Identification of modifiable work-related factors that influence the risk of work disability in rheumatoid arthritis




To define work-related factors associated with increased risk of work disability (WD) in people with rheumatoid arthritis (RA).


Questionnaires were mailed to all RA patients who used a province-wide arthritis treatment program between 1991 and 1998 (n = 1,824). The association between risk factors and WD (defined as no paid work due to RA for at least 6 months) was assessed using multiple logistic regression analysis, controlling for significant sociodemographic and disease-related variables.


Of the original 1,824 patients, 581 were eligible and responded to the questionnaire. Work survival analysis revealed a steady rate of WD starting early, with 7.5%, 18%, and 27% work disabled at 1, 5, and 10 years, respectively. Significant determinants in multiple logistic regression were physical function (Health Assessment Questionnaire), pain (visual analog scale), and 6 work-related factors: self employment, workstation modification, work importance, family support toward employment, commuting difficulty, and comfort telling coworkers about RA.


Work disability occurs early in RA. Novel work-related factors were identified, which are potentially modifiable, to help RA patients stay employed.


Work disability (WD) is a common consequence of rheumatoid arthritis (RA). This is particularly significant because RA affects individuals during their prime working years. Reported rates of WD range from 32% to 50% within 10 years of disease onset, increasing to 50–90% after 30 years of disease (1–7). Longitudinal studies consistently show that the toll of WD starts early after disease onset (2–9).

Work disability profoundly affects the quality of life of patients and their families, and has major financial consequences for the individual and society at large. A number of studies of RA have found indirect costs from lost productivity to exceed the direct medical costs of providing health care by up to 2-fold (10–16). In a systematic review of cost studies in RA (17), the mean annual indirect cost was calculated to be $9,744/year/patient (calculated in 1998 US dollars) (12–14, 16, 18, 19). To reduce the burden of WD from RA, a good understanding of the determinants of WD is necessary.

A number of studies have identified determinants of WD in patients with RA (2–6, 20–34). Disease-related determinants have been extensively studied. In almost all studies, a measure of physical function, with or without a measure of pain, were identified as significant determinants of WD, but other clinical features added little or nothing to WD models (22). However, disease-related determinants overall have a relatively small effect on WD compared with sociodemographic and work-related factors (30). Sociodemographic factors, such as age, sex, education, and marital status, despite their importance as determinants of WD, offer little opportunity for intervention. Of the work-related factors, the influence of low physical demand and high job autonomy has been demonstrated in a number of studies (1–6, 21, 24–26, 28, 30, 31). Other work characteristics have not been studied extensively, despite their importance and their potential for modification.

The purpose of the present study was to provide a comprehensive assessment of the work-related factors influencing the risk of WD in RA, after controlling for significant sociodemographic and disease-related factors. The unique feature of this study was the broad scope of work factors included, allowing the identification of novel work-related determinants of WD. As a secondary objective, the association between risk of WD and social support and 2 global psychological characteristics (self efficacy and sense of coherence) was assessed.


We performed a cross-sectional study of RA patients assessing the association between work disability due to RA at the time of survey and potential explanatory factors.

Study sample

A survey was mailed to all RA patients, age 18–65 years, who had used any of the services offered by the treatment program of The Arthritis Society, British Columbia and Yukon division, between 1981 and 1998. Patients are registered in a database with a physician-derived diagnostic code. The treatment program offers a wide range of services including physiotherapy, occupational therapy, social work, nursing, patient education, exercise classes, and support groups, as well as drug monitoring clinics and evaluations by rheumatologists. Access to these services, except for educational activities, requires the referral of a physician. Most referring physicians are rheumatologists or family physicians. Services are offered in the 4 largest cities in British Columbia, as well as in 22 outlying areas through traveling clinics. The wide range of services and the broad geographic distribution ensured that the sample represents RA patients of varying disease severity, duration, and practice settings. Patients were invited to participate if they had RA and had worked for pay at or since the onset of their RA. To verify the database-derived diagnosis, subjects were asked if they had RA. The medical records of all subjects who replied “no” or “unsure” were reviewed, and subjects not meeting the 1987 revised American College of Rheumatology (formerly American Rheumatism Association) diagnostic criteria for RA were excluded (35). To further assess the diagnostic accuracy of RA in the sample, we also reviewed the medical records of 50 randomly selected participants. Two medical records could not be located. In 11 cases, all of whom were referred to the treatment program only for visits to allied health professionals, the available medical records included insufficient information to determine whether or not they fulfilled the criteria. The 37 remaining medical records fulfilled the diagnostic criteria for RA.

