Association of regional racial/cultural context and socioeconomic status with arthritis in the population: A multilevel analysis




To examine the extent to which differences in individual- and regional-level socioeconomic status and racial/cultural origin account for geographic variations in the prevalence of self-reported arthritis, and to determine whether regional characteristics modify the effect of individual characteristics associated with reporting arthritis.


Analyses were based on the 2000–2001 Canadian Community Health Survey (>15 years, n = 127,513). Arthritis was self-reported as a long-term condition diagnosed by a health professional. A 2-level logistic regression model was used to identify predictors of reporting arthritis. Individual-level variables included age, sex, income, education, immigration status, racial/cultural origin, smoking, physical activity, and body mass index. Regional-level variables included the proportion of low-income families, low education, unemployment, recent immigrants, Aboriginals, and Asians.


At the individual level, age, sex, low income, low education, Aboriginal origin, current smoking, and overweight/obesity were positively associated with reporting arthritis; recent immigration and Asian origin were negatively associated with reporting arthritis. At the regional level, percentages of low-income families and the Aboriginal population were independently associated with reporting arthritis. Regional income and racial/cultural origin moderated the effects of individual income and racial/cultural origin; low-income individuals residing in regions with a higher proportion of low-income families reported arthritis more than low-income individuals living in better-income regions.


Both individual and regional factors were found to contribute to variations in the prevalence of arthritis, although significant unexplained variation remained. Further research is required to better understand the mechanisms that underlie these regional effects and to identify other contributing factors to the remaining variation.


Arthritis and related conditions, hereafter referred to as arthritis, are among the most prevalent chronic conditions in western nations. Arthritis is a leading cause of pain and disability, resulting in substantial economic costs for health care systems. The prevalence of arthritis in the US and Canada has been found to vary geographically (1–4). In 2003, the estimated prevalence of doctor-diagnosed arthritis among adults in the US ranged from 17.9% to 37.2% (4). Similar variations have been found in Canada, where the province of Quebec has consistently had the lowest prevalence rates of arthritis in the country, while the Atlantic provinces have consistently had the highest rates (1, 2). The reasons for these geographic variations are not clear.

These differences in prevalence persist after controlling for individual factors found to be associated with an increased likelihood of reporting arthritis (2). These factors include older age, high body mass index (BMI), smoking, lower socioeconomic status (SES) as reflected by lower educational attainment or lower income, and racial/ethnic background (3, 5–9). Arthritis is more frequent in African Americans (9) and persons of Aboriginal descent (3, 7). In contrast, persons of Asian origin or recent immigrants to western nations report a lower prevalence of arthritis (3, 7, 8, 10).

Whether the variations in arthritis prevalence result from variations in the geographic distribution of important individual characteristics in the population or from differences in the distribution of socioeconomic characteristics at a regional level, or some combination of both, has not been well investigated (11). Studies exploring variations in the distribution of health outcomes, such as self-rated health, health-related behaviors, and mortality, among others, have found that community-level SES (e.g., region, neighborhood) and cultural environments (e.g., residential segregation, acculturation) contribute in part to explaining these variations (12–18). The association of lower SES and poor health outcomes has been found even in more egalitarian societies (12).

Communities in which people live may also influence health. These influences may operate through, for example, the availability and accessibility of health services, infrastructure deprivation (e.g., lack of stores selling healthy foods at affordable prices), the prevalence of prevailing attitudes toward health and health-related behaviors, and a lack of social support (13, 19–21). It has been shown that individuals living in more deprived areas are less likely to seek medical care, independent of their individual SES (22, 23). Canada has long been a multiethnic and multicultural society. Visible minority immigrants, regardless of how long they have lived in Canada, have found themselves living in areas with larger numbers of people from their own ethnic group (24). The composition and spatial distribution of the population has been found to be related to health and health outcomes. In general, it has been hypothesized that residential segregation affects health indirectly through concentration of poverty or the social position of minority group members (25, 26).

