UBC Faculty Research and Publications

Socioeconomic status and hospital utilization among younger adult pneumonia admissions at a Canadian… McGregor, Margaret J; Reid, Robert J; Schulzer, Michael; Fitzgerald, J M; Levy, Adrian R; Cox, Michelle B Nov 25, 2006

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

Item Metadata


52383-12913_2006_Article_329.pdf [ 522.14kB ]
JSON: 52383-1.0135719.json
JSON-LD: 52383-1.0135719-ld.json
RDF/XML (Pretty): 52383-1.0135719-rdf.xml
RDF/JSON: 52383-1.0135719-rdf.json
Turtle: 52383-1.0135719-turtle.txt
N-Triples: 52383-1.0135719-rdf-ntriples.txt
Original Record: 52383-1.0135719-source.json
Full Text

Full Text

ralssBioMed CentBMC Health Services ResearchOpen AcceResearch articleSocioeconomic status and hospital utilization among younger adult pneumonia admissions at a Canadian hospitalMargaret J McGregor*1,2, Robert J Reid3,4, Michael Schulzer1, J Mark Fitzgerald1, Adrian R Levy3,5 and Michelle B Cox1Address: 1Centre for Clinical Epidemiology and Evaluation, Family Practice Research Office, 828 West 10th Avenue, Vancouver, BC, Canada, 2Department of Family Practice, Department of Medicine, University of British Columbia, 5950 University Boulevard, Vancouver, BC, Canada, 3Department of Health Care and Epidemiology, Faculty of Medicine, University of British Columbia, 5804 Fairview Avenue, Vancouver, BC, Canada, 4Group Health Cooperative, 1730 Minor Avenue, Suite 1600, Seattle, WA, USA and 5Centre for Health Evaluation & Outcome Sciences (CHÉOS), St. Paul's Hospital, 620-1081 Burrard Street, Vancouver, BC, CanadaEmail: Margaret J McGregor* - mrgret@telus.net; Robert J Reid - reid.rj@ghc.org; Michael Schulzer - Michael@stat.ubc.ca; J Mark Fitzgerald - mark.fitzgerald@vch.ca; Adrian R Levy - alevy@interchange.ubc.ca; Michelle B Cox - michellebcox@shaw.ca* Corresponding author    AbstractBackground: Although the general association between socioeconomic status (SES) and hospitalizationhas been well established, few studies have considered the relationship between SES and hospital lengthof stay (LOS), and/or hospital re-admission. The primary objective of this study therefore, was to examinethe relationship of SES to LOS and early re-admission among adult patients hospitalized with community-acquired pneumonia in a setting with universal health insurance.Methods: Four hundred and thirty-four (434) individuals were included in this retrospective, longitudinalcohort analysis of adult patients less than 65 years old admitted to a large teaching hospital in Vancouver,British Columbia. Hospital chart review data were linked to population-based health plan administrativedata. Chart review was used to gather data on demographics, illness severity, co-morbidity, functionalstatus and other measures of case mix. Two different types of administrative data were used to determinehospital LOS and the occurrence of all-cause re-admission to any hospital within 30 days of discharge. SESwas measured by individual-level financial hardship (receipt of income assistance or provincial disabilitypension) and neighbourhood-level income quintiles.Results: Those with individual-level financial hardship had an estimated 15% (95% CI -0.4%, +32%, p =0.057) longer adjusted LOS and greater risk of early re-admission (adjusted OR 2.65, 95% CI 1.38, 5.09).Neighbourhood-level income quintiles, showed no association with LOS or early re-admission.Conclusion: Among hospitalized pneumonia patients less than 65 years, financial hardship derived fromindividual-level data, was associated with an over two-fold greater risk of early re-admission and amarginally significant longer hospital LOS. However, the same association was not apparent when anecological measure of SES derived from neighbourhood income quintiles was examined. The ecological SESvariable, while useful in many circumstances, may lack the sensitivity to detect the full range of SES effectsin clinical studies.Published: 25 November 2006BMC Health Services Research 2006, 6:152 doi:10.1186/1472-6963-6-152Received: 07 April 2006Accepted: 25 November 2006This article is available from: http://www.biomedcentral.com/1472-6963/6/152© 2006 McGregor et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.Page 1 of 10(page number not for citation purposes)BMC Health Services Research 2006, 6:152 http://www.biomedcentral.com/1472-6963/6/152BackgroundModern epidemiology studies have established a clearassociation between socioeconomic status (SES) andhealth status even after standardization for all known con-founders. Low SES has been shown to be an independentpredictor of higher mortality rates [1,2], higher diseaseprevalence [3,4], higher hospitalization rates [5-7], andpoorer treatment response and prognosis [8,9] for a widerange of illnesses across many countries with differinghealth care systems [10].Far fewer studies have examined the relation between hos-pital length of stay (LOS), as a measure of health servicesutilization, and SES. Some US studies have found LOS tobe inversely related to SES [11]; others have found noeffect [12]. In one study that examined the relationshipbetween race and hospital LOS among the elderly, Afri-can-Americans were found to have a significantly shorterLOS after adjusting for age and health status [13]. In Can-ada, Brownell and Roos found a small inverse associationbetween neighbourhood-level income quintiles, an eco-logical indicator of SES, and LOS for patients admitted toeight Manitoba hospitals for 14 common