies, the results are remark-ably similar to the hazard ratios from our model (Table III).The effects of age, sex and OA on HUI3 were estimatedfrom the CCHS, a large national survey in Canada. Com-parative population data for a similar definition of OA arenot available. Schultz and Kopec54 previously estimatedthe overall effect of ‘‘arthritis and rheumatism’’ on HUI3 at0.09 using data from the 1996/7 NPHS and adjusting forco-morbidity. Similar to the current study, the effect wasstronger in men. In a graphical analysis, our model pre-dicted HUI3 scores observed in a different CCHS cyclewith acceptable accuracy (data not shown).While the results of preliminary validation are promising,there is a need for more validation studies. Future validationwill include re-evaluation of the current input parameters,comparisons of model output with the actual trends ob-served in administrative and survey data from several Ca-nadian provinces, and stochastic sensitivity analyses.The definition of OA adopted in our model is useful froma healthcare utilization perspective but it has clear limita-tions. Our model is unable to distinguish between differentOA sites, which limits its ability to model heterogeneous ef-fect of risk factors on OA incidence. In terms of modelingthe impact of OA on quality of life, many cases of symptom-5atic or even disabling OA remain undiagnosed. The effect ofour OA definition on the estimates of disease burden inpulation-based microsimulation model of osteoarthritis in Canada,ofre n ts o est ed.edg srs t ud au vist.sup e s as graian s o h Rxs rt a yp teS ep ePrince Edward Island20e24 7,584 5.8 2,112,568 8.260e64 7,152 5.5 1,244,611 4.865e69 6,842 5.2 1,151,556 4.5ARTICLE IN PRESS. A. Kopec et al.: A microsimulation model of osteoarthritisPOHEM-OA will be evaluated in a sensitivity analysis. Mod-eling joint pain and other symptoms in the population, asopposed to physician-diagnosed OA, might be a potentiallyuseful alternative in future studies.We acknowledge that the values of the specific parame-ters in POHEM-OA are potentially subject to criticism andmay need to be modified as new and better data becomeavailable. Different studies may produce different estimatesof the incidence, prevalence, relative risk, or health impactof OA. A limitation of the current version of POHEM-OA isthat it does not include other factors that may potentially af-fect the incidence of OA, such as geographic region, race,socio-economic status, injury, physical activity, or familyhistory of OA2. Data on the distribution of these factors inthe population, changes over time, and causal effects onthe risk of OA are insufficient at this time. Projections of dis-ease incidence and prevalence from our model may be in-accurate if there are temporal changes in the distribution ofthese factors, as well as other and unknown risk factors forOA. This limitation can be minimized by introducing a cor-rection for trend in incidence rates based on historicaldata. Furthermore, the effect of BMI on OA incidence inthe model is derived from data on self-reported physician-diagnosed OA of any joint. While this effect may be attenu-ated due to unavoidable misclassification of the disease, itis based on the best Canadian data currently available.At present, POHEM-OA does not contain any parametersdescribing the effects of specific interventions on changesin health behaviors, such as diet or healthcare