Independent and combined influence of homeownership, occupation, education, income, and community poverty on physical health in persons with arthritis

Authors


Abstract

Objective

To examine the independent and combined influence of individual- and community-level socioeconomic status (SES) measures on physical health status outcomes in people with self-reported arthritis.

Methods

From 2004–2005, 968 participants completed a telephone survey assessing health status, chronic conditions, community characteristics, and sociodemographic variables. Individual-level SES measures used included homeownership, occupation (professional or not), educational attainment (less than high school, high school degree, and more than high school), and income (<$15,000, $15,000–$45,000, and >$45,000). Community poverty (2000 US Census block group percentage of individuals living below the poverty line [low, medium, and high]) was used as a community-level SES measure. Outcomes were physical functioning (Medical Outcomes Study Short Form 12 version 2 physical component summary [PCS]), functional disability (Health Assessment Questionnaire [HAQ]), and the Centers for Disease Control and Prevention (CDC) Health-Related Quality of Life (HRQOL) Healthy Days physical and limited activity days, and were analyzed via multivariable regressions.

Results

When entered separately, all individual-level SES variables were significantly (P < 0.01) associated with poorer PCS, HAQ, and CDC HRQOL scores. A higher magnitude of effect was seen for household income, specifically <$15,000 per year in final models with all 4 individual SES measures and community poverty. The magnitude of effect for education is reduced and marginally significant for the PCS and number of physically unhealthy days. No effects were seen for occupation, homeownership, and community poverty.

Conclusion

Findings confirm that after adjusting for important covariates, lower individual- and community-level SES measures are associated with poorer physical health outcomes, while household income is the strongest predictor (as measured by both significance and effect) of poorer health status in final models. Studies not having participant-reported income available should make use of other SES measures, as they do independently predict physical health.

INTRODUCTION

The strong association throughout the developed world between lower levels of individual adult socioeconomic status (SES) and poorer health outcomes from many diseases (1–3), including arthritis (4–6), is well established, and recent data suggest that the socioeconomic environment of an individual's neighborhood may be important in this regard as well (7–9).

Several published studies from the UK, the US, and Canada have examined individual and community measures of SES with outcomes in persons with arthritis (8–13). In the UK, the Early Rheumatoid Arthritis Study Group found that greater deprivation measured by the Carstairs Index score was associated with a worse clinical score among 869 rheumatoid arthritis (RA) patients in the UK (8). Another UK study using data from the British RA Outcome Study Group found that area of residence deprivation (using a Townsend score) was related to RA severity at recruitment and was a predictor of response in a randomized controlled clinical trial (9). A third UK study used data from the Norfolk Arthritis Register, a large primary care cohort study. They examined the relationship between inflammatory polyarthritis and functional outcomes with SES at both the personal and area level, and found that individuals living in more deprived areas had poorer Health Assessment Questionnaire (HAQ) outcome scores (10).

In the US, a report from the Los Angeles Family and Neighborhood Survey found that having a chronic condition (including arthritis) was associated with substantially poorer self-rated health among participants in a deprived area than those in a more advantaged area (12). Neighborhood SES in the form of concentrated poverty was found to contribute to poorer physical functioning and depression scores in individuals with systemic lupus erythematosus, independent of individual SES (11). A study using the Canadian Community Health Survey found that living in low-income regions was associated with greater likelihood of reporting arthritis, with this relationship remaining after adjustment for individual SES (13).

Previous work by our group has demonstrated the impact of education and community social determinants on health outcomes in patients with self-reported arthritis (14, 15). A correlation was demonstrated between lower SES, specifically educational attainment and community poverty, and poor outcomes in health assessment measures (e.g., Medical Outcomes Study Short Form 12 version 2 [SF-12 v. 2], Centers for Disease Control and Prevention [CDC] Health-Related Quality of Life [HRQOL] Health Days measure). Formal educational attainment is, in part, a marker for behavioral variables (16–20). Community social determinants through the physical, social, and service characteristics of local neighborhoods can clearly impact residents' health.

Many studies examining the association of both individual and community social determinants with outcomes and mortality have been conducted in primarily urban areas (21, 22). Also, most studies usually include only one or two measures of individual SES (11–13), if any (23). This study focuses on a cohort of adults recruited from primary care practices from both rural and urban communities across North Carolina, and expands previous work from our group by examining 3 other measures of individual SES in addition to educational attainment and community poverty level. We include a dichotomous occupation variable, homeownership, and household income as SES measures and expand our physical health status outcomes to include both general health and arthritis-specific assessment measures.

SUBJECTS AND METHODS

Study design.