illnessesbetween 1989 and 1992 [14]. In contrast, Glazier and col-leagues found that once admitted to hospital, there wasno relationship between neighbourhood-level incomequintiles and LOS [15]. We are unaware of any Canadianstudies that have examined individual measures of SES inrelation to hospital LOS.These contrasting findings may be explained by a numberof factors. First, some studies are limited by their use ofecological measures of SES, resulting in misclassificationand potential bias toward the null, especially for smallereffects. With a greater mix of individuals at differing levelsof SES in a particular neighbourhood, this misclassifica-tion will be greater. Second, inconsistent results may bedue to differences in adjustment for important potentialconfounders beyond clinical case mix (for example, levelof function and living situation). Finally, the impact ofSES on health services utilization is likely to be influencedby access to and co-payments for hospitalization, andstudies from countries with differences in health careaccess and insurance arrangements may produce differentresults. Hofer et al. found that SES effects on hospitaliza-tion were substantially diminished when they controlledfor insurance and health status [16].In order to understand the impact of SES on LOS, it is alsoimportant to examine early re-admission. This measure isless frequently examined in relation to SES, and yet is cru-cial in understanding whether hospital stays are meetingthe needs of different socioeconomic groups equitably.less than 65 years old admitted with community-acquiredpneumonia in a health care system with universal insur-ance for hospital and physician care. We examined pneu-monia because it is one of the most common reasons formedical admission to hospital throughout the Westernworld [17]. It was hoped that by performing individual-level adjustment using broader clinical data we couldbegin to clarify the complexities of the associationbetween SES and hospital utilization.MethodsStudy setting and populationCanada has a publicly funded health care system provid-ing residents with universal insurance for medically neces-sary health care. Patients may present to any acute carehospital and receive first dollar coverage for the carereceived. Although some private facilities have opened forcertain types of surgical services, there are no private acutecare hospitals to treat medical conditions. The VancouverGeneral Hospital (VGH) is a large teaching hospital situ-ated in central Vancouver, British Columbia, which servespatients from a large geographic area including both poorand wealthy neighbourhoods. This study was a retrospec-tive longitudinal cohort analysis of adult patients ran-domly selected from the total 3,934 admissions toVancouver General Hospital who had a most responsiblediagnosis of community-acquired pneumonia (ICD-9-CM codes 481.XX – 483.XX, 485.XX, 486.XX) betweenJanuary 1, 1990 and March 31, 2001. Repeat admissionson the same individual were eligible for inclusion at thetime of the index admission and excluded thereafter.Included for computer-generated random selection wasany individual admitted for pneumonia, living in thecommunity of Vancouver or Richmond, with a valid Brit-ish Columbia Medical Services Plan number who was dis-charged alive (n = 3,934). We excluded those admittedfrom or discharged to another acute care hospital in theprovince (n = 146); outliers with an LOS greater than 3times the inter-quartile range (n = 13); or those who leftthe hospital against medical advice (n = 57). Ethicsapproval was obtained from the Ethics Board of the Uni-versity of British Columbia and the Vancouver Hospitaland Health Sciences Centre Research Advisory Commit-tee. In this study we present results on all reviewed chartsfrom patients less than 65 years old.Data sources/data collectionData were obtained from four sources that were linked, atthe individual level, using patient-specific identifiers.1) A computerized hospital discharge abstract database,used to define the cohort from which admissions werePage 2 of 10(page number not for citation purposes)The purpose of this study was to examine the relationshipof SES to hospital LOS and early re-admission for adultsrandomly sampled, provided information on age, sex, 6-BMC Health Services Research 2006, 6:152 http://www.biomedcentral.com/1472-6963/6/152digit postal code of residence, co-morbid diagnoses andhospital LOS.2) A Statistics Canada postal code conversion program[18,19] linked individuals' postal codes to enumerationarea census data on mean household income.3) Structured chart review of sampled admissions by tworegistered nurses with extensive chart review experience,gathered individual-level data on demographics, illnessseverity, functional status and other measures of case mix.4) The BC Linked Health Database, an individual-level,population-based research resource of linkable healthcare utilization and other data, provided data on all re-admissions for any cause to any BC hospital within 30days of discharge from the sampled hospitalization. Thisdatabase is developed and maintained by the Centre forHealth Services and Policy Research at the University ofBritish Columbia, in collaboration with the BritishColumbia Ministry of Health [20].Data measuresLength of stay (LOS), re-admission and socioeconomic status (SES)Hospital LOS was measured in days from the date ofadmission to the date of discharge. Our second outcomeof interest was re-admission to any