Studies on predictors of arthritis in the population have traditionally focused on individual-level characteristics, for example, SES of the family, race/cultural origin, and obesity; however, these individual factors have not completely explained the differences in the prevalence of arthritis between communities. Consequently, understanding how community characteristics are linked to the prevalence of arthritis may be crucial to deciding priorities for targeting interventions.

Few studies using both individual and community characteristics have used arthritis as an outcome (27, 28); hardly any have examined how community contexts interact with individual characteristics. In light of the paucity of reports on the associations of individual and regional characteristics with arthritis, the purpose of the present study was to contribute to the literature by using nationally representative Canadian data. We hypothesized that differences in individual- and regional-level SES and racial/cultural origin would contribute to geographic variations in the prevalence of self-reported arthritis, and that regional characteristics would modify the effect of individual characteristics associated with reporting arthritis.


Data sources.

Individual-level data were obtained from the 2000–2001 Canadian Community Health Survey (Cycle 1.1) conducted by Statistics Canada. The target population included individuals age ≥12 years living in private dwellings in each of the provinces and territories. Individuals living on Indian reservations, institutional residents, members of the Canadian armed forces, and residents of certain remote regions were excluded. The survey used a stratified 2-stage cluster design. In the first stage, separate strata were formed based on economic characteristics of the provinces and then independent clusters were drawn from each stratum. In the second stage, dwelling lists were prepared for each cluster and dwellings were selected from those lists. Generally, 1 person was randomly selected per household. Approximately 98% of the targeted population was covered. The overall response rate was 84.7% with 130,880 participants. Further details are documented elsewhere (29). Data for the population age ≥15 years were used for analyses (n = 127,513).


The individual outcome of interest was the presence or absence of arthritis. This was ascertained by asking respondents if they had arthritis or rheumatism, excluding fibromyalgia, as a long-term condition that had lasted or was expected to last ≥6 months and that was diagnosed by a health professional.

Individual-level variables.

Based on findings from the literature, 4 sets of individual variables were considered: demographic (age, sex), lifestyle characteristics (smoking, level of physical activity, and obesity), socioeconomic (household income, education), and ethnic/racial/cultural background (racial/ethnic origin and immigration status). Smoking status was categorized as current smoker (daily/occasional), former smoker (former daily/former occasional), and never smoked. Physical activity was characterized as active, moderate, or inactive based on Statistics Canada's index of physical activity (29). BMI (weight [kg]/height [m2]) was calculated from self-reported weight and height, except for pregnant women and individuals <0.914 meters or >2.108 meters in height (1.4% of the sample). BMI was collapsed into 4 categories: underweight (<18.5 kg/m2), normal weight (18.5–24.9 kg/m2), overweight (25–29.9 kg/m2), and obese (≥30 kg/m2) (30).

Individuals were asked to give an estimate of the total household income from all sources (e.g., wages and salaries, bonds, savings, employment insurance, pension, etc.) in the past 12 months (figures in Canadian dollars). Originally consisting of 11 categories, a 4-category variable was created: low (<$20,000), middle-low ($20,000–$39,999), middle-high ($40,000–$59,999), and high income (≥$60,000). Highest level of education was trichotomized into less than secondary school, secondary school, and postsecondary or higher.

Racial/ethnic origin was determined by responses to the leading statement, “People in Canada come from different cultural and racial backgrounds….” A list of 13 groups was presented and respondents were asked to indicate all that applied. Four mutually exclusive groups were created: only white, only Asian (including Chinese, South Asian, West Asian, East Asian), only Aboriginal (North American Indian, Métis, Inuit/Eskimo), and other (included all other and those of multicultural origin). We were particularly interested in exploring Aboriginal and Asian populations because the former often report an increased arthritis prevalence, whereas the latter report a lower prevalence (3, 7, 8, 10). The Asian population is the largest nonwhite group in Canada. Only a small proportion of the population identified themselves as black or Latin American (1.7% and 0.7%, respectively). Immigration status was defined as the length of time in Canada since immigration (nonimmigrants, <10 years since immigration, 10–25 years, and >25 years).

Regional-level variables.