Established in 2001, the North Carolina Family Medicine Research Network (NC-FM-RN) is a practice-based patient cohort for primary care research that is frequently enriched (2004, 2005, 2008). Individuals visiting participating practices who were 18 years of age and older and provided consent to participate in the research study were invited to complete a survey about health conditions and health behaviors (24). This cohort is often used as a population for additional research studies, and the current social determinants study stems from the NC-FM-RN cohort (Figure 1). In 2004, 4,442 NC-FM-RM cohort participants were assessed for eligibility. Of those who had agreed to be contacted for future studies, had a current address and telephone number, and spoke English fluently, 4,165 were invited by a mailed letter to participate in our study addressing health outcomes. With a 59.5% response rate, 2,479 individuals were queried about demographic items, health status, chronic conditions, health attitudes and beliefs, and community characteristics in a telephone survey lasting approximately 45 minutes. All study materials and methods were approved by the University of North Carolina at Chapel Hill Biomedical Institutional Review Board.

Figure 1.

Participant recruitment and participation. NC-FM-RN = North Carolina Family Medicine Research Network; SODE = current Social Determinants of Health study; BMI = body mass index; PCS = physical component summary; HAQ = Health Assessment Questionnaire.

This study focuses on the 1,307 participants self-reporting arthritis according to the 2002 arthritis module of the Behavioral Risk Factor Surveillance System: any type of doctor-diagnosed arthritis, including osteoarthritis, RA, gout, lupus, or fibromyalgia (25).

Measures.

Physical health status outcomes.

Physical health status was assessed using the following 4 established measures. The physical component summary (PCS) of the standard Medical Outcomes Study SF-12 v. 2 was used to measure physical functioning. It is used in this study as a continuous variable, with a higher score (range 0–100) indicating better physical health. The scale is strongly correlated with the Short Form 36 (SF-36) and is reliable in general populations (26).

The CDC HRQOL Healthy Days measures were used to assess outcomes related to physical health and activity limitation (27). Two questions are used in this study: 1) “Now thinking about your physical health, which includes physical illness and injury, for how many days during the past 30 days was your physical health not good?” and 2) “During the past 30 days, for about how many days did poor physical or mental health keep you from doing your usual activities, such as self-care, work, or recreation?” Both measures are assessed on a scale from 0–30, with a higher score or more days indicating worse health; these measures are treated as continuous variables in all analyses. The CDC HRQOL questions have been validated and have good construct, acceptable criterion, and known groups validity (28).

The HAQ measures self-reported disability in daily function by assessing 20 activities of daily living organized around 8 domains: dressing and grooming, arising, eating, walking, hygiene, reach, grip, and outside activities. The level of difficulty for each is assessed on a scale from 0 (no difficulty) to 3 (unable to do). Domain scores are summed (range 0–24) and divided by 8 to provide a continuous, averaged index value from 0–3; a higher score indicates greater disability with a 0.22 change indicating clinical relevance (29).

Main socioeconomic predictor variables.

Main socioeconomic predictors were assessed using 4 individual-level SES measures and one community-level SES measure.

Educational attainment was assessed with 7 categories and later trichotomized as less than a high school degree, high school degree or GED, and greater than a high school degree (referent).

Household income was assessed using a 6 response-category format in $15,000 increments, ranging from less than $15,000 to greater than $75,000. For use in this study, household income was trichotomized as less than $15,000 per year, between $15,000 and $45,000 per year, and greater than $45,000 per year (referent).

Occupation was based on a participant description and coded using 2000 US Census occupation classification categories. This variable was further refined into physically demanding/nonprofessional (e.g., farming, fishing, service, construction, production, and labor) and non–physically demanding/professional (e.g., management, technical, sales, and office) categories for use in this study, with non–physically demanding/professional as referent.

Homeownership was assessed by asking participants “Do you own your home?” (yes or no), with homeowner as the referent.

Community SES was derived by matching each participant's home address to the related 2000 US Census block group, a geographical entity averaging approximately 1,000 residents (30, 31), using MapMarker Plus, version 7.2 (Empower Geographics). Only results with precise geography were used. The 2000 US Census block group poverty rate (percentage of the population in households with income below the poverty level) was used as a proxy for community-level SES (32). In some studies, block group characteristics have been suggested to be better indicators of the immediate SES environment than census tract measures (33). A poverty level category was assigned by dividing the sample into tertiles: low, medium, or high, without regard to race/ethnicity. The tertile cut points were 7.5% and 14.1%, meaning that approximately one-third of the sample lived in block groups with a household poverty rate greater than 14.1%. Community poverty is entered in models as indicators for medium and high household poverty rates, with low as the referent. Residents of a given block group share the same (community) household poverty rate.

Covariates.

In this study, covariates included participant sociodemographic characteristics (age, race, and sex) as well as health characteristics (body mass index [BMI] and number of comorbid conditions). Age was calculated using the participant's self-reported date of birth and date of telephone survey and used as a continuous measure. Race was self-reported and based on the 2000 US Census race and ethnicity categories and trichotomized into non-Hispanic white (referent), non-Hispanic black, and other, where other includes individuals self-reporting Latino/Hispanic ethnicity or more than one race (American Indian/Alaska Native, Asian, black or African American, Native Hawaiian or other Pacific Islander, white, and other). BMI (kg/m2) was calculated from self-reported height and weight, and used as a continuous measure. Existing comorbid conditions were assessed by asking participants if a health professional ever told them they had any of 23 different chronic diseases (e.g., diabetes mellitus, heart disease, vision problems). For this study, the number of comorbid conditions is a sum of all self-reported nonarthritis conditions.