utilization, indifferent population groups. Implementing such parameterswould allow amore realistic simulation of the impact of healthpolicies across various segments of the population. Althoughthe available data are limited, we intend to incorporate behav-ioral effects into future versions of the model.Despite its limitations, the current model can inform policyby providing information not directly available from epidemio-logical studies. Its main purpose is to study the effects ofchanges in the distribution of risk factors for OA on the futureburden of this disease in a dynamically changing population.Current policy analyses typically make simplifying assump-tions about the population under study and the effects ofhealth interventions at the population level15,16. In particular,standard methods of calculating attributable fraction usuallyassume that the risk factor is eliminated and that the underly-ing population is static. In addition, such methods generallydo not consider the impact of disease on quality of life14.The current model can be used to project the future burdenof OA in Canada while taking into account trends in popula-tion aging and obesity prevalence. For example, we have re-cently applied the model to compare different scenarios forfuture OA incidence rates in terms of their impact on health-adjusted life expectancy (HALE) inCanada55. Preliminary re-sults show that if new cases ofOAdue to excessweight couldbe avoided, average HALE would improve by 2e4 monthsand the gain would be somewhat greater in women. Anotherexample is the use of themodel to analyze the relative contri-butions of aging and obesity to the projected increase in thenumber of persons with OA56.In conclusion, we have developed the first population-based microsimulation model of OA. We hope the modelwill provide a benchmark for the synthesis of quantitativedata on the epidemiology and population health impact ofOA. We expect to be able to improve the model as gaps inknowledge are addressed in future studies. While the initialresults from this model will have to be taken with caution,6 Jthe confidence in the results will increase as the validity ofthe model is further confirmed under a variety of conditions.Please cite this article in press as: Kopec JA et al., Development of a poOsteoarthritis Cartilage (2009), doi:10.1016/j.joca.2009.10.01070e74 6,360 4.9 1,003,709 3.975e79 5,237 4.0 740,459 2.980þ 5,794 4.4 749,088 2.925e29 8,742 6.7 2,006,021 7.830e34 10,281 7.9 2,158,989 8.435e39 12,447 9.5 2,587,642 10.040e44 12,886 9.8 2,707,970 10.545e49 11,388 8.7 2,369,433 9.250e54 10,255 7.8 2,051,946 8.055e59 8,355 6.4 1,585,225 6.1Nova Scotia 5,319 4.1 787,972 3.1New Brunswick 4,996 3.8 634,264 2.5Quebec 22,012 16.8 6,216,722 24.1Ontario 39,278 30.0 9,877,292 38.3Manitoba 8,470 6.5 907,494 3.5Saskatchewan 8,009 6.1 805,993 3.1Alberta 14,456 11.0 2,481,568 9.6British Columbia 18,302 14.0 3,421,671 13.3Yukon and NWT 2,517 1.9 76,928 0.3Total 130,880 100.0 25,787,334 100.0NWT¼Northwest Territories.Table A2Distribution of the CCHS (2001) population by age groupAge group Sample Population (weighted)N % N %12e14 6,476 4.9 1,186,119 4.615e19 11,081 8.5 2,131,999 8.3Total 130,880pulation-based microsim100.0 2ulation model of o3,651 2.8 116,3275,787,335steoarthritis in C0.5Newfoundland 3,870 3.0 461,104 1.8N % N %Province Sam le Population (w ighted)Distribution of the CCH (2001) population