Regional-level information was obtained from census data. Canada is subdivided into 288 regions based on census divisions (31). Census divisions are defined as a group of neighboring municipalities joined together for the purposes of regional planning and managing common services (e.g., police services). A census division might correspond to a county, a regional municipality, or a regional district.

Data from the 2001 Canada Census were used to determine regional-level SES and racial/cultural composition. SES indicators were income characterized as the percentage of low-income families in the region and education characterized as the percentage of individuals with less than grade 9 schooling in the region. The racial/cultural composition of a region was characterized as the proportion of the population identified as Asian and Aboriginal and the proportion of recent immigrants (living in Canada ≤10 years).

Multilevel modeling.

Two-level logistic models (32) were estimated to simultaneously investigate the influence of individual- and regional-level characteristics on the likelihood of reporting arthritis. This way, the nonindependence of individuals living in the same region was accounted for. Due to sample size restrictions, provinces could not be added to the models as a third level. However, province-specific effects were added to the models as fixed effects.

Model 1 (Figure 1) allowed us to examine whether the odds of reporting arthritis varied across regions and to estimate the magnitude of this variation, controlling for age, sex, and province. Model 1 was then expanded to adjust for additional individual characteristics (model 2 = model 1 plus individual SES, racial/cultural origin, and lifestyle characteristics). Model 2 was then further expanded to examine whether regional characteristics explained regional variations, net of individual characteristics (model 3 = model 2 plus regional characteristics). Finally, to determine whether regional characteristics moderated the relationship between individual SES and racial/cultural origin and reporting arthritis, interaction terms between individual SES and racial/cultural background and between regional SES and racial/cultural background were added (model 4 = model 3 plus interaction terms). To select relevant variables for model 4, individual SES and racial/cultural effects were allowed to vary across regions. Where significant variation was found, interaction terms were introduced for further investigation. A detailed description of the models is presented in Figure 1.

Figure 1.

Description of the multilevel models examined, where pij represents the probability of participant i in region j reporting arthritis, γ00 is the average log odds of reporting arthritis across all regions, and u0j is the difference in log odds between region j and the average (γ00). The parameters β1j to β10j represent the effect of individual predictors on the likelihood of reporting arthritis after taking into account the variability in arthritis prevalence across regions, and the coefficients γ01 to γ0q quantify the impact of regional variables on the likelihood of reporting arthritis after taking into account individual characteristics. SES = socioeconomic status.

The associations between the prevalence of arthritis and the predictors are referred to as the fixed components, and the variance in the prevalence of arthritis across regions as the random component. Individual odds ratios (ORs) and 95% confidence intervals were obtained from the fixed components.

Median ORs were calculated as a measure of heterogeneity across regions. This measure can be viewed as the median of the ORs that would be obtained by choosing 2 persons with the same covariates but from different regions and computing the OR between them, with data for the person from the region with a higher probability in the numerator. The process is iterative. A median OR >1 indicates heterogeneity between regions; the greater the value, the greater the extent of heterogeneity. This value is directly comparable with a fixed-effects OR. For example, a median OR of 2 compared with an OR of 1.5 for a regional-level variable indicates that the remaining between-regions heterogeneity is of a higher magnitude than the effect of the regional-level variable (33, 34).

All models were fitted using restricted maximum likelihood estimation in HLM version 6 (Scientific Software International, Lincolnwood, IL). Various sensitivity analyses were conducted. Analyses were repeated excluding those participants who reported multiple racial backgrounds to determine potential bias due to cross-classification across different racial/cultural groups. In addition, because income was not reported by 11% of individuals, data were reanalyzed including these individuals as a “not stated” income category.


A total of 127,513 individuals age ≥15 years were linked to data from 286 regions across Canada. The overall prevalence of arthritis was 16.0% and varied considerably across regions. The range of arthritis prevalence and individual characteristics across regions are shown in Table 1.

Table 1. Description of individual outcome, socioeconomic, and ethnicity variables used to assess differences in the prevalence of arthritis in Canada*
Individual variablesEstimated numberOverall prevalence, %Prevalence by regions, range
  • *

    Weighted to compensate for unequal sampling probabilities.