Statistical analysis.

Statistical analyses were conducted on 968 participants who self-reported arthritis, after excluding observations missing covariates or main socioeconomic status predictors. Participants with no missing data were not significantly different by demographics or SES (individual and community) variables when compared to people with one or more missing items. However, individuals with missing data tended to report significantly poorer physical health, as they scored 2.3 points lower on the PCS, 0.15 points higher on the HAQ, and reported nearly 2 more days of physically unhealthy days, as well as nearly 2 more days of limited activity days. Figure 1 delineates data available for analysis, including the number of observations available for each outcome.

All data were analyzed using Stata, version 11.0 (34). Descriptive statistics were computed to describe the sample, and t-tests and chi-square tests were performed to evaluate statistical differences between non-Hispanic white and non-Hispanic black groups; the 61 participants categorized as “other race” were not included in this examination due to sparse numbers and their being heterogeneous in nature. Correlation analyses examined bivariate associations between individual- and community-level SES and physical health status outcomes. Multiple linear regression models, specifically analyses of covariance, were used to examine the associations of the 5 main SES predictor variables, both separately and together (in block stepwise analyses), with each of the 4 physical health status outcome measures. All models were adjusted for age, sex, BMI, race, and comorbid condition count. Since data were collected at 22 family practice sites across North Carolina, we employed the cluster option in Stata regression commands to account for potential intrasite correlation. This option produces Huber-White robust SEs and provides a more conservative approach to establishing significance of parameter estimates when compared to ordinary linear regressions (35, 36).

Because of the known complex relationships between race and socioeconomic status (e.g., non-Hispanic blacks more often have lower income and education), we evaluated for effect measure modification in each physical health status outcome. Race/SES interaction terms were added to each model separately and then combined. Of all the interaction terms tested, only one parameter estimate showed nominal significance (race and medium poverty; P = 0.012) and, under multiple testing, the Bonferroni criterion would not support a finding of effect measure modification. Analyses were, therefore, not stratified by race. However, we do present the characteristics of the study sample as an entire group (n = 968), and separately by non-Hispanic white (n = 731) and non-Hispanic black (n = 176).

For all categorical explanatory variables in this study, the referent was set to what is perceived as the most advantageous class (i.e., professional employment, homeowner, household income >$45,000, greater than a high school degree, low community poverty). We anticipated that the less advantageous categories would show outcome changes that indicated poorer health. Depending on the direction of each variable's scale, a deleterious change from the referent may be negative or positive.

RESULTS

Sociodemographic and outcome characteristics of the 968 participants with arthritis are shown in Table 1. The sample was mean ± SD 57 ± 14 years old, with a mean ± SD BMI of 31 ± 7 kg/m2 and a mean ± SD of 2 ± 2 comorbid conditions. The majority were women (74%), non-Hispanic white (76%), and educated (52% beyond high school), with a household income less than $45,000 (66%). Approximately half of the participants (49%) reported nonprofessional occupations that typically would be physically demanding (e.g., service industries, farming, manufacturing). Homeownership was reported by 78% of the participants.

Table 1. Participant sociodemographic and outcome characteristics*
VariablesEntire group (n = 968)Non-Hispanic white (n = 731)Non-Hispanic black (n = 176)P
  • *

    Values are the percentage unless otherwise indicated. BMI = body mass index; n/a = not applicable; PCS = physical component summary; SF-12 v. 2 = Short Form 12 version 2; HAQ = Health Assessment Questionnaire.

  • Other, n = 61.

  • Sample size varies for outcomes: PCS, n = 933; unhealthy physical days, n = 955; limited activity days, n = 600; HAQ, n = 967.

Age, mean ± SD years56.86 ± 13.6756.85 ± 13.7856.30 ± 13.850.631
BMI, mean ± SD kg/m230.57 ± 7.3729.82 ± 6.9233.94 ± 8.27< 0.001
Comorbid condition count, mean ± SD2.11 ± 1.582.07 ± 1.582.36 ± 1.580.029
Women73.5571.1484.090.001
Homeowner78.4182.3561.93< 0.001
Educational attainment   < 0.001
 More than a high school degree52.0755.9534.66 
 High school degree30.3729.0036.36 
 Less than a high school degree17.5615.0528.98 
Physically demanding occupation49.2845.2871.02< 0.001
Household income   < 0.001
 <$15,00025.8318.7453.98 
 $15,000–$45,00040.6042.4135.23 
 >$45,00033.5738.8510.80 
Race    
 Non-Hispanic white75.52n/an/a 
 Non-Hispanic black18.18n/an/a 
 Other6.30n/an/a 
Community poverty rate, mean ± SD12.18 ± 8.5911.28 ± 7.7315.94 ± 10.91< 0.001
Community poverty   < 0.001
 Low (<7.5%)33.4736.2521.02 
 Medium (7.5–14.1%)33.5734.4731.82 
 High (>14.1%)32.9529.2747.16 
Physical health status outcomes, mean ± SD    
 PCS (SF-12 v. 2)40.74 ± 12.1241.08 ± 12.3739.38 ± 10.990.099
 Unhealthy physical days8.66 ± 10.658.66 ± 10.868.59 ± 9.950.941
 Limited activity days9.21 ± 10.329.09 ± 10.539.87 ± 9.580.489
 HAQ0.67 ± 0.730.63 ± 0.710.84 ± 0.82< 0.001

