The KORONA registry was established in 2009 by the Clinical Research Center for RA at Hanyang University, South Korea, with funding from the Ministry of Health and Welfare. It comprises a cohort of 5,077 patients diagnosed with RA who were recruited from 23 institutions across the country. All RA patients were aged ≥ 18 years and satisfied the 1987 American College of Rheumatology criteria for RA [17]. Upon enrollment, all patients completed an initial questionnaire capturing their demographic information, medical history, and disease-specific outcomes. Patient-related variables included age, sex, comorbidities, marital status, mean annual household income, education, and insurance type. Disease-related variables encompassed disease onset and diagnosis dates; disease activity measures; laboratory test results; and prescription information for disease-modifying antirheumatic drugs, nonsteroidal anti-inflammatory drugs, glucocorticoids, and biologics. Disease activity was assessed using the Disease Activity Score (DAS) 28-joint count erythrocyte sedimentation rate, EuroQol-5 dimensions (EQ-5D) with an additional depression dimension, a health assessment questionnaire (HAQ), and the physician’s visual analog scale, and joint assessments were performed by the patient’s rheumatologists and well-trained medical staff. Patient self-evaluation data were integrated into the database and updated during annual follow-up visits.

Data regarding body mass index, and health behaviors such as smoking, exercise, alcohol consumption, and healthcare utilization within the past 2 years, including hospitalization and various medical check-ups were collected. Medication compliance, and use of alternative medicines and supplemental medicines were also recorded. All items were assessed at enrollment and at annual follow-up visits. However, only baseline data were used for the current study. All data were anonymized, transferred to the Korean Society of Rheumatology for approval, and provided to the researchers. Hospitalization was defined as admissions to the hospital for RA-related reasons, excluding surgeries and normal delivery, within the past 2 years. Medication compliance was measured on a scale from 1 to 5, with 1 indicating the highest compliance level, representing patients who consistently took their medication within a 6-month period. Skipped medication refers to the number of days patients skipped medication, with 1 indicating that they took all of the medication, 2 indicating that they skipped 5 or fewer days, 3 indicating 6–15 skipped days, 4 indicating 16–30 skipped days, and 5 indicating more than 31 skipped days. Check-up represents whether patients underwent the following tests within 2 years: chest X-ray, mammography, abdominal sonography, dual-energy X-ray absorptiometry, electrocardiogram, echocardiogram, computed tomography, magnetic resonance imaging, positron emission tomography, endoscopy, fluoroscopy, and others. Alternative medicine within a year was scored as 1 for herbal medication, 2 for acupuncture, 3 for moxibustion or cupping therapy, 4 for pharmaco-acupuncture, 5 for bee venom acupuncture, and 6 for others. We also recorded the use of the following supplements within a year: glucosamine, chondroitin, omega-3, vitamins, calcium, iron, ginseng, and others.

Socioeconomic determinants included educational attainment and annual household income. Income was recorded in Korean Won and dichotomized as ≤ 2,000,000 Won or > 2,000,000 Won for data analysis (1 USD = 1,094 Won, 1,000,000 Won = 914 USD in 2013 currency). Education was recorded in years and dichotomized as < 10 years or ≥ 10 years for data analysis.

The KORONA protocol was approved by the Institutional Review Boards of Hanyang University Hospital and all participating hospitals. Informed consent was obtained from all patients before registration. The study protocol complied with the Declaration of Helsinki and was fully approved by the Institutional Review Board of Seoul National University Bundang Hospital (IRB number: X 1812-510-906).

