Abstract
Objective Socioeconomic status (SES) is associated with differences in health outcomes for individuals with rheumatoid arthritis (RA). We aimed to determine the effect of area-level SES on RA disease activity, disability, quality of life (QOL), and biologic/targeted synthetic disease-modifying antirheumatic drug (b/tsDMARD) use in individuals with early RA managed within a protocolized, treat-to-target (T2T), longitudinal observational cohort study.
Methods Adult patients with RA diagnosed after June 2003 were included. SES was defined as quintiles of Index of Relative Social Advantage and Disadvantage (IRSAD) based on residential address at baseline. Covariates included baseline age, sex, smoking status, BMI, and the Rheumatic Disease Comorbidity Index (RDCI). Longitudinal multivariable random effects regression models were constructed with restricted cubic splines examining nonlinear responses in outcome variables. b/tsDMARD use was examined using time-to-event models for recurrent events.
Results Of 255 participants, 66.7% were female, with mean age 53.9 years, and 68% had seropositive disease. There was an ordered trend across SES quintiles, such that higher quintiles were associated with lower Disease Activity Score in 28 joints based on C-reactive Protein (DAS28-CRP; P = 0.03), lower modified Health Assessment Questionnaire (mHAQ; P = 0.001), and higher 36-item Short Form Health Survey physical component summary (SF-36 PCS; P < 0.001). SES quintile was not significantly associated with b/tsDMARD initiation or switching.
Conclusion Disadvantageous SES was associated with higher disease activity, disability, and poorer QOL. Our results suggest an inequity in health outcomes for patients with RA despite T2T management within a universal healthcare system.
Despite the widespread availability of highly efficacious treatments for rheumatoid arthritis (RA), individuals with disadvantageous socioeconomic status (SES) continue to experience poorer health outcomes. Previous studies have found that individuals with low SES experience greater RA disease activity1-8 and disability,2-7,9,10 among other patient-reported outcome measures (PROMs), including greater pain,7 fatigue,11 and lower quality of life (QOL).12
The mechanisms by which SES may contribute to poorer health outcomes in patients with RA remain under debate. Potential explanations for the association between low SES and poorer RA outcomes include effects of smoking,13 obesity,14 greater allostatic load (ie, biological “wear and tear”),15 early adverse childhood experiences,16 and differences in medication use (eg, delays in initiation of biologic disease-modifying antirheumatic drugs [bDMARDs] related to variation in practice or affordability).17
The effect of SES has been studied in individuals with early RA (and early inflammatory arthritis) in multiple observational cohorts.2,3,6,7,9,18 Although some are inception cohorts,19 study aims, participants, and treatment protocols differ, which may limit the capture of sufficiently homogenous data to enable testing of relevant hypotheses. Similarly, detailed medication use data may not be available to interpret and understand disease outcome data.
The hypothesis that SES influences disease outcomes in individuals with early RA would be best evaluated in a prospective longitudinal inception cohort, managed with a contemporary standardized approach to treatment, such as the treat-to-target (T2T) strategy. Simultaneous collection of comprehensive RA disease outcome measures and medication use, as well as sociodemographic details, would enable the exploration of how some theoretical mechanisms, such as differences in medication use, contribute to any relationship between SES and RA outcomes identified.
Our primary aim was, therefore, to investigate the effect of SES on disease outcomes in a prospective, single-center study of an early RA inception cohort managed by a T2T protocol within a universal healthcare system. The secondary aims were to investigate the effect of SES on disability, QOL, bDMARD/targeted synthetic DMARD (tsDMARD) initiation or switching, glucocorticoid (GC) use, and visit frequency.