For each physical health status outcome assessed, the participants reported a mean ± SD PCS score of 40.74 ± 12.12 and a mean ± SD HAQ score of 0.67 ± 0.73. On average, they reported a mean ± SD of 8.66 ± 10.65 unhealthy physical days and 9.21 ± 10.32 limited activity days per month on the CDC HRQOL Healthy Days scale.

Also shown in Table 1 are the sociodemographic characteristics stratified by non-Hispanic white and non-Hispanic black racial subgroups, with P values reported for each test to indicate substantial differences by race. Results show that non-Hispanic blacks are more likely than non-Hispanic whites to have lower education, to be non-homeowners, to have a nonprofessional, physically demanding occupation, to have lower household income, and to live in a community with a high poverty rate. Non-Hispanic blacks had greater disability than their non-Hispanic white counterparts, as noted with higher HAQ scores. While there were no significant differences by race for the PCS, number of physically unhealthy days, or number of limited activity days, there was a trend toward lower PCS scores for non-Hispanic blacks.

Correlation analyses (data not shown) revealed that individual and community SES variables were weakly correlated with each other (ranging from 0.10 to 0.40; P < 0.01), and also weakly correlated with the physical health outcomes (ranging from 0.06 to 0.36; P < 0.01). Correlation analyses between the physical health outcomes indicate that they are moderately correlated with each other (ranging from 0.55 to 0.73; P < 0.001).

We examined the independent effect of each SES variable alone or in concert on the physical health status outcomes. A staged set of models was developed to allow for in-depth examination, with results of these analyses shown in Tables 2, 3, and 4. Since the referent is conceptualized as the most advantaged class in all cases, the PCS (higher score implies better health) tends to have negative parameter estimates for the SES variables, while the other outcomes (higher scores imply poorer health) tend to have positive estimates.

Table 2. Parameter estimates for each of the 5 socioeconomic status variables singly with 4 physical health outcomes*
 SF-12 v. 2 PCS (range 0–100)Physically unhealthy days (range 0–30)Limited activity days (range 0–30)Disability status: HAQ score (range 0–3)
  • *

    Values are the β (95% confidence interval). Five separate models are run for each physical health outcome. All models are adjusted for age, sex, body mass index, race, and comorbid condition count. SF-12 v. 2 = Short Form 12 version 2; PCS = physical component summary; HAQ = Health Assessment Questionnaire.

Educational attainment    
 High school degree−2.72 (−4.13, −1.32)2.18 (0.57, 3.78)1.21 (−0.70, 3.13)0.08 (−0.01, 0.16)
 P≤ 0.0010.0100.2020.067
 Less than a high  school degree−4.83 (−6.74, −2.92)3.56 (1.83, 5.28)3.51 (0.39, 6.64)0.24 (0.08, 0.40)
 P≤ 0.001≤ 0.0010.0290.005
Physically demanding  occupation−2.19 (−3.44, −0.94)1.67 (0.08, 3.26)1.84 (−0.07, 3.75)0.12 (0.03, 0.22)
 P0.0020.0410.0580.010
Non-homeowner−2.74 (−4.18, −1.31)2.01 (0.28, 3.73)2.20 (0.28, 4.13)0.15 (0.07, 0.23)
 P≤ 0.0010.0250.027≤ 0.001
Household income    
 $15,000–$45,000−3.97 (−5.72, −2.22)1.87 (0.65, 3.10)3.25 (1.45, 5.06)0.14 (0.02, 0.25)
 P≤ 0.0010.005≤ 0.0010.020
 <$15,000−8.51 (−10.43, −6.58)5.86 (4.39, 7.33)7.41 (5.38, 9.44)0.48 (0.36, 0.61)
 P≤ 0.001≤ 0.001≤ 0.001≤ 0.001
Community poverty    
 Medium (7.5–14.1%)−0.32 (−2.35, 1.71)0.65 (−0.76, 2.06)0.68 (−1.19, 2.54)0.01 (−0.14, 0.15)
 P0.7500.3470.4590.919
 High (>14.1%)−1.20 (−2.89, 0.49)1.36 (0.28, 2.43)2.10 (−0.27, 4.47)0.10 (0.03, 0.17)
 P0.1530.0160.0800.008
Table 3. Parameter estimates for each of the 4 individual SES paired with community SES with 4 physical health outcomes*
 SF-12 v. 2 PCS (range 0–100)Physically unhealthy days (range 0–30)Limited activity days (range 0–30)Disability status: HAQ score (range 0–3)
  • *

    Values are the β (95% confidence interval). Four separate models are run for each physical health outcome. All models are adjusted for age, sex, body mass index, race, and comorbid condition count. SES = socioeconomic status; SF-12 v. 2 = Short Form 12 version 2; PCS = physical component summary; HAQ = Health Assessment Questionnaire.