acute care hospital inthe province of British Columbia for any cause within 30days of discharge from the index admission.SES was measured in two ways:1) Individual-level financial hardship measured by receiptof income assistance or provincial disability pensionTo qualify for income assistance provided by the Govern-ment of British Columbia, a person's time-limitedemployment insurance must have expired and they musthave depleted their savings. In 1996 (the approximatemidpoint of the study period), the annual income of a sin-gle BC resident on income assistance was C$7,081 [21].To qualify for a provincial disability pension, an individ-ual must prove a longstanding physical or mental disabil-ity and be ineligible for a federal disability pension. Theannual income for a single person in BC living on a pro-vincial disability pension in 1996 was C$10,784 (Table 1)[21]. An individual is not eligible to receive income assist-ance if they are receiving a disability pension, and eligibil-ity for either of these benefits applies only to those lessthan 65 years old.This variable was derived from chart review data andcoded as "yes" if there was a chart notation that the patientthe admission face sheet at the time of admission, underthe heading of "employer" and supplemented with socialworker and clinical service providers' notes.2) Neighbourhood-level income quintilesPatient residential postal codes were aggregated into cen-sus enumeration areas based on the 1996 definitionsusing software provided by Statistics Canada [18,19].Incomes (defined as income per single person equivalent)were based on census data on average household incomeand the distribution of households by size in each enu-meration area. Within the enumeration areas representedby the study population, the lowest income ranged fromC$10,950 to C$29,001 (Table 1). While not an individ-ual-level measure of income, the use of this technique hasbeen found to correlate well with individual SES in largepopulation-based analyses and has been used extensivelyin health care research as a surrogate measure for SES [16].Other variables examinedWe examined age, sex, smoking status and substancedependency as potential confounders. An individual wasconsidered to be a smoker if there was any history ofsmoking documented in the year prior to admission. Sub-stance abuse was coded as "yes" if there was documenta-tion of alcohol, cocaine, heroin, prescription or illicitsubstance dependency at the time of admission.We also measured pneumonia severity, using chart reviewdata to construct a "Pneumonia Severity Index" (PSI) [22].This index is calculated from individual demographic (ageand sex), disease co-morbidity (neoplastic disease,hepatic disease, renal failure, congestive heart failure andcerebrovascular disease), and a mix of physiologic andlaboratory measures. The index has been used clinically todetermine when pneumonia patients should be hospital-ized [22] and by researchers to explain variations in mor-tality and hospital LOS for pneumonia patients [23].We used an imputation process to derive missing values inthe PSI data. Ten percent of these data were missing andoccurred at random with no relation to age. Cases with asingle missing variable were filled in first. Using logisticregression on the data subset, the missing variable wasregressed on all other variables in the PSI list. The derivedregression was then applied to each case for which thisvariable was singly missing, and the predicted probabilityof the presence of the variable was estimated. Wheneverthis probability exceeded 0.5, the variable was entered aspresent for the case. For cases with multiple missing vari-ables, a similar logistic technique was applied, replacingone variable at a time while omitting the other missingPage 3 of 10(page number not for citation purposes)collected income assistance or received a disability pen-sion. This information was recorded by clerical staff onvariables from the model, and building up the completedataset in a stepwise manner.BMC Health Services Research 2006, 6:152 http://www.biomedcentral.com/1472-6963/6/152The variable "living alone" was coded "yes" if a patientwas classified in the chart to be living alone at the time ofadmission. Functional impairment or "any documentedproblems with activities of daily living (ADLs)" was codedas "yes" if there was documentation in the nursing notesof difficulty with feeding, mobility, bathing, dressing and/or toileting on admission. We note that one of our SESmeasures looked at "receipt of disability" and thereforehad the potential for correlation with functional impair-ment. However, disability pensions are commonly pro-vided for psychological disabilities rather than thefunctional impairments coded here.We also used a modified Deyo adaptation of Charlson co-morbidities for the ICD-9-CM codes listed as contributingto hospital stay [24,25]. Secondary diagnoses extractedfrom the hospital discharge database were used to con-struct this index. We excluded co-morbidities from theCharlson index that were already captured by the PSI var-iable discussed above, including neoplastic disease, renalData analysisAfter examining the data for outliers and possible dataentry errors, we generated descriptive statistics and calcu-lated crude mean and median lengths of stay for eachstudy population characteristic. Appropriate parametricand non-parametric tests of comparison were used forunivariate testing of LOS by the various factors. The distri-bution