by province of r sidenceTabl A1are used.ning POHEM, more recise es imates for all ages and yearsand shown for selec ed ages nd/or ears only. When run-shown for illustration purpose . The estimates are oundedPopulation parametersTables A1eA7 shows the distribution of key POHEM-OAvariables in the CCHS. These tables include all CCHS re-spondents. Weighted data reflect the distribution in the Ca-nadian household population. Data in Tables A8eA10 areAppendithe Canad Institute f Healt esearch.Source of port: Th tudy w funded by a nt frommanuscripThe autho hank Cla e Nade for help in re ing theAcknowl ementThere a o conflic f inter to be disclosConflict interest100.0anada,3ribu CC 001 sexle eig6713 14Distribution of (2 pula f edIncome Saabn S ) p MBMI (w4 48 0pI 39 69 99 80.5e0.599 800.0Table A8en Wo0.10.92.97.96.8 24.8 6able Arence nce r OA p 0 pe ars fted aen Wo10y rates in Canada p population for seleand years (in projected rates)2001 200 2011 20160.59 0.5 0.49 0.470.73 0.7 0.66 0.630.84 0.8 0.76 0.722.07 2.0 1.87 1.785.73 5.4 5.19 4.94win de the hmhe cur e m n:EM romARTICLE IN PRESS7Osteoarthritis and Cartilage Vol. jj, No. jj0.6e0.699 6,031 4.6 1,096,935 4.30.7e0.799 8,545 6.5 1,578,011 6.10.8e0.899 12,469 9.5 2,388,627 9.30.9e0.999 57,344 43.8 11,504,065 44.6Please citeOsteoarthritthis article in press as: Kois Cartilage (2009), doi:102,422,431.91.9427,027425,220(continued on nepec JA et al., Devel.1016/j.joca.2009.101.71.70.3e0.390.4e0.492,97 2.3 501,022 1.90.2e0.29 1,91 1.5 308,631 1.20.1e0.19 1,08 0.8 180,531 0.7HUI3 Sample Population (weighted)N % N %<0.0 799 0.6 138,751 0.50.0e0.099 888 0.7 151,656 0.69 2S (2001)ndex Markopulatio3 (HUIn by the Health)UtilitiesDistribution of the CCHTable A7Total 130,8 0 100. 25,787,334 100.0Not stated 1,8 0 1. 279,962 1.2Not applicable 42,8 6 32. 7,189,595 27.9N % N %Underweight (BMI< 20) 6,040 4.6 1,473,150 5.7Acceptable (BMI 20e24.9) 35,404 27.1 7,944,017 30.8Overweight (BMI 25) 44,730 34.2 8,882,610 34.46 8Sample Population eighted)DistributioTof the CCHle A6(2001 opulation by B ITotal 130,880 100.0 25,787,334 100.0Not stated 14,088 10.8 2,632,077 10.2Not applicable 6,476 4.9 1,186,119 4.6$80,000 or more 4,524 3.5 1,081,049 4.2$50,000e79,999 13,578 10.4 2,945,634 11.4$30,000e49,999 24,220 18.5 5,191,844 20.1N % N %None 5,704 4.4 1,327,650 5.1<$15,000 34,229 26.2 6,141,039 23.8$15,000e29,999 28,061 21.4 5,281,922 20.5mple Population (weighted)Distribution of the CTable A5CHS (2001) population by incomeTotal 130,880 100.0 25,787,335 100.0Not stated 1,264 1.0 217,774 0.8Some post-secondary 9Post-secondary grad. 52,832,586 47.50.22,107,60111,108,4148.243.1Education Sample Population (weighted)N % N %Less than secondary 44,338 33.9 7,594,745 29.5Secondary 22,860 17.5 4,758,801 18.5the CCHS 001) po tion by level o ucationTable ATotal 0,880 00.0 25,787,335 100.0Female 0,366 53.8 13,073,184 50.7Male 0,514 46.2 12,714,150 49.3N % N %Sex Samp Population (w hted)Dist tion of the HS (2 ) population byTable Axt page)opment of a po.010the simulation start year (2001).