Educational attainment    
 High school degree−2.72 (−4.13, −1.31)2.12 (0.47, 3.77)1.11 (−0.85, 3.07)0.08 (−0.01, 0.16)
 P≤ 0.0010.0140.2520.088
 Less than a high school degree−4.89 (−6.79, −2.99)3.53 (1.75, 5.32)3.50 (0.26, 6.73)0.25 (0.08, 0.41)
 P≤ 0.001≤ 0.0010.0360.005
 Community poverty    
  Medium (7.5–14.1%)0.32 (−1.53, 2.17)0.21 (−1.13, 1.55)0.32 (−1.69, 2.33)−0.02 (−0.16, 0.12)
  P0.7210.7440.7430.770
  High (>14.1%)−0.86 (−2.40, 0.68)1.10 (−0.03, 2.23)1.93 (−0.51, 4.38)0.08 (0.01, 0.16)
  P0.2580.0550.1150.029
Income    
 $15,000–$45,000−4.04 (−5.80, −2.27)1.84 (0.62, 3.07)3.22 (1.44, 5.00)0.14 (0.02, 0.26)
 P≤ 0.0010.005≤ 0.0010.021
 <$15,000−8.56 (−10.52, −6.60)5.78 (4.28, 7.28)7.24 (4.93, 9.55)0.48 (0.34, 0.62)
 P≤ 0.001≤ 0.001≤ 0.001≤ 0.001
 Community poverty    
  Medium (7.5–14.1%)0.63 (−1.29, 2.56)0.06 (−1.30, 1.42)−0.03 (−1.99, 1.93)−0.04 (−0.18, 0.10)
  P0.5020.9300.9730.536
  High (>14.1%)0.05 (−1.48, 1.57)0.42 (−0.78, 1.61)0.94 (−1.53, 3.40)0.02 (−0.05, 0.10)
  P0.9490.4770.4390.562
Physically demanding occupation−2.17 (−3.42, −0.92)1.61 (−0.01, 3.24)1.80 (−0.11, 3.72)0.12 (0.04, 0.21)
 P0.0020.0520.0640.008
 Community poverty    
  Medium (7.5–14.1%)−0.03 (−2.04, 1.99)0.44 (−1.03, 1.90)0.47 (−1.59, 2.53)−0.01 (−0.15, 0.13)
  P0.9790.5410.6400.905
  High (>14.1%)−1.02 (−2.69, 0.65)1.22 (0.17, 2.27)1.99 (−0.52, 4.50)0.09 (0.02, 0.16)
  P0.2160.0250.1140.016
Non-homeowner−2.69 (−4.16, −1.22)1.92 (0.16, 3.69)2.13 (0.14, 4.12)0.15 (0.07, 0.22)
 P≤ 0.0010.0340.037≤ 0.001
 Community poverty    
  Medium (7.5–14.1%)−0.07 (−2.20, 2.07)0.49 (−0.99, 1.96)0.40 (−1.59, 2.39)−0.01 (−0.15, 0.14)
  P0.9480.4990.6790.938
  High (>14.1%)−0.98 (−2.69, 0.72)1.22 (0.11, 2.33)1.93 (−0.47, 4.33)0.09 (0.02, 0.15)
  P0.2450.0330.1100.016
Table 4. Parameter estimates for all 5 socioeconomic status variables cumulatively in blocks arranged according to level of influence*
 SF-12 v. 2 PCS (range 0–100)Physically unhealthy days (range 0–30)Limited activity days (range 0–30)Disability status: HAQ score (range 0–3)
  • *

    Values are the β (95% confidence interval). All models in blocks 1–4 are adjusted for age, sex, body mass index, race, and comorbid condition count. SF-12 v. 2 = Short Form 12 version 2; PCS = physical component summary; HAQ = Health Assessment Questionnaire.