 With arthritis3,929,07416.00.0–32.2
 Without arthritis20,694,24784.017.8–100
Age, years   
 Less than secondary6,400,34926.214.0–72.0
 Postsecondary or higher11,144,76745.613.6–64.4
 <10 years1,405,3955.80–18.6
 10–25 years1,513,5916.20–17.9
 >25 years2,244,5859.10–19.6
Racial/cultural origin   

Results from analyses of model 1 indicated that a significant variation in arthritis prevalence existed across regions (χ2 = 1023.15, P < 0.001), controlling for age, sex, and province, with an estimated median OR of 1.38 (model 1 estimates not shown).

Significant variation across regions remained after controlling for the remaining individual-level variables (i.e., lifestyle factors, SES, and racial/cultural origin) with an estimated median OR of 1.22 (Table 2, model 2). Increasing age, female sex, and being overweight or obese were each significantly and positively associated with reporting arthritis. A strong association was also found for the effect of smoking status. There was a strong individual income gradient for reporting arthritis, and a gradient of effect was observed for level of education. Recent immigrants (<10 years) were 47% less likely to report arthritis than nonimmigrants. However, immigrants who had lived in Canada in excess of 10 years were just as likely to report arthritis as nonimmigrants. Aboriginals were 40% more likely to report arthritis than white individuals, whereas Asians were 28% less likely.

Table 2. Results from multilevel models*
Fixed effectsModel 2 (individual)Model 3 (individual + regional)χ2PMedian OR
OR95% CIOR95% CI
  • *

    All models include provinces in the fixed part at the individual level. OR = odds ratio; 95% CI = 95% confidence interval; BMI = body mass index.

  • Increments in 10 units.

 Age (increments in 10 years)1.801.65–2.101.741.64–1.93   
 Sex (reference: male)       
 Smoking (reference: never smoker)       
  Current smoker1.501.40–1.601.521.46–1.60   
  Former smoker1.171.11––1.29   
 Physical activity (reference: inactive)       
 BMI (reference: 18.5–24.9), kg/m2       
 Income (reference: ≥$60,000)       
 Education (reference: postsecondary or higher)       
  Less than secondary1.091.01––1.08   
 Immigration (reference: nonimmigrants)       
  <10 years0.530.41–0.680.510.43–0.62   
  10–25 years0.860.70–1.050.860.76–0.97   
  >25 years1.050.96––1.09   
 Racial/cultural origin (reference: white)       
 Percentage of low-income families  1.301.03–1.64   
 Percentage of low education  0.970.94–1.01   
 Percentage of Aboriginal population  1.121.04–1.19   
 Percentage of Asian population  1.021.00–1.04   
Random effects       
 Model 2    602.43<0.0011.22
 Model 3    580.99<0.0011.14

The effect of the inclusion of regional characteristics is shown in Table 2, model 3. Inclusion of the regional variables had a minimal effect on the contribution of individual-level variables to the likelihood of reporting arthritis. Of the regional predictors examined, only regional income and proportion of Aboriginals were significantly associated with arthritis. Lower regional income was significantly associated with an increased likelihood of reporting arthritis, independent of an individual's income, with a 30% greater likelihood with every 10 percentage-point increase in the proportion of low-income families. In addition, for residents of regions with a higher proportion of Aboriginals, the likelihood of reporting arthritis increased by 12% with every 10 percentage-point increase, independent of an individual's racial/cultural origin.

In evaluating whether the average effect of individual SES and racial/cultural origin on the likelihood of reporting arthritis varied across regions, only individual income and racial/cultural origin showed significant regional variation (model estimates not shown). This result justified an exploration of cross-level interactions for these effects. Significant cross-level effects were found between regional income and individual income and racial/cultural origin and between regional racial/cultural background and individual racial/cultural origin. Table 3 presents the ORs of regional income and regional racial/cultural background across categories of individual-level income and racial/cultural origin for the significant associations only. Inclusion of the cross-level interaction terms had minimal effects on the contribution of individual-level and regional-level variables to the likelihood of reporting arthritis. Although somewhat reduced, there remained significant regional variation (estimated median OR 1.13).