(2) At each birthday in the person’s simulated life, annu-(1) POHalized hpulation-basedselects a record fazard h is calculatedmicrosimulation model othe CCHS database inn OA oc s in th icrosimulatioas: h ¼ h0RR.f osteoarthritis in CThe follotermining wg steps scribe basic algorit for de-INCIDENCE EL70 14.43 13.73 13.07 12.44 11.8480 34.41 32.75 31.17 29.67 28.2490 81.33 77.40 73.67 70.11 66.73MOD60 5 4.7050 0 1.6940 0 0.6930 0 0.6020 2 0.45Men90 1 66.8180 6 28.0820 0.26 0.25 0.24 0.24 0.2330 0.35 0.34 0.33 0.32 0.3140 0.69 0.68 0.66 0.64 0.6350 1.78 1.74 1.67 1.65 1.6160 4.70 4.58 4.47 4.36 4.2570 12.28 11.98 11.69 11.40 11.1231.03 30.2 29.52 28.7973.83 72.0 70.23 68.50WomenAge 6 2021cludingMortalit er 1000 cted agesTable A90 28.1 31.130 1.4 1.240 3.5 3.150 6.8 8.260 13.1 16.470 18.3 23.280 23.8 29.9Ian20 0.5 0.6Age M menlec gesRefe incide ates of er 100 rson-ye or se-T 990 5 8.280 4 6.170 29.2 39.83.0 560 1 2.250 8.840 3.030 0.920 0.1Age M menseline lence ( OA fo cted agBa preva %) of r sele esTotal 130,880 100.0 25,787,334 1Table A7 (continued )HUI3 Sample Population (weighted)N % N %1.0 32,272 24.7 6,855,037 26.6Missing 1,689 1.3 231,822 0.9n thisada,the longitudinal NPHS 1996e2004. The model predicts cur-DT,nd se-lected musculoskeletal disorders in the United States. Arthritis Rheum1998;41(5):778e99.ARTICLE IN PRESS8 J. A. Kopec et al.: A microsimulation model of osteoarthritis2. Felson DT, Lawrence RC, Hochberg MC, McAlindon T, Dieppe PA,Minor MA, et al. Osteoarthritis: new insights. Part 1: the diseaseand its risk factors. Ann Intern Med 2000;33(8):635e46.3. Kopec JA, Rahman MM, Sayre EC, Cibere J, Flanagan WM,Aghajanian J, et al. Trends in physician-diagnosed osteoarthritis inci-dence in an administrative database in British Columbia, Canada,1996e1997 through 2003e2004. Arthritis Rheum 2008;59(7):929e34.4. Badley EM, Wang PP. Arthritis and the aging population: projections ofarthritis prevalence in Canada 1991e2031. J Rheumatol 1998;25(1):138e44.5. Hootman JM, Helmick CG. Projections of US prevalence of arthritis andassociated activity limitations. Arthritis Rheum 2006;54(1):226e9.6. Perruccio AV, Power JD, Badley EM. Revisiting arthritis prevalence pro-jectionseit’s more than just the aging of the population. J Rheumatol2006;33(9):1856e62.7. Woolf AD. The bone and joint decade. Strategies to reduce the burdenof disease: the Bone and Joint Monitor Project. J Rheumatol Suppl2003;67:6e9. Review.8. Bijlsma JW, Knahr K. Strategies for the prevention and management ofosteoarthritis of the hip and knee. Best Pract Res Clin Rheumatol1. Lawrence RC, Helmick CG, Arnett FC, Deyo RA, FelsonGiannini EH, et al. Estimates of the prevalence of arthritis arent BMI within 14 groups defined by BMI category and age,separately for men and women, based on BMI history andother covariates, such as region of residence, income andeducation. The response variable is change in BMI com-pared to 2 years prior, treated as a continuous, normally dis-tributed variable. The model includes the most recent BMIvalue and up to three prior changes in BMI. The modelcan be presented in a simplified form as:DBMIt ;tþ2 ¼ aþ b1BMIt þ b2DBMIt2;t þ b3DBMIt4;t2þ b4DBMIt6;t4 þ b5Incomet6þ b6Educationt6 þ b7Regiont6In this equation, DBMIi;j ¼ BMIj  BMIi is the difference inBMI between time i and time j.There are 112 regression equations arising from 28 strata(modeling groups) generated by the 14 ageBMI cate-gories and the sex variable (male and female). Withineach stratum, there are four models, depending on thenumber of prior BMI values used.Referencesequation h0 is the age/sex-specific OA incidence ratefor persons with normal BMI estimated from BCLHDrates using calibration methods (Table A9 in the Ap-pendix), and RR is the relative risk (hazard ratio) ofOA based on person’s BMI category (Table III in themanuscript).