Block 1    
 Occupation−2.03 (−3.29, −0.76)1.54 (−0.03, 3.11)1.66 (−0.13, 3.46)0.12 (0.02, 0.21)
 P0.0030.0540.0680.018
 Non-homeowner−2.55 (−3.95, −1.16)1.85 (0.12, 3.58)2.01 (0.17, 3.85)0.14 (0.06, 0.22)
 P≤ 0.0010.0370.0340.002
Block 2    
 Occupation−2.03 (−3.29, −0.78)1.50 (−0.10, 3.10)1.65 (−0.16, 3.46)0.12 (0.03, 0.20)
 P0.0030.0640.0720.013
 Non-homeowner−2.53 (−3.96, −1.10)1.79 (0.04, 3.54)1.97 (0.05, 3.88)0.14 (0.05, 0.22)
 P≤ 0.0010.0460.0440.003
 Community poverty    
  Medium (7.5–14.1%)0.19 (−1.93, 2.30)0.30 (−1.22, 1.82)0.23 (−1.93, 2.39)−0.02 (−0.16, 0.12)
  P0.8550.6870.8240.784
  High (>14.1%)−0.83 (−2.51, 0.86)1.09 (0.02, 2.18)1.84 (−0.69, 4.37)0.08 (0.01, 0.14)
  P0.3190.0460.1450.027
Block 3    
 Occupation−0.83 (−2.04, 0.38)0.63 (−0.81, 2.06)0.99 (−0.43, 2.40)0.06 (−0.02, 0.15)
 P0.1690.3740.1630.138
 Non-homeowner−2.25 (−3.84, −0.67)1.61 (−0.14, 3.36)1.85 (0.17, 3.52)0.12 (0.03, 0.22)
 P0.0070.0690.0320.014
 Community poverty    
  Medium (7.5–14.1%)0.56 (−1.39, 2.51)0.05 (−1.37, 1.47)0.03 (−2.15, 2.21)−0.03 (−0.17, 0.11)
  P0.5560.9460.9760.624
  High (>14.1%)−0.64 (−2.19, 0.91)0.96 (−0.17, 2.08)1.75 (−0.77, 4.27)0.07 (0.01, 0.14)
  P0.4000.0920.1630.048
 Education    
  High school degree−2.54 (−4.00, −1.09)1.98 (0.36, 3.60)0.96 (−0.95, 2.87)0.06 (−0.03, 0.15)
  P0.0020.0190.3060.160
  Less than a high school degree−4.16 (−6.19, −2.14)3.00 (1.36, 4.64)2.87 (0.09, 5.65)0.20 (0.02, 0.37)
  P≤ 0.001≤ 0.0010.0440.027
Block 4    
 Occupation0.15 (−1.05, 1.35)0.07 (−1.27, 1.42)0.35 (−0.98, 1.67)0.01 (−0.06, 0.09)
 P0.8000.9120.5920.723
 Non-homeowner−0.75 (−2.19, 0.69)0.58 (−1.30, 2.46)0.36 (−1.61, 2.33)0.03 (−0.08, 0.14)
 P0.2900.5300.7060.585
 Community poverty    
  Medium (7.5–14.1%)0.87 (−1.02, 2.76)−0.12 (−1.50, 1.26)−0.17 (−2.24, 1.90)−0.05 (−0.18, 0.09)
  P0.3500.8570.8670.462
  High (>14.1%)0.11 (−1.40, 1.62)0.38 (−0.85, 1.60)0.95 (−1.54, 3.44)0.02 (−0.06, 0.10)
  P0.8800.5280.4360.568
 Education    
  High school degree−1.58 (−3.00, −0.15)1.42 (−0.22, 3.06)0.05 (−2.06, 2.16)0.01 (−0.08, 0.10)
  P0.0310.0860.9600.814
  Less than a high school degree−2.19 (−4.41, 0.03)1.67 (−0.24, 3.58)1.07 (−1.93, 4.06)0.08 (−0.12, 0.27)
  P0.0530.0830.4670.418
 Income    
  $15,000–$45,000−3.52 (−5.17, −1.86)1.36 (0.09, 2.63)3.01 (0.92, 5.11)0.13 (−0.01, 0.26)
  P≤ 0.0010.0370.0070.056
  <$15,000−7.43 (−9.54, −5.32)4.79 (2.81, 6.78)6.65 (3.71, 9.59)0.44 (0.27, 0.61)
  P≤ 0.001≤ 0.001≤ 0.001≤ 0.001

Each of the 5 individual-level SES variables was entered singly into models for each of the 4 outcomes, adjusting for age, sex, BMI, race, and comorbid condition count. Most of the resulting estimates (Table 2) are significantly different from zero, with most being P < 0.01. The effect of education on each of the physical health outcomes is shown in Table 2. People with less than a high school education scored ∼4.8 points lower on the PCS compared to those with education beyond high school. They also reported ∼3.6 more days per month of poor physical health and 3.5 more days of limited activity related to health. Finally, participants with less than a high school education scored 0.24 higher on the HAQ scale of disability, indicating greater disability.

While the significant role of educational attainment, occupation, and homeownership is observed for all physical health outcomes, a household income of <$15,000 has the largest negative effects on the PCS and physically unhealthy days outcomes. The role of community poverty is not as strongly associated with worse physical functioning (PCS) as are the individual-level SES markers; however, high community poverty is significantly associated with a greater number of physically unhealthy days and HAQ disability, with a trend for greater limited activity days.

