Study population
Information on psychosocial conditions was retrieved from the Epidemiological Investigation of Rheumatoid Arthritis (EIRA) study. EIRA is a population-based case-control study covering the middle and southern parts of Sweden and includes cases of RA diagnosed by rheumatology specialists within 1 year from diagnosis that meet the American College of Rheumatology (ACR) 1987 and/or 2010 European League Against Rheumatism (EULAR)/ACR criteria for RA [1, 14]. The mean time from first symptom to diagnosis was 10 months. During 1996–2015, 3724 cases and 5935 controls participated in the study. As described in detail elsewhere, the study participants answered a detailed questionnaire regarding their socioeconomic conditions and lifestyle habits [7]. Incomplete responses from the participants were followed up through telephone interviews. The response rate to the study questionnaire was 93% for cases.
Exposures
An index on the level of social support outside work at the time of RA diagnosis was calculated based on four questions developed by Henderson [15] and Undén and Orth-Gomer [16], with a score range of 4 to 20. Similarly, an index for the level of decision latitude at work was based on six questions (Supplementary Tables S1 and S2) developed by Karasek and Theorell, which focused on influence on the working situation, demands on occupational skills, and possibility of further education in the workplace, with a score range of 6 to 24 [9]. Questions on decision latitude at work were only included during the first decade of the EIRA study (part 1, 1996–2006, 1998 cases); furthermore, only individuals active on the labour market were asked to respond to these questions. Other questions, including questions on social support, were included during a longer EIRA study period (1996–2015, 3724 cases). Other environmental and lifestyle factors were also captured at diagnosis and classified for the present analyses as follows: university degree (yes/no), smoking habits (ever smoking, yes/no), alcohol habits (ever drinking, yes/no), and symptom duration at inclusion (categorized in quartiles). Rheumatoid factor (RF) status was determined using standard procedures at each clinic and of anti-citrullinated protein antibody (ACPA) positivity as the presence of anti-citrullinated peptide (anti-CCP) antibodies as assessed by standard enzyme-linked immunosorbent assay (anti-CCP2 assay, Immunoscan-RA Mark 2 ELISA test; Euro-Diagnostica, Malmö, Sweden).
Outcome
Information on the outcome, RA disease activity during follow-up, was retrieved from the Swedish Rheumatology Quality register (SRQ), a clinical quality register used to collect data from follow-up visits during routine care. Patient characteristics are registered at diagnosis, and information on disease activity, patient-reported outcome measures (PROMs), and treatment is registered at diagnosis and subsequent follow-up visits. The national coverage of patients with RA in the register is around 86% [17]. Information from SRQ was available until 2020, allowing a follow-up time of 60 months from diagnosis for all patients in the study.
Information on disease activity, defined according to the DAS28-CRP and its components, was captured from the SRQ from calendar years 1996 to 2020 at visits 3 (±2) months, 12 (±3) months, and 60 (±12) months after inclusion. Since each individual might have more than one visit within each timespan, we selected the visit occurring closest to the pre-set time.
Remission at each time point was defined as a DAS28-CRP ≤ 2.4 at the recorded visit in SRQ [18]. The patients’ assessments of pain (VAS pain) were also retrieved from the SRQ and unacceptable pain was defined as a VAS pain >40 [19].
Statistical analyses
We categorized the main exposures using sex-specific quartiles of the scores among the controls to define the cut-off levels (frequency tables available in Supplementary Tables S3-S6) where the cut-off for the lowest quartiles identified the exposures as compared to the remaining three quartiles. Data from controls were used to avoid possible influence from the RA disease.
Baseline characteristics are presented as medians and interquartile range (IQR).
Logistic regression was used to estimate whether low social support or low decision latitude at work was associated with DAS28-CRP remission at 3, 12, and 60 months, first after adjusting only for age and sex and subsequently with additional adjusting for smoking habits, alcohol habits, symptom duration, and educational level. Additional analyses investigating a possible association between the two main exposures and DAS28-CRP remission were made where the two main exposures were treated as continuous variables instead of dichotomous variables, both in a crude model and in a fully adjusted model.
In a sensitivity analysis, the odds ratio (OR) for remission among patients in the lowest quartiles was compared to the highest quartiles of the exposures. An additional sensitivity analysis was performed, comparing the odds for remission among patients with both low social support and low decision latitude at work as compared with patients without either of these exposures.
Additional analyses for the association between social support and remission were conducted in subgroups based on sex, ACPA status, smoking habits, educational level, alcohol habits, and symptom duration.
The medians of the separate components of the DAS28-CRP as well as of VAS pain were compared between patients with low and not low social support and subsequently for patients with low and not low decision latitude at work. In complementary analyses, the OR for having VAS pain above median for each time point for individuals reporting low social support as compared to not low social support was investigated. The association was tested using logistic regression in a model adjusted for age and sex and further in a fully adjusted multivariable model. Further, the OR for having VAS pain above median for each time point was investigated for individuals reporting low decision latitude at work as compared to not low decision latitude at work, with the same method. Subsequently, in models for the OR of having unacceptable pain (VAS pain >40 mm) at each time point, social support and decision latitude at work, respectively, were investigated, first in a model adjusted for age and sex and further in a fully adjusted model.
The frequencies of the exposures in individuals lost to follow-up and individuals still in the study after 5 years were also compared.
As the rate of missingness for the exposure was low, complete case analyses were performed. Individuals lacking data on social support and individuals without data on decision latitude at work mainly due to incomplete questionnaires were excluded. Missingness for the covariates did not exceed 5%.
Statistical analyses were performed using SAS 9.4 (SAS Institute, Cary, NC, USA). All tests were two-sided and the significance level was set to 0.05. Two-sided p-values were calculated using median two-sample tests.