The baseline characteristics of patients who had ever used biologics before enrollment (ever-users) and those who had never used biologics during the study period (never-users) were analyzed. The categorical and continuous variables are presented as frequencies and percentages, and means and standard deviations, respectively. Unadjusted associations between the covariates and the primary outcome were assessed using the chi-square test for categorical data and Student’s t-test or Mann–Whitney U test for continuous data. Similar analyses were conducted after stratifying the data based on income and education. Pairwise deletion was used to handle missing data, including data of monthly household income for 2 participants. This allowed the use of all available data while only removing data points missing from any analyses. Multivariable logistic regression analysis was performed to identify the factors associated with biologics use after adjusting for income, education, healthcare habits, and comorbidities. To determine additive risks resulting from the interaction between education and income, we determined Relative Excess Risk due to Interaction (RERI), attributable proportion (AP) due to interaction, and synergistic index (SI) [18]. In brief, RERI quantifies the degree of additive interaction between two risk factors. If an interaction between risk factor A and risk factor B is present, the combined effect of A and B is greater or smaller than the sum of the individual effects of A and B. A RERI value of 0 indicates no interaction or exact additivity, a RERI value of > 0 indicates positive interaction or more than additivity, and a RERI value of < 0 indicates a negative interaction. AP represents the proportion of disease in the doubly exposed group that is attributable to the interaction. The SI was computed as the ratio between the combined effect and the sum of the individual effects. When an additive interaction effect is absent, AP equals 0 and SI equals 1.

To comprehensively analyze the underlying mechanism, we employed structured equation modeling (SEM) to simultaneously examine multiple factors. SEM allows for the consideration of both the independent and dependent effects of these factors. The measured items for latent variables are presented in Supplementary Table 1. For example, the latent variable of SES included observed variables such as income, education, insurance, and marital status. Our developed model included a direct pathway between SES and biologics use, as well as indirect pathways through latent variables, such as comorbidity, lifestyle, and healthcare habits. Goodness-of-fit indices, including chi-square, comparative fit index (CFI), incremental fit index (IFI), Tucker–Lewis coefficient (TLI), normed fit index (NFI), and relative fit index (RFI), were used to assess the model’s fit. p < 0.05 was considered to indicate statistical significance. All statistical analyses were conducted using R (version 4.1.0, R Foundation for Statistical Computing Platform) or IBM^® SPSS^® Amos™, version 23 statistical software (IBM Corp., Armonk, NY, USA).

We investigated the mechanisms underlying the influence of SES on biologics use. Table 2 presents the RA patients categorized based on income and education levels and a description of their insurance coverages. Comparative analyses of biologics usage and the time to initiation of biologics were conducted. We also analyzed disease activity and healthcare utilization habits. Regarding biologics use, the percentage of ever-users was statistically significantly lower in both the low-income and low-education groups than in the high-income and high-education counterparts (6.8% vs. 8.6%, p = 0.014, 6.9% vs. 8.5%, p = 0.04, Supplementary Fig. 1). Although the time to initiate biologics, which is measured as the interval between the date of RA diagnosis and the first biologics treatment, exhibited no statistically significant differences regardless of income or education levels.

RA patients with lower income or lower education exhibited higher levels of disease activity, poorer physical and mental health, and a reduced quality of life compared with their higher income or higher education counterparts (Table 2, Supplementary Table 3). Specifically, RA patients with lower income had higher baseline disease activity (DAS 28: 3.6 ± 1.7 vs. 3.4 ± 1.5, p = 0.001), lower EQ-5D scores, higher HAQ scores, and greater physician assessment scores than those with higher incomes. Similarly, disease activity was higher in RA patients with lower education than in those with higher education. Overall, despite the trend of higher disease activity in patients with lower income or lower education, we paradoxically found lower biologics use in those groups.

We further explored the relationship between income and education with healthcare utilization and health behaviors. There were no significant differences in alcohol consumption, smoking habits, or exercise frequency based on income and education. However, in terms of medication compliance and skipped medication, individuals in the low-income, low-education group showed lower scores compared to their counterparts, which indicates better medication adherence. Moreover, the consumption of supplementary healthcare products incurring additional costs was found to be higher among those in the high-income, high-education group. Moreover, the low-education and low-income patients were more likely to have higher hospitalization rates, and patients with lower education levels were more inclined to undergo health check-ups, indicating that the NHI system is effective in providing healthcare to low SES patients in Korea.

Furthermore, we analyzed comorbidities based on income and education levels. Patients with low income and low education had a higher prevalence of heart failure, hypertension, asthma, hepatitis, diabetes, bone fractures, and osteoporosis (Supplementary Table 4).

As presented in Model 1 of Table 3, univariate logistic regression analysis revealed associations between an increase in biologics use and factors such as age, low income, low education, and positive rheumatoid factor. Conversely, alcohol consumption and smoking were associated with a decrease in biologics use. Rheumatoid factor positivity and smoking did not reach statistical significance in the multivariate analysis.