The relationship between each of the 4 individual SES measures (education, household income, occupation, and homeownership) and physical health outcomes was then examined in the context of community poverty, adjusting for covariates (Table 3). The first set of models examined educational attainment as the individual-level SES measure and community poverty level on physical health outcomes. Individuals with less than a high school degree had significantly poorer scores on all 4 health status outcomes compared to individuals with greater than a high school degree for physical functioning (β = −4.89, P ≤ 0.001), physically unhealthy days (β = 3.53, P ≤ 0.001), limited activity days (β = 3.50, P = 0.036), and HAQ disability (β = 0.25, P = 0.005). Individuals with a high school degree had significantly poorer PCS and physically unhealthy days, as well as a trend for a poorer HAQ score compared to those with greater than a high school degree. In two models, living in the poorest communities (i.e., highest community poverty rate) was associated with poorer health status outcomes independent of education level. The high poverty group had a statistically significant negative impact on HAQ disability scores (β = 0.08, P = 0.029) and a trend for greater physically unhealthy days (β = 1.10, P = 0.055). Comparing these results to those in Table 2, we observed a similar educational attainment effect, but a reduced community poverty effect. Educational attainment appears to more strongly explain all physical health outcomes, although both remain significant for HAQ disability.

This dynamic is observed in the occupation and homeownership sets of regression models as well. When compared to the referent groups, having a physically demanding job and not owning a home both were associated with poorer physical health outcomes. As before, when these groups were individually adjusted for community-level poverty, poorer HAQ disability scores and greater numbers of physically unhealthy days were seen in the highest-level poverty groups. Compared to results shown in Table 2, we see that the effects of occupation and homeownership seem to remain independent and decrease only slightly.

Most striking are those models including both household income and community poverty (Table 3). The effects for household income approximate those given in Table 2 and remain significant, while the significant effects for community poverty are eliminated. The information provided by community poverty is apparently subsumed by the individual income and no independent association with any physical health outcome remains. When compared to those making more than $45,000, participants earning less than $15,000 had lower PCS scores by 8.56 points (P ≤ 0.001) and higher HAQ disability scores by nearly half a point (β = 0.48, P ≤ 0.001). Additionally, those with the lowest household income had an increased number of physically unhealthy days, 5.78 days more per month (P ≤ 0.001), and increased limited activity days, 7.24 days more per month (P ≤ 0.001). Lesser but significant (P < 0.01) associations were seen for results comparing participants from households earning $15,000–$45,000 to the more than $45,000 income group.

Finally, a series of models successively adds SES variables until all 5 SES variables are in each model (Table 4). Considering both the relative sizes of parameter estimates and their P values for each outcome in Tables 2 and 3, the order of the introduction of SES variables has been arranged so the variables that tended to be less influential are entered first (i.e., occupation, homeownership), community poverty next, and then education; the final variable entered is household income.

The block 1 models have occupation and homeownership entered together and they are independently significant for the PCS and HAQ. Community poverty was added in the second set of models (block 2), and living in the poorest community is associated with a greater number of physically unhealthy days (β = 1.09, P = 0.046) and greater HAQ disability score (β = 0.08, P = 0.027). Moreover, significant independent effects are maintained in occupation and homeownership. In block 3, educational status is included in the models. While the effects for less than a high school degree are significant, homeownership and high community poverty also show independent significance for some of the outcomes. However, the significant independent effect of occupation is eliminated with the addition of education into this model.

Block 4 adds household income to the other variables to create models that contain all the SES variables. Income is significant at both levels (<$15,000 and $15,000–$45,000) relative to the >$45,000 level, while all other SES variables lose significance, with the exception of high school education on PCS, where it remains a significant individual-level SES variable.

DISCUSSION

Previous research has revealed that individual- and community-level SES independently influence health outcomes, with lower SES (particularly educational attainment, income, and area-level deprivation) being associated with a greater risk of rheumatic conditions (13, 14, 37–39) and worse health outcomes for people with arthritis (8–11, 15, 40). In addition, low occupational status was found to be associated with poorer health and chronic disease, including arthritis (41). Yet there remained a need to go beyond these known relationships to better understand to what extent household income influences physical health outcomes, given multiple measures of individual-level SES factors (e.g., occupation, homeownership, and education).

This study identified the importance of individual SES, particularly household income (at both levels, <$15,000 and $15,000–$45,000) relative to the >$45,000 level, while most other SES variables lose significance. One observed exception was educational attainment, where those with less than a high school education had worse physical functioning and education remained a significant individual-level SES variable. It is important to consider that even while small differences in outcomes may be statistically significant if the sample size is large as in our case, this statistical significance may not constitute what an individual with arthritis would report as an important difference in physical health outcome, such as physical functioning or disability. Previous research has proposed that a minimally important difference (MID) that a patient would report is a universal value of an effect size (ES) of 0.5 (42), but is likely to be dependent on the particular instruments reported and the anchors used to elicit MID responses from patients (43). HAQ changes of approximately 0.20 have been reported as clinically important (44, 45), corresponding to an ES of 0.27. The PCS (SF-36) MID corresponds to an ES of 0.49 (46). Given this, we consider that the parameter estimates may be interpreted as changes in the physical health measures resulting from a treatment consisting of the condition expressed by the SES variable relative to its referent class, and we calculated quasi ES for all outcomes under this paradigm. For example, the PCS parameter estimate for the lowest income (<$15,000) is −7.43, meaning that people with low income would have an average PCS that is 7.43 units less than those in the highest income group (>$45,000). Using the sample SD of PCS (12.12) from Table 1, a quasi ES would be 0.61 (7.43/12.12). Likewise, the quasi ES for physically unhealthy days, limited activity days, and the HAQ are 0.45, 0.64, and 0.60, respectively. Because expert opinion varies widely on what constitutes a meaningful difference and how this should be assessed, quasi ES are one ad hoc way to capture the magnitude of the effects, standardizing the parameter estimates by corresponding SDs. Our findings represent substantial effects judged against other estimates of MID.