Data collection

All data were obtained from a self-report questionnaire that was mailed to the sample population. A second questionnaire was mailed to nonresponders, 2 months after the first mailing. The questionnaire was developed from validated instruments assessing each of the factors, when available, as described below. The questionnaire was pretested with RA patients treated at the Mary Pack Arthritis Treatment Program in Vancouver. It could be completed in 20–30 minutes and was found to be clear and easy to fill out.

Assessment of variables

The outcome variable, work disability, was determined by asking subjects about their work status at the time of the survey and the reason for stopping work. Work disability was defined as not having performed any paid work for at least 6 months due to RA. This duration was selected because it is the criterion for long-term work disability used by most private insurance companies (36). This definition assesses complete WD and thus does not include reduction in work hours to part-time work, or stopping work for reasons other than RA, such as early retirement or lifestyle choices unrelated to RA.

Explanatory variables were grouped under the headings of work-related, sociodemographic, disease-related, and psychosocial factors. Information on the factors was collected at the time of survey for employed individuals. Unemployed individuals were asked to respond based on their status at the time of work loss, except for a few factors for which recall to the time of work loss was not expected to be accurate (self-reported joint count, physical function, self efficacy, and sense of coherence, measured as described below).

Work-related factors were the main focus of our study. Physical demand was assessed using the corresponding subscale of the 1978 US Social Security Administration study of disabled and nondisabled adults (SDA) (37, 38) and with the physical effort scale of the Job Content Questionnaire (39). Job autonomy was assessed using the questionnaire developed by Yelin et al (30), which has been used in other studies of WD in RA (2, 21, 28, 29), and by the decision latitude scale of the Job Content Questionnaire (39). Participants were asked whether they were self employed. Commuting difficulty was assessed using the method of Allaire et al (21), who had identified commuting difficulty as a significant risk factor for WD.

Having received job accommodations from the employer to help workers perform their job despite arthritis was assessed using the employer help subscale of the SDA, which measures type and amount of help received (37, 38). Also, to specifically assess ergonomic modifications, we designed 4 questions asking subjects whether modifications had been performed for their RA in the following 4 categories: work equipment, workstation, physical environment, and organization of work. Examples were provided to illustrate each category.

Attitudes toward work were assessed using an 8-item scale from the SDA (37, 38) that characterizes what motivates individuals to work by asking subjects how much they agree with statements exploring reasons why people work. The importance of working was assessed using the work importance subscale of the Preliminary Diagnostic Questionnaire (PDQ), which has been validated as an independent subscale (40). The PDQ was designed to measure the vocational potential of work-disabled clients applying for vocational rehabilitation services. The work importance subscale assesses how important the concept of work is to the individual, as well as whether they receive support for working from family and friends. Family support toward continued employment was also assessed by asking subjects to rate how much their family or partner supported their decision to continue working. This question was added because it quantified better the amount of support received from family members or partners for continuing to work, than did the question about support from family included in the PDQ scale.

Psychosocial characteristics of the job were evaluated using the demand-control model of job strain by Karasek and Theorell (41), which measures decision latitude (skill discretion and decision authority), work demand (psychological load and physical effort), social support at work from coworkers and supervisors, and job satisfaction using a validated instrument (the Job Content Questionnaire) (39).

Psychological and social variables were assessed as a secondary objective of the study. The following variables were measured: perceived self efficacy, using the Arthritis Self-Efficacy Scale (42); sense of coherence, using the short form of the Sense of Coherence scale (43); and perceived general social support, using the social support subscale of the Arthritis Impact Measurement Scale (44).