METHODS
Participants. The Royal Adelaide Hospital (RAH) is a tertiary public hospital based centrally in the city of Adelaide that receives patients from urban and rural locations throughout South Australia. The RAH early arthritis cohort (EAC) study is an ongoing longitudinal observational cohort study, established in 1998, enrolling DMARD-naïve individuals with onset of polyarthritis within 12 months of their initial rheumatology clinic appointment, who agree to receive disease management using a standardized T2T approach.20 This treatment escalation algorithm has been updated as major societal guidelines have changed; for example, b/tsDMARDs were added to the protocol as they became available and some older conventional synthetic DMARDs (csDMARDs) have been removed. Participants are managed with initial csDMARD triple therapy with methotrexate, sulfasalazine, and hydroxychloroquine. A T2T strategy is then followed with dose escalation of the initial csDMARDs, followed by the addition of leflunomide and/or b/tsDMARDs until the treatment target of Disease Activity Score in 28 joints (DAS28) remission (< 2.6) is reached; prior to 2009, criteria-based treatment intensification was used.21 The timeline of treatment target changes is shown in Supplementary Figure S1 (available with the online version of this article). Clinic visits occur every 3-6 weeks and then 3-6-monthly once the treatment target has been reached. Patients have access to rheumatology nurses during and between visits. Use of b/tsDMARDs for eligible patients under the Australian Pharmaceutical Benefits Scheme (PBS) is permitted after 6 months of therapy with csDMARDs.22 The PBS caps patients’ medication copayments (out-of-pocket cost to the patient) for prescriptions (currently set at A $31.60 [US $22.15],23 with lower copayments set at A $7.70 [US $5.40] for those with lower incomes, pensioners, or veterans). The remaining cost of medicines is paid by the Australian Government.23
For this study, we included newly diagnosed individuals who entered the EAC cohort from June 1, 2003 onward, reflecting the era when PBS-subsidized access to b/tsDMARDs for patients with RA commenced.24 We excluded individuals whose residential address at baseline was outside of the state of South Australia, enabling us to determine within-state relative area-level SES.
Variables. In addition to routine care at clinic visits, clinical data are collected prospectively by written survey and by trained metrologists. Study entry date was defined as the date the patient started their first DMARD as part of the triple therapy regimen. The exposure (independent) variable was baseline area-level SES, which was determined by geocoding participants’ residential address to Statistical Area 1 (SA1) areas (Australian Census tracts with, on average, 400 inhabitants).25 The Australian Bureau of Statistics’ (ABS) publicly available Socio-Economic Indexes for Areas (SEIFA) was used for this purpose. The subindex, known as the Index of Relative Social Advantage and Disadvantage (IRSAD), based on population data from the 2011 Australian Census, was used to classify each participant into South Australian standardized area-level SES quintiles. IRSAD is a predefined socioeconomic index incorporating information on average household income, education, occupation, and employment status, among other factors.26 The 2011 SEIFA data were chosen in preference to other years of Census data as 2011 best matched the median year of participants treatment commencement.
The primary outcome measure was DAS28 based on C-reactive Protein (DAS28-CRP). Secondary PROMs included the modified Health Assessment Questionnaire (mHAQ), a measure of disability due to arthritis with demonstrated prognostic value for patients with RA,27 and the 36-item Short Form Health Survey (SF-36) physical and mental component summaries (PCS and MCS, respectively), validated for assessing health-related QOL in patients with RA.28 SF-36 PCS and MCS were normalized to Australian population data.29 Initiation or switching of b/tsDMARDs was an additional secondary outcome, highlighting instances where participants required escalation or change in treatment in response to persistent RA disease activity. Oral and parenteral GC use was also recorded.
Baseline demographic and disease covariates included were age (categorized as < 45, 45-65, and ≥ 65 years), sex, race and ethnicity, smoking status, BMI (calculated as weight in kilograms divided by height in meters squared [kg/m2]; categorized as < 25, 25-29, and ≥ 30), comorbidity score using the Rheumatic Disease Comorbidity Index (RDCI; scores 0-9),30 patient and physician global assessments, pain visual analog scales, and days from symptom onset and diagnosis of polyarthritis to DMARD commencement.
Data analysis. Descriptive statistics were used to summarize the sample’s characteristics and outcome measures at baseline.