of each SES measure across other independentdrivers of LOS was examined with bivariate linear regres-sion models.SES and LOSWe examined the association between individual-level(receipt of income assistance or disability pension) andneighbourhood-level (income quintiles) SES measuresand LOS using two separate multiple linear regressionmodels. Due to the skewed distribution of our outcomevariable (LOS), regression analyses were carried out afterlogarithmic transformation of this variable. Forward andbackward stepwise linear regression was run entering allTable 1: Socioeconomic status and other characteristics of study sample (n = 434) §Characteristic n (%)Employable persons on income assistance (C$7,081) ‡ 84 (19)Persons receiving a provincial disability pension (C$10,784) ‡ 64 (15)Receiving income assistance or provincial disability pension 148 (34)Neighbourhood-level income quintiles (average income range per single person equivalent)†1 (C$10,950 – C$29,001) 183 (45)2 (C$29,037 – C$33,944) 79 (19)3 (C$33,966 – C$38,834) 38 (9)4 (C$38,881 – C$45,760) 45 (11)5 (C$46,045 – C$122,256) 65 (16)Missing 24Male 189 (44)Mean Age, +/- SD 45.1, +/-12.6Current smoker 205 (47)Documented substance abuse issues 111 (26)Pneumonia Severity IndexMean, +/- SD 70.6, +/- 34.0Median, min – max 64.0, 12 – 219Documented ADL problem 19 (4)Number of co-morbidities0 307 (71)1 112 (26)2 14 (3)3 1 (0.2)Living alone 107 (25)§ Adult pneumonia admissions <65 years to Vancouver General Hospital, January 1, 1990 to March 31, 2001‡ Annual benefits in British Columbia, 1996 [21]. These two categories of BC benefits are mutually exclusive; an individual cannot receive both at the same time. At ages 65 and over, individuals are no longer eligible for income assistance or provincial disability assistance.† Range of 1996 census enumeration area average income per single person equivalent for Vancouver Census Metropolitan Area [18]SD = standard deviation; min = minimum value; max = maximum value; ADL = activities of daily livingPage 4 of 10(page number not for citation purposes)failure, hepatic disease, congestive heart failure and cere-brovascular disease.measured co-variates. Variables that were not significantat p < 0.05 in the model were dropped. We exponentiatedBMC Health Services Research 2006, 6:152 http://www.biomedcentral.com/1472-6963/6/152the parameters of the model to generate the estimatedadjusted multiplicative effects and 95% confidence inter-vals for each SES variable on LOS.SES and re-admissionLogistic regression models were used to examine theadjusted effect of each of our two SES measures on re-admission within 30 days. Variables significant at p < 0.10in univariate regression analysis were initially included inthe multiple regression model and then omitted from thefull model if they were not significant at p < 0.05.Finally, all models were tested for co-linearity and two-factor interaction effects. Co-linearity was tested by ana-lyzing the correlation matrix, dropping highly correlatedvariables (correlation coefficient ≥ 0.8), re-analyzing withdifferent combinations and multiple models until stableresults were obtained.ResultsCharacteristics of the study populationA total of 148 patients (34% of the population understudy) had individual-level financial hardship, with 84(19%) patients on income assistance and 64 (15%) col-lecting a provincial disability pension (Table 1). Forty-fivepercent of individuals with available data (183/410), hadpostal codes corresponding to the lowest neighbourhood-level income quintile and the median income quintilewas the second income quintile, representing an esti-mated range of C$29,037 – C$33,944 (Table 1). Table 2describes the proportion of those identified with financialhardship by neighbourhood-level income quintiles.A higher frequency of male admissions (63% vs. 53%, p =0.05), living alone (41% vs. 16%, p < 0.001) and identi-fied problems in activities of daily living (7% vs. 3%, p <0.05) was seen among those with individual-level finan-cial hardship (Table 3). This group also had a significantlyhigher mean PSI score on admission (75.5 vs. 68.1, p <0.05). In contrast, there were no significant differences inthe distribution of the above variables across the lowestincome quintile grouping (quintile 1) compared to quin-tiles 2 to 5 (Table 3).SES and length of stay (LOS)Those with individual-level financial hardship had alonger median length of stay of 6 versus 4 days comparedto those without – a difference that was significant in uni-variate testing (p < 0.01) (Table 4). After adjustment forPSI, number of co-morbidities, any problem with activi-ties of daily living, living alone and year of admission,financial hardship was associated with an estimated 15%(95% CI -0.4, +32, p = 0.057) longer LOS (Table 5). Sex,smoking and substance abuse were not significant in themultiple regression analysis and therefore not included inthe final model.There was no significant association of neighbourhood-level income quintiles with LOS in univariate (Table 4) ormultiple regression analysis (Table 5).SES and hospital re-admission within 30 daysThree hundred and fifty one records (81%) were success-fully linked to secondary hospital discharge data. Linkageof reviewed cases admitted in the last three years of thestudy time period was hindered due to technical difficul-ties with the linkage process for those years. Among thosethat were linked, there were 43 (12%) all-cause hospitalre-admissions within 30 days of