Sociodemographic and disease-related factors previously found to be important were measured to be included as covariates in our predictive model of WD. Age, sex, ethnic group, immigration status, marital status, number of dependents, education level, family and personal income, availability of private disability insurance, and replacement rates were assessed in the self-administered questionnaire. All income amounts were collected using categories of income with $10,000 strata. Income was adjusted to dollar value at the time of the survey (1998). The replacement rate, a measure of economic incentive to remain employed with a physical disability, was calculated by asking people what percentage of their take-home pay they could replace through work disability payments if they stopped working. Although the accuracy of people's answer may not be perfect, the information obtained reflected their perception. Only 9% replied they did not know. The ratio of personal to family income was also measured, as another economic incentive to remain employed. Education level was assessed using categories representing the highest level of education achieved and by asking the total number of years of education. Disease-related factors measured included RA duration, age at onset, comorbid conditions, joint tenderness (using a validated self-report measure of joint count from the Rapid Assessment of Disease Activity in Rheumatology instrument) (45), level of pain, global assessment of disease activity (using a visual analog scale), and physical function (using the Health Assessment Questionnaire [HAQ]) (46).

Statistical analysis

The interrelationship between explanatory variables was measured using Pearson correlation coefficients. Univariate association between explanatory variables and outcome (WD from RA) was measured using chi-square test for categorical variables and t-test for continuous variables. All numerical scales were included as continuous variables.

Multivariate analysis was conducted using multiple stepwise logistic regression analysis, controlling for significant sociodemographic and disease-related variables, using a forward stepwise method with backward elimination. In outcome variable assessment, the dependent variable was work disability from RA, as defined above. Individuals who stopped work for reasons other than RA, such as retirement not due to RA or choosing to stop work to care for children, were included in the non–work-disabled group because they did not meet the definition for WD. Explanatory variables with a level of significance of P < 0.2 on univariate analyses were included in the multivariate analysis. Multivariate analyses were first conducted for each group of variables (work-related, sociodemographic, disease-related, psychosocial). Variables from each group with a level of significance of P < 0.2 and primary variables of interest were included in the final model. Variables in the final model were selected using a P value < 0.05. All analyses were conducted using SPSS software (Chicago, IL).

Duration of employment after RA diagnosis was analyzed using survival analysis. For this, time from RA diagnosis to work cessation was measured. The event analyzed was WD from RA, and work cessation for other reasons were censored events.


Questionnaires were mailed to 1,824 RA patients, of whom 951 (52%) responded. Nonresponders did not differ from participants in terms of age or sex. Of the 951 responders, 130 were not eligible because they had not worked while having RA and 240 declined to participate. Therefore, 581 patients formed the study sample. Subjects who declined to participate were asked to provide limited data. The nonparticipants did not differ from participants in terms of sex, age at onset of RA, marital status, or duration of work while having arthritis. However, at the time of the survey, they were significantly older (mean age 54 versus 48 years) and less likely to be still working (28% versus 57%). Participants were mostly women (79%), had a mean age of 48 years, mean duration of RA of 9.8 years, and a mean HAQ score of 1.1. Disease and sociodemographic characteristics of work-disabled and non–work-disabled individuals are presented in Table 1. Work disabled participants had stopped working, on average, 5.2 years prior to the survey.

Table 1. Sociodemographic and disease characteristics of work-disabled and non–work-disabled subjects (n = 581)*
 Work disabled (n = 203)Not work disabled (n = 378)P
  • *

    RA = rheumatoid arthritis; VAS = visual analog scale; RADAR = Rapid Assessment of Disease Activity in Rheumatology; HAQ = Health Assessment Questionnaire.

  • CND = Canadian dollar, 1 CND = 0.67 American dollar in 1998.

  • Replacement rate = percentage of take-home pay one could replace through work disability payments.

Sociodemographic characteristics   
 Women, %83780.19
 Age, mean, years46.748.10.08
 Education, mean, years12.614.3< 0.001
 Marital status  0.12
  Single, %1211 
  Married, %6673 
  Divorced/widowed, %2316 
 Immigrated to Canada, %33240.03
 Ethnicity, % white79840.19
 Household income, mean, CND44,50061,000< 0.001
 Personal income, mean, CND25,60036,400< 0.001
 Disability insurance, %6948< 0.001
 Replacement rate, %32260.057
Disease characteristics   
 Age at RA onset, mean, years37.837.60.87
 RA duration, mean, years910.2< 0.001
 Pain, mean VAS (0–10 cm)8.34.3< 0.001
 Disease activity, mean VAS (0–10 cm)8.24.5< 0.001
 Mean tender joint score, RADAR (0–60)2816< 0.001
 Physical function, mean HAQ (0–3)1.60.8< 0.001
 Comorbidities, % with ≥146330.002

We found that 203 patients (35%) were work disabled at the time of survey. Work survival analysis by time since RA diagnosis revealed that work disability occurred early and continued at a steady rate throughout the course of RA (Figure 1). We found that 7.5% of the study sample was work disabled 1 year after RA diagnosis; 9.4% were work disabled after 2 years; 18% after 5 years, and 27% after 10 years.