Since comorbidity itself may directly affect biologics use, a multivariable logistic regression analysis was conducted to identify biologics use-associated factors while adjusting for comorbidities (Models 2 and 3 of Table 3). Instead of including all comorbidities in the model, we used the age-adjusted Charlson comorbidity index (CCI) [19]. Due to the correlation between lower income and low education, we separately assessed regression models with income and education to eliminate the collinearity effect. Following the adjustment for all covariates, older age (OR = 0.98, 95% CI: 0.97–0.99, p < 0.001 in both models), low education (OR = 0.77, 95% CI: 0.65–0.98, p = 0.032) in the education model, and low income (OR = 0.76, 95% CI: 0.60–0.96, p = 0.021) in the income model were significant.

While both low income and low education individually showed associations with lower biologics utilization, we further investigated the additive risks resulting from their interaction by using RERI (Supplementary Table 5). Income and education exhibited a significant additive interaction for biologics use. Notably, among low-income patients, biologics use increased by 11% if they had high education. Similarly, among low-education patients, biologics use increased by 9% if they had a high income. When both high income and high education were present (high-income, high-education group, biologics use increased significantly by 68%, indicating that these patients had 68% higher odds of using biologics than those with both low income and low education. This suggests a disparity in biologics use based on education and income. RERI was 0.98, indicating a positive additive interaction of income and education. Furthermore, AP was 0.92 (95% confidence interval [CI]: 0.62–1.22), indicating that 92% of the odds ratio (OR) for biologics use in patients with both high income and high education could be attributed to the additive interaction. The OR for biologics utilization in individuals with high income and high education was 3.27 (SI, 3.27, 95% CI: 0.19–57.09) times higher than the sum of the risks in the patients with low-income or low-education alone. This indicates a synergistic effect of high education and high income, which increased biologics use 3.3-fold when both were present.

Supplementary Figure 2 presents the structural and measurement results of the SEM model. Despite a chi-square value for the SEM model fit of 53.507 with 4 degrees of freedom, which does not support a good model fit, the p value for probability was considerably substantial (> 0.05), indicating support for the proposition that the overall model fitted the data. The other standardized fit indices demonstrated that the model had a satisfactory fit (CFI = 0.999, IFI = 0.999, TLI = 0.995, RFI = 0.994, and NFI = 0.999).

On examining the direct structural effects of the variables (Table 4), several significant findings were noted. First, SES significantly and negatively affected comorbidity (β = −0.054, p < 0.001, critical ratio = −17.248), suggesting that patients with higher SES have lower comorbidity rates. Second, both comorbidity (β = 0.016, p < 0.001, critical ratio = 5.731) and SES (β = 0.032, p < 0.001, critical ratio = 4.830) positively affected lifestyle and healthcare habits, suggesting that patients with comorbidities and those with higher SES show more favorable lifestyle and healthcare behaviors. Third, disease activity had a positive effect on the use of biologics (β = 0.001, p = 0.009, critical ratio = 2.600), whereas age exhibited a negative effect on biologics use (β = 0.000, p = 0.027, critical ratio = −2.211). SES had a positive effect on biologics use (β = 0.034, p = 0.007, critical ratio = 4.603). These findings collectively suggest that patients with higher disease activity, younger age, and higher SES are more likely to use biologics for their RA treatment. The mediation analysis revealed that age and disease activity partially mediates the association between SES and biologics use. However, no mediating effect of comorbidities, lifestyle, or healthcare habits was noted in the linkage between SES and biologics. In the pathway analyses, several factors were identified as significant predictors of biologics use (Supplementary Table 6 [H117-153]). Alcohol consumption, smoking, and the use of alternative medicine significantly and negatively affected biologics use. Additionally, older age and comorbidities measured on the basis of age-adjusted CCI were associated with a lower likelihood of biologics use. By contrast, higher disease activity, longer disease duration, hospitalization, good medication compliance, higher income, and higher education significantly and positively affected biologics use. These findings highlight the complex interplay between various factors that influence biologics use.