We used multilevel, multivariable longitudinal regression with random effects set at the participant ID-level to model the effect of SES on disease outcomes (random intercept and random slopes over time). Ordinal trends across the 5 SES quintiles were evaluated by the linear (βlinear) orthogonal polynomial contrast. Models included the following additional covariate terms relevant to our theorized model of the effect of SES on RA outcomes: baseline age, BMI, smoking status, and comorbidity (Supplementary Figure S2, available with the online version of this article). Our results therefore theoretically reflect the direct effect of SES, adjusted for the effect of participant comorbidity. Allowing for variability between participants at baseline and in their disease activity trajectory, this included random effect terms for both model intercept and slope. Restricted cubic splines were used in constructing models to allow description of nonlinear responses in outcomes across follow-up time, measured in years. All available outcome data across participants’ clinic visits were used in constructing the models. Initial models were iteratively refined through review of the Akaike information criterion and Bayesian information criterion to determine the appropriate spline degrees of freedom, with knots placed at default centiles, to prevent overfitting while also considering model parsimony. Predicted values for disease outcomes from these models, with additional covariates set to sample mean values, were generated to 10 years of follow-up, and presented graphically. Sensitivity analyses were performed to examine whether SES was primarily associated with significant differences in either subcomponent of the DAS28-CRP score (ie, the DAS28 Patient-derived [DAS28-P] index,31 which indicates contribution of patient-reported components [tender joint count and patient global health] to overall disease score, and the 2-component DAS28 [2C-DAS28], focused on clinician/laboratory-derived components [swollen joint count and CRP]32). We also sought to establish if any significant interaction between SES and years since baseline visit was present.
Next, time-to-event analyses for recurrent events were used to determine the relationship between SES and b/tsDMARD initiation or switching. The “failure” event was defined as initiation or switch in b/tsDMARD per participant. Recurrent events were first analyzed by participant ID-level random effect time-to-event models (Royston-Parmar models using restricted cubic splines to model the baseline hazard function), with the time scale being years from date of treatment commencement. Informed by the results from the primary analysis, which investigated the effect of SES on disease activity, the SES variable was dichotomized to quintiles 1-3 and quintiles 4-5 for this latter analysis. Other covariates included baseline age (< 65 or ≥ 65), sex, and comorbidity score.
GC use was analyzed by calculating annual cumulative doses by combining all oral prednisolone equivalent doses and intramuscular/intraarticular parenteral GC use documented in clinical records and in written responses from participants on visit surveys. Parenteral GC doses were converted to oral prednisolone equivalent doses, enabling an overall cumulative dose to be determined, using accepted conversion factors.33 A multivariable panel Poisson regression was performed with total annual (by calendar year) oral prednisolone equivalent dose, exposure time (in years), and additional covariates of SES quintile, age (< 60 or ≥ 60 years), year group of cumulative dose (2003-2011, 2012-2017, or 2018-2023), and duration of follow-up at time of visit (≤ 2, 3-5, or > 5 years).
Similarly, annual (by calendar year) EAC clinic visit frequency was also analyzed using a multivariable panel Poisson regression with the RDCI included as an additional covariate with patient-level random effects.
Stata MP v18 (StataCorp) and the merlin R package34 were used to undertake all analyses.
RESULTS
Baseline characteristics. The final sample included 255 individuals, mostly female (66.7%) with median age 54 (IQR 44-65) years, followed for a median 7.7 (IQR 4.5-12.6) years. Most participants were seropositive with either rheumatoid factor (RF; 67.6%) and/or anticitrullinated protein antibodies (ACPA; 67.3%) detected. SES quintiles were evenly distributed among participants (Table 1; Supplementary Table S1, available with the online version of this article). The median duration from first symptom onset to treatment commencement was approximately 5 months. There was no effect of SES on time from symptom onset or diagnosis to treatment commencement, nor in years of follow-up within the study (data not shown).
Baseline characteristics.
Effect of SES on disease outcome measures. In longitudinal models, higher SES quintile was associated with lower coefficients for DAS28-CRP (Figure 1; Supplementary Table S2, available with the online version of this article). The “total effect” of SES (derived from a model adjusting for only baseline age and sex [ie, baseline model]) was found to have a significant ordinal relationship between increasing SES quintile and lower DAS28-CRP (βlinear −0.14 [95% CI −0.23 to −0.05], P = 0.003). The βlinear coefficient refers to the ordinal trends across the 5 SES quintiles, evaluated using orthogonal polynomial contrasts; the linear polynomial contrasts are referred to in these results. A “direct effect” of SES persisted after considering potential mediators of this relationship in a model including covariates for comorbidity, smoking status, and BMI (ie, adjusted model); however, the effect size was numerically reduced (βlinear −0.10 [95% CI −0.19 to −0.01], P = 0.03; Figure 1; Supplementary Table S3).