(3) A random number u between 0 and 1 is generatedand the time of event (in years) is estimated as t ¼ln(1 u)/h. If t  1 no event occurs during the year.Note that competing events, such as death, couldcensor this event.(4) Steps 2e3 are repeated at every subsequent birthdayuntil the person develops OA or dies.(5) Steps 1e4 are repeated for every individual record inour start-up database.BMI MODELThe BMI model is part of the POHEM software and wasdeveloped by Statistics Canada using biannual data from2007;21(1):59e76.Please cite this article in press as: Kopec JA et al., Development of a poOsteoarthritis Cartilage (2009), doi:10.1016/j.joca.2009.10.0109. Vignon E, Valat JP, Rossignol M, Avouac B, Rozenberg S, Thoumie P,et al. Osteoarthritis of the knee and hip and activity: a systematic in-ternational review and synthesis (OASIS). Joint Bone Spine 2006;73(4):442e55.10. Zhang W, Moskowitz RW, Nuki G, Abramson S, Altman RD, Arden N,et al. OARSI recommendations for the management of hip and kneeosteoarthritis, Part II: OARSI evidence-based, expert consensusguidelines. Osteoarthritis Cartilage 2008;16(2):137e62.11. ArthritisFoundation,AssociationofStateandTerritorialHealthOfficials,Cen-ters for Disease Control and Prevention. National Arthritis Action Plan. APublic Health Strategy. 1999. Internet address, http://ww2.arthritis.org/resources/ActionPlanInterior.pdf; 2008 [accessed 20.01.09].12. Lohmander LS, Gerhardsson de Verdier M, Rollof J, Nilsson PM,Engstro¨m G. Incidence of severe knee and hip osteoarthritis in rela-tion to different measures of body mass. A population-based prospec-tive cohort study. Ann Rheum Dis 2009;68(4):490e6.13. Luo W, Morrison M, de Groh M, Waters C, DesMeules M, Jones-McLean E, et al. The burden of adult obesity in Canada. ChronicDis Can 2007;27(4):135e83.14. Bartlett C, Doyal L, Ebrahim S, Davey P, Bachmann M, Egger M, et al.The causes and effects of socio-demographic exclusions from clinicaltrials. iiieiv, ixex. Health Technol Assess 2005;9(38):1e152.15. Andrews G, Simonella L, Lapsley H, Sanderson K, March L. Evidence-based medicine is affordable: the cost-effectiveness of current com-pared with optimal treatment in rheumatoid and osteoarthritis. J Rheu-matol 2006;33(4):671e80.16. Segal L, Day SE, Chapman AB, Osborne RH. Can we reduce diseaseburden from osteoarthritis? An evidence-based priority-setting model.Med J Aust 2004;180(Suppl 5):S11e7.17. Fone D, Hollinghurst S, Temple M, Round A, Lester N, Weightman A,et al. Systematic review of the use and value of computer simulationmodelling in population health and health care delivery. J PublicHealth Med 2003;25(4):325e35.18. Weinstein MC. Recent developments in decision-analytic modelling foreconomic evaluation. Pharmacoeconomics 2006;24(11):1043e53.19. Feuer EJ, Etzioni R, Cronin KA, Mariotto A. The use of modeling to un-derstand the impact of screening on U.S. mortality: examples frommammography and PSA testing. Stat Methods Med Res 2004;13(6):421e42.20. Brown JB, Russell A, Chan W, Pedula K, Aickin M. The global diabetesmodel: user friendly version 3.0. Diabetes Res Clin Pract 2000;50(Suppl 3):S15e46.21. Le Lay A, Despiegel N, Francois C, Duru G. Related Articles. Can dis-crete event simulation be of use in modelling major depression?Cost Eff Resour Alloc 2006;5(4):19e30.22. Bansback NJ, Regier DA, Ara R, Brennan A, Shojania K, Esdaile