Figure 1.

Cumulative survival in employment according to years since diagnosis of rheumatoid arthritis (RA). Median survival time was 23 years (95% confidence interval 20.8–25.2). Survival (standard error) at 1, 2, 5, 10, and 15 years were as follows: 92.5% (1.1) were still employed 1 year after diagnosis; 90.5% (1.2) at 2 years; 82% (1.7) at 5 years; 73% (2.2) at 10 years; and 63% (2.6) at 15 years.

Most of the work-related factors studied were significant univariate predictors of WD (Table 2). People who were self employed, who rated their jobs as allowing greater autonomy, who had received ergonomic modification of their workstation or their equipment to adapt it to their arthritis, people for whom work was more important, and people whose family was more supportive of them continuing to work were less likely to be work disabled. People whose job was more physically demanding, who had greater difficulty commuting to and from work, who rated their work conditions more adversely (i.e., less skill discretion or decision latitude, greater psychological load or physical effort, less social support from coworkers, and less job satisfaction) and who had disability insurance were more likely to be work disabled. Individuals who had received more accommodations from their employer to help them with their job (employer help scale) or who had used vocational services were more likely to be work disabled. Use of other services aimed at keeping people employed (getting extra job training or additional education, having an assessment of their work by an occupational therapist) did not differ significantly between groups. Work-disabled subjects also did not differ in their attitudes toward work (motivating factors for work), social support from supervisors, or labor union membership.

Table 2. Univariate comparison of work-related factors in work-disabled and non–work-disabled individuals*
Work-related factorsWork disabled (n = 203)Not work disabled (n = 378)P
  • *

    RA = rheumatoid arthritis; OT = occupational therapist.

  • Job Content Questionnaire (JCQ) evaluating psychosocial work conditions according to the Karasek model.

Self employed, %8190.001
Physical demand (mean of 10 items; range 1–3; 3 = greater)2.171.93< 0.001
Job autonomy (mean of 9 items; range 1–3, 3 = less)2.021.760.002
Commuting difficulty (mean of 3 items; range 0–6; 6 = more)1.90.6< 0.001
Adverse work conditions, JCQ, mean ± SD (15 items; range 1–4; 4 = more adverse)   
 Overall score (mean of 15 items)2.39 ± 0.42.19 ± 0.35< 0.001
 Decision authority subscale2.22 ± 0.81.85 ± 0.7< 0.001
 Skill discretion subscale2.02 ± 0.51.94 ± 0.50.07
 Psychological load subscale2.81 ± 0.582.67 ± 0.580.005
 Physical effort subscale2.89 ± 0.952.35 ± 0.92< 0.001
Job satisfaction, JCQ, mean ± SD (mean of 4 items; range 1–5; 5 = less)2.15 ± 1.051.86 ± 0.860.001
Social support at work, JCQ, mean ± SD   
(mean of 8 items; range 1–4; 4 = less support)   
 Supervisor support2.08 ± 0.731.99 ± 0.650.15
 Coworker support1.94 ± 0.541.78 ± 0.490.001
Told their employer/supervisor about RA, %79710.03
Comfort telling others about RA, mean ± SD (range 1–7; 7 = less comfort)   
 Comfort telling employer/supervisor3.65 ± 2.33.52 ± 2.250.55
 Comfort telling coworkers3.25 ± 2.23.35 ± 2.20.62
Importance of work to person with RA, mean ± SD (mean of 8 items; range 1–4; 4 = more)3.16 ± 0.523.55 ± 0.41< 0.001
Attitudes toward work (motivation), mean ± SD (mean of 8 items; range 1–5; 5 = more)3.01 ± 0.523.09 ± 0.530.12
Family support toward employment, mean (range 1–7; 7 = more)5.36.3< 0.001
Employer help (8 items; yes/no)   
 Number of help items received1.361.050.042
 Received at least one help item, %36290.088
 Got someone to help do job, %27150.001
 Allowed more breaks/rests, %29200.038
 Shortened work day, %23150.02
 Changed time to start and end work, %17140.60
 Changed job to something one can do, %15140.60
 Helped learn new skills, %1590.037
 Got special equipment, %8180.002
 Arranged special transportation, %111
Ergonomic modifications, % (4 items; yes/no)   
 Modified work equipment10.917.50.02
 Modified workstation14240.007
 Modified work organization20190.90
 Modified physical environment680.59
Use of other services to help workers, %   
 Vocational services1550.001
 Work assessment by OT28160.50
 Extra job training1612.50.28
 Obtained further education18190.80
Labor union membership, % member40370.52