Graph of coefficients with 95% CIs from multivariable regression models of DAS28-CRP, mHAQ, and SF-36 PCS and MCS analyses. Q1 refers to lowest SES; Q5 refers to highest SES. DAS28-CRP: Disease Activity Score in 28 joints based on C-reactive protein; IRSAD: Index of Relative Social Advantage and Disadvantage; MCS: mental component summary; mHAQ: modified Health Assessment Questionnaire; PCS: physical component summary; Q: quintile; RDCI: Rheumatic Diseases Comorbidity Index; SES: area-level socioeconomic status; SF-36: 36-item Short Form Health Survey.
Similarly, higher SES quintile was associated with reduced mHAQ (Figure 1). As above, the total effect of SES on mHAQ (βlinear −0.06 [95% CI −0.09 to −0.03], P < 0.001) was slightly numerically greater than the direct effect (adjusted model: βlinear −0.06 [95% CI −0.09 to −0.02], P = 0.001; Figure 1; Supplementary Tables S2-S3, available with the online version of this article).
Higher SES quintile was associated with greater SF-36 PCS (baseline model: βlinear 1.44 [95% CI 0.64-2.24], P < 0.001), with little change in the adjusted model (βlinear 1.46 [95% CI 0.65-2.27], P < 0.001; Figure 1; Supplementary Tables S2-S3, available with the online version of this article). Notably, higher SES was not associated with significantly higher SF-36 MCS in the baseline (βlinear 0.74 [95% CI 0.07-1.56], P = 0.07) or adjusted models (βlinear 0.49 [95% CI −0.29 to 1.26], P = 0.22; Figure 1; Supplementary Tables S2-S3).
Age, sex, current smoking status, and BMI were not independently associated with differences in predicted outcome measures in these adjusted final models. Increased comorbidity was associated with significantly higher DAS28-CRP and mHAQ, and lower SF-36 PCS and MCS. Older age at baseline was associated with lower SF-36 PCS, and female sex was associated with lower SF-36 PCS and MCS. Predictions from the models with covariates set to mean sample values were obtained and are presented in Figure 2. Participants achieved maximal improvement in disease activity and disability over the first 2 years following treatment baseline. In the mHAQ outcome model, a random effect for slope could not be determined owing to convergence issues, so the above results of the mHAQ model include random effects for the intercept only.
Predicted disease activity, disability, and quality of life measures according to participants’ SES quintile to 10 years of follow-up. Q1 refers to lowest SES; Q5 refers to highest SES. DAS28-CRP: Disease Activity Score in 28 joints based on C-reactive protein; MCS: mental component summary; mHAQ: modified Health Assessment Questionnaire; PCS: physical component summary; Q: quintile; SES: area-level socioeconomic status; SF-36: 36-item Short Form Health Survey.
In sensitivity analyses, SES was associated with significant differences in coefficients for subcomponents of DAS28-CRP in multivariable models using DAS28-P and 2C-DAS28, suggesting that both subcomponents of the disease activity score are affected by SES (Supplementary Figure S3, available with the online version of this article). Additionally, inclusion of a multiplicative interaction term for SES and years since baseline visit in the multivariable model did not significantly affect the primary outcome results and the coefficient for the interaction term was not statistically significant (data not shown).
Time to first b/tsDMARD initiation and/or switching analysis. Among 74 (30.7%) participants who had at least 1 instance of b/tsDMARD use, there were 182 occasions where b/tsDMARDs were initiated or the medication was switched. The median number of b/tsDMARDs used was 2 (IQR 1-3). Overall, there were over 2000 person-years of follow-up with an observed crude incidence rate of b/tsDMARD initiation or switching of 9.1/100 participant years.
The results of the time-to-event models of recurrent b/tsDMARD events are presented in Table 2 and constructed based on theorized relationships (Supplementary Figure S4, available with the online version of this article). Model 1, which estimated the total effect of SES on bDMARD events, indicated that bDMARD events were somewhat fewer in higher SES participants, but this was not statistically significant (hazard ratio [HR] 0.77 [95% CI 0.51-1.15], P = 0.20). Model 2 includes adjustment for age and sex with little change in the coefficients. Model 3 is a sensitivity analysis to determine the presence of a “direct effect” of SES on b/tsDMARD initiation or switching events, therefore additionally including adjustment for comorbidity. In this model, both younger age (< 65 years) and greater comorbidity score at baseline were associated with a greater HR for b/tsDMARD initiation or switching events, and there was no statistically significant effect of sex.