discharge from the indexadmission. Twelve re-admissions (3%) occurred withinthe first 10 days and 22 (6%) occurred within the first 20days. Apart from year of admission, there were no differ-ences between linked and unlinked cases with the excep-tion of a higher proportion of individuals in the first(lowest) income quintile among the unlinked (16/83,53%, n = 83) compared to linked cases (139/351, 40%, n= 351).Among those re-admitted, just over one half (53%, n =23) had individual-level financial hardship compared to30% (n = 91) of those not re-admitted. In the multipleregression analysis, financial hardship was associatedwith re-admission (adjusted OR 2.65, 95% CI 1.38, 5.09)(Table 6). Male sex was also associated with early readmis-sion (adjusted OR 2.05, 95% CI 1.01, 4.18) in the multi-ple regression model.There was no significant association of neighbourhood-level income quintiles with re-admission in either univar-iate or multiple regression analysis (Table 6). When we re-ran the models using income deciles vs. quintiles, resultswere the same. When we re-ran the models excludingcases with imputed data, there was also very little differ-Table 2: Distribution of individual-level financial hardship by neighbourhood income quintiles among study sample (n = 434) §Neighbourhood-level income quintileIndividual-level financial hardship n (%)1 78 (60)2 16 (12)3 15 (12)4 12 (9)5 8 (6)Missing 19Page 5 of 10(page number not for citation purposes)ence in the magnitude of estimated effect for either out-§ Adult pneumonia admissions <65 years to Vancouver General Hospital, January 1, 1990 to March 31, 2001BMC Health Services Research 2006, 6:152 http://www.biomedcentral.com/1472-6963/6/152come. There were no significant interaction effects for anyof the models.DiscussionOur study found that approximately one in three admis-sions were on social assistance or collecting disability pen-sion. This is substantially higher than the reported averageof 7.77% of the BC population collecting social assistanceor disability pension between 1995 and 2000 [26]. Wealso found a disproportionately high number of pneumo-nia admissions had income quintiles in the bottom 20%of incomes in Vancouver. Both findings are consistentwith the literature in a number of Western countries, thathospitalized pneumonia [27] and general medical admis-sions [5-7] tend to be poorer than the general population.Individual-level financial hardship and hospital utilizationOur study found a significant difference in median, unad-justed LOS among pneumonia patients less than 65 yearsold with individual-level financial hardship (6 vs. 4 days,p < 0.01). This association had marginal significance afteradjustment for case mix, functional impairment, and liv-Table 4: Length of stay by socioeconomic status measure among study sample (n = 434) §SES measure Mean LOS, +/- SD; Median LOS, min – max; nIndividual-level financial hardshipYes 7.4, +/- 6.7; 6.0, 1 – 55; n = 148No 5.4, +/- 4.3; 4.0, 1 – 36; n = 286p-value <0.01Neighbourhood-level income quintiles1 6.2, +/- 5.2; 5.0, 1 – 33; n = 1832 6.1, +/- 4.3; 5.0, 1 – 22; n = 793 5.2, +/- 2.9; 5.0, 1 – 15; n = 384 7.1, +/- 8.5; 5.0, 1 – 55; n = 455 5.7, +/- 5.4; 4.0, 1 – 36; n = 65p-value NSTable 3: Distribution of length of stay predictors by socioeconomic status measure among study sample (n = 434) §Individual-level financial hardshipCharacteristic Income assistance or provincial disability pension(n = 148) n (%)No income assistance or provincial disability pension(n = 286) n (%)p-valueMean Age, +/- SD 43.9, +/- 11.9 45.7, +/- 13.0 NSMale 93 (63) 152 (53) 0.05Mean PSI, +/- SD 75.5, +/- 37.4 68.1, +/- 31.8 <0.05Co-morbid Dx >1 8 (5) 7 (2) NSDocumented ADL problem 11 (7) 8 (3) <0.05Living alone 61 (41) 46 (16) <0.001Neighbourhood-level income quintileQuintile 1 (n = 183)Quintiles 2–5 (n = 227)p-valueMean Age, +/- SD 44.7, +/- 12.9 45.6, +/- 12.6 NSMale 102 (56) 129 (57) NSMean PSI, +/- SD 73.9, +/- 35.8 68.1, +/- 32.4 NSCo-morbid Dx > 1 7 (4) 8 (4) NSDocumented ADL problem 10 (5) 8 (4) NSLiving alone 40 (22) 53 (23) NS§ Adult pneumonia admissions <65 years to Vancouver General Hospital, January 1, 1990 to March 31, 2001SD = standard deviation; NS = not significant; PSI = pneumonia severity index;Co-morbid Dx = co-morbidity diagnoses; ADL = activities of daily livingPage 6 of 10(page number not for citation purposes)§ Adult pneumonia admissions <65 years to Vancouver General Hospital, January 1, 1990 to March 31, 2001BMC Health Services Research 2006, 6:152 http://www.biomedcentral.com/1472-6963/6/152ing alone, and ultimately only a small amount of the var-iation in outcome was explained by financial hardship (Rsquared change from 23.9% to 24.5% after inclusion inthe model). This suggests that once admitted to hospitalwith pneumonia, there is a very minor effect of financialhardship on hospital stay per se. Rather, the impact of SESis confounded by a range of other characteristics (illnessseverity, disease co-morbidity, functional status and livingalone) that both lengthen LOS and are disproportionatelypresent among those with financial hardship. This findingis quite different from those of Stelianides and colleagueswho prospectively examined 107 patients hospitalized forpneumonia in France [27]. These authors found anadjusted 5.7 days longer LOS associated with low SES.However the mean LOS of all patients reported in thisstudy was substantially higher (15 days) than in our studyand there was no adjustment for factors beyond