Table 3. Associations of regional and individual variables in multilevel models*
 Percentage of low-income families (increments in 10 units)Percentage of Aboriginal population (increments in 10 units)
  • *

    Values are the odds ratios of reporting arthritis (95% confidence interval). Odds ratios were obtained from a model with individual and regional predictors.

  • P < 0.05.

  • P < 0.01.

Individual racial/cultural  origin  
 Aboriginal1.03 (1.01–1.05)0.96 (0.94–0.98)
 Asian0.95 (0.92–0.98)1.04 (0.98–1.10)
 Other1.15 (1.12–1.18)1.03 (0.97–1.06)
 White1.07 (1.05–1.09)1.03 (0.98–1.05)
Individual income  
 <$20,0001.09 (1.02–1.16) 
 $20,000–$39,9991.06 (0.91–1.21) 
 $40,000–$59,9991.06 (0.95–1.17) 
 ≥$60,0001.06 (0.93–1.19) 

An increasing proportion of low-income families in a region increased the likelihood of reporting arthritis among individuals with low income; this was not the case for individuals with higher income. Lower regional income was also associated with an increased likelihood of reporting arthritis by Aboriginals and whites. A negative relationship between prevalence of arthritis and regional income was observed for persons of Asian origin.

The proportion of Aboriginals in a region was an important factor only for Aboriginal individuals. Aboriginals living in areas with a high proportion of Aboriginals were less likely to report arthritis than Aboriginals living in areas with a lower proportion of Aboriginals.

In replicated analyses, which included individuals in the “not stated” income category, minimal changes in results were observed. Similarly, replicated analyses excluding individuals with multicultural backgrounds did not alter results.


This study examines the distribution of arthritis in a population taking into account regional-level risk factors as well as individual characteristics and demonstrates that regional variation in the prevalence of arthritis is in part explained by the socioeconomic, racial/cultural origin, and lifestyle characteristics of individuals, as well as the socioeconomic and racial/cultural composition of regional populations. This is at least the case in Canada, but we suspect that similar results would be found for other jurisdictions. A novel finding is an interaction between individual- and regional-level SES and racial/cultural origin. Living in the most economically disadvantaged regions was associated with higher odds of reporting arthritis, even after controlling for a range of individual-level variables including income and education. In addition, living in regions with a greater proportion of Aboriginals was associated with a decreased likelihood of reporting arthritis for those of Aboriginal origin.

At the individual level, our findings are in line with other studies of predictors of arthritis in the population. As expected, the odds of reporting arthritis increased with increasing age and were higher in females. A socioeconomic gradient was also seen, with individuals of lower income and those with less than secondary education being more likely to report arthritis. The association between reporting arthritis and individual SES is well established (3, 5, 7). The higher prevalence of arthritis among obese individuals is also well known (5), and reported findings have suggested that obesity has contributed to more cases of arthritis in recent years than in previous decades (6). Reports in the literature have been inconsistent, however, for the relationship between smoking and arthritis (35–38). For the most part, studies have focused on rheumatoid arthritis (39–42). It has been suggested that exposure to tobacco smoke may trigger the production of rheumatoid factor and, subsequently, contribute to the development of rheumatoid arthritis (41). In persons with arthritis, heavy smoking is also associated with the severity of rheumatoid arthritis (40, 41). We found smoking to be positively associated with the reporting of arthritis.

Our finding of an increased risk of reporting arthritis in Aboriginals and a decreased risk in Asians is in accordance with other studies (7, 8, 10). That recent immigrants are less likely to report arthritis than nonimmigrants is also consistent with results from the literature (43, 44). For example, Newbold and Danforth (44) reported a narrowing of the health gap in Canada between individuals who are native born and immigrants as their years in Canada increase—a worsening of immigrant health over time. Some researchers hypothesize that convergence in health outcomes might arise from a process of acculturation, in which recent immigrants gradually take on the characteristics of their “new” society. McDonald and Kennedy (45, 46) found that in Canada, the relative incidence of most chronic conditions among immigrants increases with every year since immigration.