Results from this study suggest that while household income does indeed play a dominant role, the other individual-level measures of SES independently predict physical health measures when examined in their own right. We propose that occupation, homeownership, and educational attainment remain important and good measures to include in further research examining the relationship between SES and arthritis-related outcomes. We believe that perhaps household income not only captures the objective number of dollars available to the individual and household, but also contains elements of occupational status, knowledge, and literacy, as well as the availability, level, and extent of resources and social capital contained in the other SES measures. At the very least, study findings suggest that having a household income above $45,000 provides an individual with arthritis the financial means to seek out goods and services, either within or outside their community, which is necessary for better general and arthritis-specific health outcomes.

This study uniquely illustrates both the separate and combined effect of individual and community socioeconomic factors on a variety of physical health outcome measures, including general health outcomes (e.g., SF-12 v. 2 PCS, the CDC HRQOL measures) and an arthritis-specific outcome (e.g., the HAQ). This study also used 4 measures of individual-level SES (occupation, homeownership, educational attainment, and household income) that have previously not been found in the arthritis literature to date. Additionally, this study has a moderately large number of participants (n = 968) who, as a group, are racially and geographically diverse. Although sampling procedures preclude complete generalizability of results to the general population, we do believe that the racial and geographic diversity increases the generalizability more than would be expected in a homogeneous sample.

Despite these strengths, there are a few limitations worthy of note, as with any large telephone survey study. First, this is a cross-sectional study, which does not allow for examination and determination of a causal pathway of individual and/or community SES influence on physical health outcomes. We caution that poor health may cause individuals to have access to fewer resources and place them in a lower SES condition. Second, this study may have been unintentionally influenced by two sources of bias: reporting bias and recall bias. The potential for reporting bias may come from our attempt to obtain information about household income. While participants are told at the beginning of the study that their participation and all information provided to the research study staff will be kept confidential, many participants of all income levels may remain uncomfortable with income disclosure (47). Reliance upon telephone survey data may introduce some level of participant recall bias. Third, this study did not ask participants to self-report health behaviors (e.g., smoking or level of physical activity), health insurance status, or disease duration (i.e., time since arthritis diagnosis). These factors are known to be unequally distributed by SES and could also be influentially associated with physical health among individuals with arthritis. Finally, the use of aggregate US Census data to proxy community SES can be considered a crude measure of community, given that it does not account for contextual community life details (e.g., neighborhood safety and publicly available opportunities for physical activity).

In conclusion, our study findings suggest that SES measures, particularly household income, play an important role in physical health outcomes among people with arthritis. Our study sheds light on the complex interplay between common markers of SES, i.e., occupation, homeownership, education, and income, as well as community poverty. Because the national health care agenda places a priority on reducing health disparities caused by inequalities in SES and race, our findings may be of particular interest to clinicians and/or public health practitioners seeking to reduce poorer health outcomes due to these social factors. Furthermore, additional research is needed to examine the influence of individual and community SES over the life course as it relates to general physical health status in people with arthritis.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Callahan 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 conception and design. Callahan.

Acquisition of data. Callahan, Martin, Schoster.

Analysis and interpretation of data. Callahan, Martin, Shreffler, Kumar, Kaufman, Schwartz.

Acknowledgements

We would like to thank Robert DeVellis, Thelma Mielenz, Randy Randolph, Philip Sloane, and Morris Weinberger for their contributions and input to the study. We thank the following participating family practices in the NC-FM-RN for their assistance: Black River Health Services, Burgaw; Bladen Medical Associates, Elizabethtown; Blair Family Medicine, Wallace; Cabarrus Family Medicine, Concord; Cabarrus Family Medicine, Harrisburg; Cabarrus Family Medicine, Kannapolis; Cabarrus Family Medicine, Mt. Pleasant; Chatham Primary Care, Siler City; CMC-Biddle Point, Charlotte; CMC-North Park, Charlotte; Community Family Practice, Asheville; Cornerstone Medical Center, Burlington; Dayspring Family Medicine, Eden; Family Practice of Summerfield, Summerfield; Goldsboro Family Physicians, Goldsboro; Henderson Family Health Center, Henderson; Orange Family Medical Group, Hillsborough; Person Family Medical Center, Roxboro; Pittsboro Family Medicine, Pittsboro; Prospect Hill Community Health Center, Prospect Hill; Robbins Family Practice, Robbins; and Village Family Medicine, Chapel Hill. Finally, we thank the individuals who willingly participated in the study.

Ancillary