clinicalcase mix.Our finding that those with individual-level financialhardship had a greater than 2.5-fold adjusted odds of re-admission within 30 days of discharge suggests a numberof scenarios. Patients may experience destabilization oftheir illness following discharge due to challenging socialcircumstances such as poor housing and inadequate nutri-tion. Also, these individuals may be less likely to adhereto post-discharge treatment plans because of a decreasedability to access needed ambulatory care. Alternatively,patients with financial hardship may be more vulnerableto experiencing a new illness that is not directly related totheir initial hospitalization with pneumonia.From a health policy perspective, and regardless of themechanism, the study results suggest that there may beinsufficient provision of post-discharge services after ahospitalization for pneumonia. One might argue thatthere is a false economy to discharging patients who aremedically stable but "socially precarious" and that failureTable 6: Adjusted odds ratios (95% CI) for re-admission associated with SES measure among study sample (n=351) §SES measure Adjusted OR (95% CI) p-valueIndividual-level financial hardship (n=351) ‡Yes 2.65 (1.38, 5.09) <0.01No (reference)Neighbourhood-level income quintiles (n=335)†1 0.75 (0.29, 1.92) NS2 1.03 (0.39, 2.74) NS3 0.52 (0.13, 2.15) NS4 1.02 (0.29, 3.54) NS5 (reference)§ Adult pneumonia admissions <65 years to Vancouver General Hospital, January 1, 1990 to March 31, 2001 readmitted to hospital within 30 days Table 5: Adjusted percent differences (95% CI) in LOS by SES measure among study sample (n = 434) §SES measure Adjusted % difference in LOS (95% CI)p-value R2 SES measure R2 modelIndividual-level financial hardship (n = 434) ‡Yes 15 (-0.4, +32) 0.057 0.028 0.245No (reference)Neighbourhood-level income quintiles (n = 410)†1 -11 (-27, +8) NS 0.004 0.2482 6 (-16, +33) NS3 4 (-21, +37) NS4 13 (-13, 46) NS5 (reference)§ Adult pneumonia admissions <65 years to Vancouver General Hospital, January 1, 1990 to March 31, 2001‡ Model adjusted for PSI, 15% (+11,+20); # co-morbidities, 24% (+10,+40); documented ADL problem, 85% (+34,+154); living alone, 20% (+3,+41); year of admission, -5% (-7,-3)† Model adjusted for PSI, 15% (+11,+20); # co-morbidities, 24% (+9,+40); documented ADL problem, 96% (+42,+171); living alone, 24% (+5,+45); year of admission, -5% (-7,-3) SES = socioeconomic status; LOS = length of stay; 95% CI = 95% confidence interval; NS = not significant; PSI = pneumonia severity index; ADL = activities of daily livingPage 7 of 10(page number not for citation purposes)of discharge‡ Model adjusted for male sex, OR = 2.05 (1.01, 4.18)† Model adjusted for male sex, OR = 2.15 (1.05, 4.40); # co-morbidities, OR = 1.72 (1.06, 2.79)BMC Health Services Research 2006, 6:152 http://www.biomedcentral.com/1472-6963/6/152to address the latter context results in early post-dischargemedical destabilization with costly re-admission. Thepiloting of programs to provide "sub-acute" medical andsocial support, or systems that follow low income patientsin the community after discharge to address potentialcomplications, may be useful in decreasing rates of re-admission among this population. Prospective research,to better understand the causes for re-admission amongthose with financial hardship, and to pilot re-admissionprevention programs among this population, would alsobe an important line of further research.Ecological SES (income quintiles) and hospital utilizationIn contrast to the link between individual-level financialhardship and hospital LOS and re-admission, we foundthat the neighbourhood-level income quintile variablewas remarkably inert. We suggest that this may be due toa lack of precision of this ecological variable in identifyingindividuals experiencing a more extreme degree of finan-cial hardship. Seventy-eight of the one hundred andtwenty-nine (60%) individuals with financial hardshipalso had postal codes corresponding to the lowest incomequintile, and the association was significant (Pearson Chi-square = 19.09, p < 0.001). On the other hand, a substan-tial number of those with financial hardship (n = 51,40%) also had postal codes that fell into neighbourhood-level income quintiles outside of the poorest quintile.Other possible explanations for the inconsistency inresults between individual and ecological measures of SESinclude the following. Firstly, the low income range repre-sented by postal codes in the bottom quintile was sub-stantially higher than the income represented by thosewith individual-level financial hardship (C$10,950 –29,001 vs. C$7,081). Secondly, the group with individual-level financial hardship had a disproportionate number ofmissing postal codes, compared to the rest of the popula-tion under 65 years (13% vs. 6%). This may reflect agreater proportion of individuals with financial hardshipand unstable housing for which a residential postal codecould not be assigned. Finally, while postal code of resi-dence may be good at identifying neighbourhood charac-teristics such as access to transit and green space, onewould expect pneumonia to be more related to individ-ual-level characteristics like housing conditions, nutritionand poverty.These differences of precision and type of measure may beless important for studies with large sample sizes. Roos etal. examined administrative data for the urban populationof Manitoba (N = 794,555) and demonstrated a signifi-cant inverse socioeconomic