JM,et al. An overview of economic evaluations for drugs used in rheuma-toid arthritis: focus on tumour necrosis factor-alpha antagonists. Drugs2005;65(4):473e96.23. Marshall JK, Pellissier JM, Attard CL, Kong SX, Marentette MA. Incre-mental cost-effectiveness analysis comparing rofecoxib with nonse-lective NSAIDs in osteoarthritis: Ontario ministry of healthperspective. Pharmacoeconomics 2001;19(10):1039e49.24. Zabinski RA, Burke TA, Johnson J, Lavoie F, Fitzsimon C, Tretiak R,et al. An economic model for determining the costs and conse-quences of using various treatment alternatives for the managementof arthritis in Canada. Pharmacoeconomics 2001;19(Suppl 1):49e58.25. Haglund U, Svarvar P. The Swedish ACCES model: predicting the healtheconomic impact of celecoxib in patients with osteoarthritisor rheumatoid arthritis. Rheumatology (Oxford) 2000;39(Suppl 2):51e6.26. Chancellor JV, Hunsche E, de Cruz E, Sarasin FP. Economic evaluationof celecoxib, a new cyclo-oxygenase 2 specific inhibitor. Pharmacoe-conomics 2001;19(Suppl 1):59e75.27. Maetzel A, Krahn M, Naglie G. The cost effectiveness of rofecoxib andcelecoxib in patients with osteoarthritis or rheumatoid arthritis. ArthritisRheum 2003;49(3):283e92.28. Will BP, Berthelot J-M, Nobrega KM, Flanagan W, Evans WK. Canada’sPopulation Health Model (POHEM): a tool for performing economicevaluations of cancer control interventions. Eur J Cancer 2001;37(14):1797e804.29. Wolfson MC. POHEMda framework for understanding and modellingthe health of human populations. World Health Stat Q 1994;47(3e4):157e76.30. Law A, Kelton WD. Simulation Modeling and Analysis. 3rd edn.McGraw-Hill; 199931. Gupta A, Harding A, Eds. Modeling Our Future. Population Aging,Health and Aged Care. Amsterdam: Elsevier; 200732. WillBP,LePetitC,BerthelotJ-M,TomiakEM,VermaS,EvansWK.Diagnos-tic and therapeuticapproaches fornonmetastaticbreast cancer inCanada,and their associated costs. Br J Cancer 1999;79(9e10):1428e36.33. Will BP, Nobrega KM, Berthelot J-M, Flanagan W, Wolfson MC,Logan DM, et al. First do no harm: extending the debate on the pro-vision of preventive tamoxifen. Br J Cancer 2001;85(9):1280e8.pulation-based microsimulation model of osteoarthritis in Canada,34. Evans WK, Will BP, Berthelot J-M, Logan DM, Mirsky DJ, Kelly N.Breast cancer: better care for less cost: is it possible? Int J TechnolAssess Health Care 2000;16(4):1168e78.35. Will BP, Berthelot J-M, Le Petit C, Tomiak E, Verma S, Evans WK.Estimates of the lifetime costs of breast cancer treatment in Canada.Eur J Cancer 2000;36(6):724e35.36. Berthelot J-M, Will BP, Evans WK, Coyle D, Earle CC, Bordeleau L.Decision framework for chemotherapeutic interventions for meta-static non-small cell lung cancer. J Natl Cancer Inst 2000;92(16):1321e9.37. Flanagan W, Le Petit C, Berthelot J-M, White KJ, Coombs BA, Jones-McLean E. Potential impact of population-based colorectal cancerscreening in Canada. Chronic Dis Can 2003;24(4):81e8.38. Maroun J, Ng E, Berthelot J-M, Le Petit C, Dahrouge S, Flanagan WM,et al. Lifetime costs of colon and rectal cancer management in Can-ada. Chronic Dis Can 2003;24(4):91e101.39. Statistics Canada. Canadian Community Health Survey (CCHS) Cycle1.1 Internet address, http://www.statcan.gc.ca/cgi-bin/imdb/p2SV.pl?Function=getSurvey&SurvId=3226&SurvVer=0&InstaId=15282&InstaVer=1&SDDS=3226&lang=en&db=imdb&adm=8&dis=2. [accessed:20.01.09]40. George MV, Loh S, Verma RBP, Shin