Regression models of biologic or targeted synthetic DMARD initiation or switching events.
GC use analysis. There were 212 (83.1%) participants who used any oral (n = 89) or any parenteral (either intramuscular or intraarticular, n = 205) GC between their baseline and final clinic visit. Although the annual cumulative GC dose independently significantly decreased with older age, longer disease duration, and more recent calendar year, there was no relationship with SES quintile (adjusted βlinear −0.14 [95% CI −0.36 to 0.08], P = 0.20; data not shown). Similarly, a sensitivity analysis assessing only oral GC use identified no effect of SES quintile.
Visit frequency analysis. There was no significant overall effect of SES quintile on EAC visit frequency in the multivariable model (overall linear contrast of predictive margins −0.02, 95% CI −0.06 to 0.02, P = 0.28). Greater comorbidity was significantly associated with higher incident rate ratio (IRR) for visit frequency (1.06 [95% CI 1.03-1.09], P < 0.001), whereas longer duration of follow-up and appointments during more recent calendar years (2018-2023) were associated with lower IRRs (data not shown).
DISCUSSION
We have shown that in this early RA inception cohort, individuals living in lower SES areas had significantly greater disease activity (ie, predicted DAS28-CRP) than those living in more affluent/less disadvantaged areas, including after adjustment for relevant covariates on the theoretical causal pathway such as comorbidity, smoking status, and BMI (Supplementary Figure S2, available with the online version of this article). A similar relationship was observed with mHAQ and QOL scores. These findings are of particular significance in the context of this study, where it might have been expected that the effect of SES would be mitigated as all participants were managed in a dedicated early RA clinic, with initial triple therapy and protocolized treatment escalation. In addition, participants had access to subsidized medications and were treated in a public hospital outpatient clinic with no out-of-pocket costs. This study has therefore shown that lower SES is associated with poorer health outcomes despite effectively controlling for access to medication and healthcare as well as any between- or within-rheumatologist variation in treatment strategy. These results therefore suggest presence of a health inequity—a “gap” in health outcomes according to SES—that requires further explanation and exploration. We undertook additional analyses to explore the cause of this gap and found that medication use and visit frequency were not associated with SES. Further, there was also no effect of SES on time from symptom onset or diagnosis to treatment commencement in this cohort. The visit frequency analysis found no disadvantageous effect of SES on EAC visit frequency.
To further examine how SES affects composite disease activity scores, we examined whether there were differences in either participant- or clinician/pathology-derived subcomponents of DAS28-CRP (either DAS28-P or 2C-DAS28) according to SES. We did not find evidence to support clinically significant differences in these subcomponents by SES quintile, suggesting that SES exerts a global effect on RA disease activity.
Greater comorbidity was independently associated with poorer RA outcomes and greater b/tsDMARD use, and this may be a clue to understanding the unexplained gap in outcomes associated with SES. Greater comorbidity may contribute to greater care complexity, which, particularly in individuals with lower SES, may lead to greater burden of treatment35; for example, escalation of RA-directed immunosuppression in response to greater perceived disease activity. This is an emerging area of research in RA and may enable better understanding of currently unexplained gaps in patient outcomes. Future qualitative research would likely contribute to better understanding of the interplay of these issues.
Studying SES in a protocol-driven EAC mitigates against some of the limitations of studying the effects of SES using national registry data. Despite the appealing methodological advantages in trying to study the effect of SES on outcomes in EACs, there are relatively few studies focusing on this aim, and none published from Australian cohorts to our knowledge, making this study unique. We considered multiple potentially relevant mediators of the effect of SES on RA outcomes in designing this study, which could have affected the findings. Use of a uniformly applied treatment escalation protocol within 1 health service will have minimized interclinician or intraclinician treatment variation. The study’s sample was restricted to include participants managed after the listing of bDMARDs on the PBS, ensuring that the availability of these treatments was uniform across the sample.