In addition to individual-level effects on the reporting of arthritis, we found significant independent and additive effects of regional-level variables. There was a higher risk of reporting arthritis for individuals living in regions with lower incomes and with higher percentages of Aboriginal people. Inclusion of these regional variables in our statistical models improved model fit; however, the median OR did not reach 1.0, suggesting that there remains variation in the reporting of arthritis even after accounting for these variables.

We found significant interactions between regional-level income and proportion of Aboriginals in the population and between individual-level income and racial/cultural origin variables. The effects of living in socioeconomically deprived regions on the likelihood of reporting arthritis appear to be larger for individuals of low SES. Individuals with low income who reside in regions with low SES reported arthritis more often than those with low income living in regions with better SES. These findings suggest a need for the development of arthritis prevention and management strategies that recognize both individual- and regional-level risk factors.

Regional SES also appears to independently modify the effect of racial/cultural origin. We found that Aboriginals residing in socioeconomically deprived regions reported arthritis more often than those residing in socioeconomically better regions. An unexpected finding, in light of conventional discourse on SES and health, is that regional income was negatively associated with arthritis prevalence for persons of Asian origin. This may be explained by a healthy immigrant effect. Recent immigrants tend to settle in regions with lower SES and a significant proportion of recent immigrants to Canada have been Asian (44, 47).

Interestingly, we found that for individuals of Aboriginal origin, arthritis prevalence decreases with the proportion of Aboriginals in the region. Nonetheless, in no region does the model predict lower arthritis prevalence for Aboriginals than the other racial/cultural origin groups studied. This suggests that Aboriginals living in predominantly non-Aboriginal regions may be particularly vulnerable. Roberts (48) found that while being black increased the probability of having a low-birthweight child, black women living in predominantly white neighborhoods were more likely to have a low-birthweight infant than those living in predominantly black neighborhoods. Neeleman et al (49) suggested the extent to which exposure to risk factors experience in the surroundings modifies the effects of those risk factors on adverse health outcomes. Therefore, factors that may be associated with being atypical in the social environment modify the risk of adverse outcomes. We note that our survey data include off-reservation Aboriginals only. However, the vast majority (78%) of Aboriginals in Canada live off reservations (50).

The cross-sectional nature of the data limits our ability to draw causal inferences, particularly for the association between income and arthritis, which may be bidirectional. For the purposes of our study, regions were administratively defined. The main drawback of this is that these boundaries may not match residents' perceptions and lived experiences of their “regional” boundaries. However, the advantage of administrative boundaries is that they are often used for routine data collection, making the data easily accessible for research.

Understanding regional differences in arthritis can provide valuable information for planners who determine the allocation and provision of medical services. It may also aid in the identification of regions that may need to be particularly targeted with prevention and/or management programs.

Using an explicit multilevel analytic framework, we have demonstrated that both individual and regional factors contribute to observed variations in the prevalence of arthritis, although significant unexplained variation remains. Our results are specific to the Canadian population; however, the findings are likely to be relevant to other developed nations, which share similar demographic and health characteristic distributions. Our findings support the recognition of individual demographic and socioeconomic circumstances and lifestyle factors as contributors to the variations in arthritis prevalence. Having found that regional income and racial/cultural origin were both significantly associated with individual arthritis status and significant moderators of the individual effects of income and racial/cultural origin, this study also contributes to a growing body of literature that recognizes the importance of “place” in the health of populations. Although the regional effects we report were modest, many strategies that address the broader determinants of health may be most effectively developed at the regional level. Further research is required to better understand the mechanisms that underlie these regional effects and to identify other contributing factors to the remaining variation in the prevalence of arthritis.


Ms Cañizares had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study design. Cañizares, Power, Perruccio, Badley.

Acquisition of data. Badley.

Analysis and interpretation of data. Cañizares, Power, Perruccio, Badley.

Manuscript preparation. Cañizares, Power, Perruccio, Badley.

Statistical analysis. Cañizares, Badley.