gradient for pneumonia hos-pitalization rates using the same postal code derivedThis finding has important implications for health serv-ices researchers who are often restricted, by data availabil-ity, to the use of ecological measures when adjusting forSES. While the imprecision of neighbourhood-level SESmeasures resulting in random misclassification of obser-vations is unlikely to attenuate SES effects in large popu-lation studies [16,29], our research suggests that theyshould be used with caution in smaller clinical studies.Beyond the problem of misclassification, it should benoted that ecological measures presuppose the presenceof a fixed address – something that is often missingamong the most marginalized patients, whose social cir-cumstances are also most likely to affect the health out-comes examined. Exclusion of marginalized groupsthrough measurement of SES with ecological measuresalone may thus lead to an underestimation and/or under-adjustment of SES effects.Distribution of other patient characteristics by socioeconomic statusIndividuals with financial hardship were more likely to bemale, to live on their own, to have a disability and topresent with a higher degree of illness severity as meas-ured by the PSI. The independent association of all thesevariables with LOS confirms the importance of measuringthese effects as potential confounders of socioeconomicstatus on hospital utilization.Study strengths and weaknessesThis study was limited by potential misclassification andunintended bias introduced by the retrospective nature ofthe data collection. It is possible for example, that therewere those on social assistance or disability pensions, whofor some reason, did not report this on admission or inthe course of their clinical stay. If this was the case andthese individuals had a disproportionate frequency ofshorter lengths of stay or re-admissions, then the resultsmay be confounded in a direction away from the null.Another weakness is that we did not explore the reasonsfor individuals having missing postal code data at the timeof chart review. This would be important to understandwhether the missing data for this variable confounded therelation of neighbourhood income quintiles with our out-comes. However, our use of chart review data allowed usto employ both a more precise individual-level and anecological SES measure, and compare the two. It alsoallowed us to use a richer set of clinical data to constructa disease-specific illness severity measure (such as PSI)and to examine other important potential confoundingvariables. The linkage of these chart review data with sec-ondary provincial health data contributes to a more com-plete understanding of the relationship between SES andPage 8 of 10(page number not for citation purposes)neighbourhood-level income measure used in our study[28].hospital utilization.BMC Health Services Research 2006, 6:152 http://www.biomedcentral.com/1472-6963/6/152ConclusionIndividuals with financial hardship have a longer hospitalLOS and are more likely to experience re-admission tohospital, perhaps as a result of their social circumstances.However, there is no demonstrated income gradient asso-ciated with hospital LOS among patients admitted forpneumonia when income is measured as an ecologicalvariable. The ecological income quintile variable, whileuseful in many circumstances, may be insufficiently sensi-tive to pick up SES effects in smaller clinical studies.Competing interestsThe author(s) declare that they have no competing inter-ests.Authors' contributionsMJM contributed to the research design, coordinated theimplementation of the research and played a major role inthe manuscript writing. RJR and ARL assisted with thestudy design and implementation. RJR, ARL and JMF allcontributed to the data interpretation. MS supervised thedata analysis and MBC performed the data analysis. Allauthors contributed to the manuscript writing and gavefinal approval of the version to be published.AcknowledgementsThe study was funded by the Vancouver Foundation FY 01–02 and an addi-tional seed grant from CIHR FY02-03. We gratefully acknowledge the assistance of Karen Cardiff and Eva Somogyi who performed the chart review, Susan Sirrett and Diane Lofthouse who assisted in the data extrac-tion from the VGH/QUIST database, David Jung who assisted in the design and implementation of the data coding sheet, Edwin Mak who performed the PSI imputation analysis, Nino Pagliccia who performed the data linkage of chart review data with the BC Linked Health Database, Kim McGrail from the Centre for Health Services and Policy Research who assisted with postal code data linkage to SES quitiles and deciles and assisted with critical review of the manuscript, Clyde Hertzman, and Sam Sheps from the Department of Healthcare and Epidemiology, University of British Colum-bia, who assisted with critical review of the manuscript.References1. Wilkins R, Berthelot JM, Ng E: Trends in mortality by neighbour-hood income in urban Canada from 1971 to 1996.  HealthReports 2002, Suppl 13:45-71.2. Mustard CA, Derksen S, Berthelot JM, Wolfson M, Roos LL: Age-specific education and income gradients in morbidity andmortality in a Canadian province.  Soc Sci Med 1997, 45:383-397.3. Marmot MG, Rose G, Shipley M, Hamilton PJ: Employment gradeand coronary heart disease in British civil servants.  J EpidemiolCommunity Health 1978, 32:244-249.4. Meddings DR, Hertzman C, Barer ML, Evans RG, Kazanjian A,McGrail K, Sheps SB: Socioeconomic status, mortality, and thedevelopment of cataract at a young age.  