The predicted differences in RA disease activity and disability found between individuals in the lowest and highest SES quintiles were not greater than previously specified minimum clinically important differences (MCID) for DAS28-CRP or mHAQ.36 This could reflect the cohort recruited into this study—individuals classified as low SES but whose personal circumstances are the “best case scenario.”7 Put another way, the entry requirements of the cohort, including agreeing to adhere to scheduled follow-up appointments, blood tests, surveys, phone calls, among others, may have differentially limited participation of individuals from lower SES backgrounds, who are already facing complexity in their social and economic circumstances. Therefore, our results may be an underestimate of true differences in disease activity according to SES. Nonetheless, between individuals from the lowest and highest SES quintiles, there was a difference in SF-36 PCS of 7.43 (95% CI 3.82-11.05, P < 0.001; Supplementary Table S2, available with the online version of this article), which is largely driven by differences between the fourth and fifth (ie, the highest) quintiles and approximates previous estimates of the MCID in individuals with RA.37 Better community access to physical therapy, assistive devices, and reduced effects of comorbidity in higher SES participants may contribute to this finding.
Management of participants through a single clinical service limits the generalizability of our findings, although the demographics and disease characteristics of this cohort are similar to those of other RA cohorts. Awareness of the availability and cost of health insurance to participants is important in interpreting the results of other studies on the effect of SES on health outcomes. Our findings may underestimate the effect of area-level SES in other contexts without access to universal health coverage.
Area-level SES may misclassify individuals but can be beneficial when person-level data are missing, including data about education or income. We chose to use a socioeconomic index drawn from the ABS’ 2011 Census data, as this was also the median year of participants’ treatment commencement and thus the closest approximation of their baseline SES. This may risk misclassifying the SES of some individuals who entered the study in later years. Nonetheless, when comparing 2011 and subsequent SEIFA 2016 data, 75% participants’ IRSAD quintile did not change. In the remainder, there was only minor change (ie, a single quintile increase or decrease) in their area-level SES quintile. Additionally, in the Australian context, area-level SES has been previously used to identify inequitable health care by other groups in conditions such as metabolic syndrome and diabetes mellitus.38,39
Finally, predictions from our models have been drawn from baseline to 10 years following treatment commencement (Figure 2), though some participants did not remain in the cohort for this duration. This may affect the reliability of the model’s predictions toward this timepoint.
Our results highlight the importance of socioeconomic factors in individuals with early RA. As RA-specific medications appear to be appropriately used, nonpharmacological interventions may be considered to begin addressing this gap, such as healthcare navigation and improving access to allied health services. Additionally, as we face an aging population with an increasing burden of treatment complexity, addressing the effect of comorbidities is of paramount importance, particularly the association between lower SES and comorbidity. Ultimately, although addressing the upstream structural social determinants of health remains central in aiming to reduce healthcare inequities, healthcare providers’ continued advocacy at both the individual and community levels remains vital in confronting these challenges.
Footnotes
R.J. Black and C.L. Hill contributed equally as co-senior authors.
CONTRIBUTIONS
OR: conceptualization, methodology, formal analysis, writing – original draft; SL: methodology, formal analysis, writing – review and editing; JS: formal analysis, writing – review and editing; KG: data curation, writing – review and editing; LM: conceptualization, writing – review and editing; AL: conceptualization, writing – review and editing; SP: conceptualization, resources, writing – review and editing; RJB: conceptualization, methodology, supervision, writing – review and editing; CLH: conceptualization, methodology, supervision, writing – review and editing.
FUNDING
Royal Adelaide Hospital Rheumatology Unit has received research funding from Melrose Health and the Health Services Charitable Gifts Board. OR was supported by an Australian Government Research Training Scholarship (Australian Research Training Place, Arthritis SA Post Graduate Research Scholarship).
COMPETING INTERESTS
The authors declare no conflicts of interest relevant to this article.
ETHICS AND PATIENT CONSENT
Approval to conduct this study was obtained from the Central Adelaide Local Health Network Human Research Ethics Committee (ref: 17652). Patient consent is obtained at the time of recruitment to participate in the EAC longitudinal study. A waiver of consent was approved for this substudy to enable access to medical records for the duration of the project.
DATA AVAILABILITY
Data cannot be shared for ethical/privacy reasons.
- Accepted for publication January 6, 2026.
- Copyright © 2026 by the Journal of Rheumatology
This is an Open Access article, which permits use, distribution, and reproduction, without modification, provided the original article is correctly cited and is not used for commercial purposes.
REFERENCES
SUPPLEMENTARY DATA
Supplementary material accompanies the online version of this article.