YE. Population Projections forCanada, Provinces and Territories 2000e2026. Statistics Canada,Demography Division. Catalogue no. 91-520-XPB ISBN 0-660-60634-8. Ministry of Industry; 2001.41. Felson DT, Anderson JJ, Naimark A, Walker AM, Meenan RF. Obesityand knee osteoarthritis. The Framingham study. Ann Intern Med1988;109(1):18e24.42. Statistics Canada. National Population Health Survey (NPHS) Internetaddress, http://www.statcan.ca/english/concepts/nphs/; 1988. Lastupdated: 2004-08-20.43. National PopulationHealthSurvey, Cycle6SyntheticMasterFile, 2007. Sta-tistics Canada. Health Statistics Division. Internet address, http://data.library.ubc.ca/java/jsp/database/production/detail.jsp?id ¼ 1092; 1988.44. ChamberlayneR,GreenB,BarerML,HertzmanC,LawrenceWJ,ShepsSB.Creating a population-based linked health database: a new resource forhealth services research. Can J Public Health 1998;89(4):270e3.45. Kopec JA, Rahman MM, Berthelot JM, Le Petit C, Aghajanian J,Sayre EC, et al. Descriptive epidemiology of osteoarthritis in BritishColumbia, Canada. J Rheumatol 2007;34(2):386e93.46. Bedson J, Mottram S, Thomas E, Peat G. Knee pain and osteoarthritis inthe general population: what influences patients to consult? FamPract 2007;24(5):443e53.47. Flanagan WM, Kopec JA, Fine`s P, Rahman MM, Sayre EC, Cibere J,et al. Estimating prevalence of physician-diagnosed osteoarthritis inCanada using microsimulation (Abstract). Osteoarthritis Cartilage2008;16(Suppl 4):S153.48. Feeny D, Furlong W, Torrance GW, Goldsmith CH, Zhu Z, DePauw S,et al. Multiattribute and single-attribute utility functions for the healthutilities index mark 3 system. Med Care 2002;40(2):113e28.49. Austin PC, EscobarM, Kopec JA. The use of the Tobit model for analyzingmeasures of health status. Qual Life Res 2000;9(8):901e10.50. Oliveria SA, Felson DT, Reed JI, Cirillo PA, Walker AM. Incidence ofsymptomatic hand, hip, and knee osteoarthritis among patients ina health maintenance organization. Arthritis Rheum 1995;38(8):1134e41.51. Wilkins K. Incident arthritis in relation to excess weight. Health Rep2004;15:39e49.52. Rahman MM, Aghanjanian J, Kopec JA, Cibere J. Validation of osteoar-thritis diagnosis in administrative data using a clinically and radiolog-ically defined population-based cohort of osteoarthritis (Abstract).Osteoarthritis Cartilage 2008;16(Suppl 4):S150.53. Harrold LR, Yood RA, Andrade SE, Reed JI, Cernieux J, Straus W, et al.Evaluating the predictive value of osteoarthritis diagnoses in an ad-ministrative database. Arthritis Rheum 2000;43(8):1881e5.54. Schultz SE, Kopec JA. Impact of chronic conditions. Health Rep 2003;14(4):41e53.55. Kopec JA, Fine`s P, Flanagan WM, Sayre EC, Rahman M, Bansback N,et al. Projecting the burden of osteoarthritis in Canada using microsi-mulation (Abstract). Ann Rheum Dis 2009;68(Suppl 3):99.56. Fine`s P, Flanagan W, Kopec JA, Rahman MM, Sayre EC, Cibere J,et al. Projecting number of osteoarthritis patients from 2001 to 2021in Canada using microsimulation (Abstract). Arthritis Rheum 2007;56(Suppl 9):S341.ARTICLE IN PRESS9Osteoarthritis and Cartilage Vol. jj, No. jjPlease cite this article in press as: Kopec JA et al., Development of a poOsteoarthritis Cartilage (2009), doi:10.1016/j.joca.2009.10.010pulation-based microsimulation model of osteoarthritis in Canada,

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.52383.1-0132726/manifest

Comment

Related Items