Soc Sci Med 1998,46:1451-1457.5. Lin S, Fitzgerald E, Hwang SA, Munsie JP, Stark A: Asthma hospital-ization rates and socioeconomic status in New York State(1987-1993).  J Asthma 1999, 36:239-251.6. Pappas G, Hadden WC, Kozak LJ, Fisher GF: Potentially avoidablehospitalizations: inequalities in rates between US socioeco-nomic groups.  Am J Public Health 1997, 87:811-816.8. Gordon NH, Crowe JP, Brumberg DJ, Berger NA: Socioeconomicfactors and race in breast cancer recurrence and survival.  AmJ Epidemiol 1992, 135:609-618.9. Callahan LF, Pincus T: Formal education level as a significantmarker of clinical status in rheumatoid arthritis.  ArthritisRheum 1988, 31:1346-1357.10. Bobak M, Hertzman C, Skodova Z, Marmot M: Socioeconomic sta-tus and cardiovascular risk factors in the Czech Republic.  IntJ Epidemiol 1999, 28:46-52.11. Epstein AM, Stern RS, Weissman JS: Do the poor cost more? Amultihospital study of patients' socioeconomic status anduse of hospital resources.  N Engl J Med 1990, 322:1122-1128.12. Weingarten JP, Clay JC, Heckert DA: Impact of socioeconomicstatus on health care utilization: factors influencing length ofstay.  J Health Hum Serv Adm 1997, 19:384-409.13. Buckle JM, Horn SD, Oates VM, Abbey H: Severity of illness andresource use differences among white and black hospitalizedelderly.  Arch Intern Med 1992, 152:1596-1603.14. Brownell MD, Roos NP: Variation in length of stay as a measureof efficiency in Manitoba hospitals.  CMAJ 1995, 152:675-682.15. Glazier RH, Badley EM, Gilbert JE, Rothman L: The nature ofincreased hospital use in poor neighbourhoods: findings froma Canadian inner city.  Can J Public Health 2000, 91:268-273.16. Hofer TP, Wolfe RA, Tedeschi PJ, McMahon LF, Griffith JR: Use ofcommunity versus individual socioeconomic data in predict-ing variation in hospital use.  Health Serv Res 1998, 33:243-259.17. Garibaldi RA: Epidemiology of community-acquired respira-tory tract infections in adults. Incidence, etiology, andimpact.  Am J Med 1985, 78:32-37.18. Wilkins R: PCCF+ version 4D user's guide. Automated geo-graphic coding based on the Statistics Canada postal codeconversion files, including postal codes to December2003.[82-F0086-XDB].  Ottawa, Health Analysis and Measure-ment Group, Statistics Canada; 2004. 19. Wilkins R: Use of postal codes and addresses in the analysis ofhealth data.  Health Reports 1993, 5:157-177.20. Chamberlayne R, Green B, Barer ML, Hertzman C, Lawrence WJ,Sheps SB: Creating a population-based linked health database:a new resource for health services research.  Can J Public Health1998, 89(4):270-3.21. Minister of Public Works and Government Services Canada 2004:National Council of Welfare Reports, Welfare Incomes2003.    [http://www.ncwcnbes.net/htmdocument/reportWelfareIncomes2003/WI2003_e.pdf].22. Fine MJ, Auble TE, Yealy DM, Hanusa BH, Weissfeld LA, Singer DE,Coley CM, Marrie TJ, Kapoor WN: A prediction rule to identifylow-risk patients with community-acquired pneumonia.  NEngl J Med 1997, 336:243-250.23. Feagan BG, Marrie TJ, Lau CY, Wheeler SL, Wong CJ, VandervoortMK: Treatment and outcomes of community-acquired pneu-monia at Canadian hospitals.  CMAJ 2000, 162:1415-1420.24. Deyo RA, Cherkin DC, Ciol MA: Adapting a clinical comorbidityindex for use with ICD-9-CM administrative databases.  J ClinEpidemiol 1992, 45:613-619.25. Charlson ME, Pompei P, Ales KL, MacKenzie CR: A new method ofclassifying prognostic comorbidity in longitudinal studies:development and validation.  J Chronic Dis 1987, 40:373-383.26. Government of British Columbia, Ministry of Employmentand Income Assistance: BC Employment and AssistanceSummary Report, Issue: 2006-6   [http://www.eia.gov.bc.ca/research]27. Stelianides S, Golmard JL, Carbon C, Fantin B: Influence of socioe-conomic status on features and outcome of community-acquired pneumonia.  Eur J Clin Microbiol Infect Dis 1999,18:704-708.28. Roos LL, Wald R, Uhanova J, Bond R: Physician visits, hospitaliza-tions, and socioeconomic status: ambulatory care sensitiveconditions in a Canadian setting.  Health Serv Res 2005,40:1167-1185.29. Mustard CA, Derksen S, Berthelot JM, Wolfson M: Assessing eco-logic proxies for household income: a comparison of house-hold and neighbourhood level income measures in the studyof population health status.  Health Place 1999, 5:157-171.Page 9 of 10(page number not for citation purposes)7. Billings J, Zeitel L, Lukomnik J, Carey TS, Blank AE, Newman L:Impact of socioeconomic status on hospital use in New YorkCity.  Health Aff 1993, 12:162-173.Publish with BioMed Central   and  every scientist can read your work free of charge"BioMed Central will be the most significant development for disseminating the results of biomedical research in our lifetime."Sir Paul Nurse, Cancer Research UKYour research papers will be:available free of charge to the entire biomedical communitypeer reviewed and published immediately upon acceptancecited in PubMed and archived on PubMed Central BMC Health Services Research 2006, 6:152 http://www.biomedcentral.com/1472-6963/6/152Pre-publication historyThe pre-publication history for this paper can be accessedhere:http://www.biomedcentral.com/1472-6963/6/152/prepubyours — you keep the copyrightSubmit your manuscript here:http://www.biomedcentral.com/info/publishing_adv.aspBioMedcentralPage 10 of 10(page number not for citation purposes)


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items