Work-disabled individuals were more likely to have reduced or changed their work specifically because of arthritis. They were more likely to have reduced the number of hours of work per week (P = 0.002), to have changed the type of work they did (P < 0.001), to have changed jobs (P < 0.001), or to have changed employer (P = 0.01). These are likely early markers of future work disability. Interestingly, the 2 groups didn't differ significantly in temporary discontinuations of work, defined as having temporarily stopped work for >2 months because of arthritis.

In terms of the psychological and social variables assessed as a secondary objective of the study, individuals with greater perceived self efficacy (for pain, symptoms, and function, as well as overall score) and those with a greater sense of coherence were less likely to be work disabled (Table 3). Work-disabled individuals did not differ in the amount of perceived general social support.

Table 3. Secondary analysis: univariate comparison of psychological and social variables in work-disabled and non–work-disabled individuals*
VariablesWork disabled (n = 203)Not work disabled (n = 378)P
  • *

    Values represent mean ± standard deviation.

Self efficacy (range 1–10; 10 = more)   
 Overall score (mean of 3 subscales)5.0 ± 1.86.4 ± 2.0< 0.001
 Pain subscale (mean of 5 items)4.8 ± 2.15.8 ± 2.2< 0.001
 Symptoms subscale (mean of 5 items)5.3 ± 1.96.2 ± 2.1< 0.001
 Function subscale (mean of 9 items)4.9 ± 2.27.2 ± 2.5< 0.001
Sense of coherence (mean of 13 items; range 1–7; 7 = better)5.0 ± 1.15.3 ± 1.00.003
General social support (mean of 4 items; range 1–5; 5 = less support)5.07 ± 2.35.02 ± 1.90.80

In multiple stepwise logistic regression, physical function (as measured by HAQ) and pain (as measured by visual analog scale) were the only sociodemographic or disease-related variables selected as significant predictors of WD. The following 6 work-related factors entered the model as significant predictors of WD (Table 4). Self-employed individuals were 5 times less likely to be work disabled (odds ratio [OR] 0.21, P = 0.001). Individuals whose workstation had been ergonomically modified were 2.6 times less likely to be work disabled (OR 0.39, P = 0.015). People who reported greater difficulty commuting to and from work were more likely to be work disabled (P = 0.015), as were people who were more comfortable telling coworkers about their RA (P = 0.001). The likelihood of being work disabled decreased as the importance of work to the individual increased (P = 0.014) and as support from family toward continued employment increased (P= 0.029). People who rated their job as “requiring lots of physical effort” were more likely to be work disabled (OR 1.57), however this did not reach statistical significance (P = 0.13). There was a trend for individuals with disability insurance coverage to be work disabled (OR 1.9, P = 0.06). No interaction was demonstrated between any of the variables included in the final logistic regression model, or with other key variables, such as physical demand, job autonomy, or disability insurance. Self employment was inversely correlated with having disability insurance (r = −0.216, P < 0.001) and with replacement rate (r = −0.258, P < 0.001), however no interaction was demonstrated between these variables, and forcing disability insurance into the model did not alter the relationship between self employment and the risk of WD. Entering continuous variables as categorical did not alter the model. To test the sensitivity of the impact of the predictors to the coding of the dependent variable, we repeated the multivariate regression analysis using “working” versus “not working for any reason” as the dependent variable. We found no significant change in the OR of the explanatory variables selected in the final model.

Table 4. Multivariate logistic regression analysis modeling the odds of work disability*
VariableOR95% CIP
  • *

    OR = odds ratio; 95% CI = 95% confidence interval; VAS = visual analog scale; HAQ = Health Assessment Questionnaire; RA = rheumatoid arthritis.

  • OR expressed as the OR for every 1 unit increase in variable measured for continuous variables; and as the OR of yes/no for binary variables.

Pain VAS (0–10 cm)1.71.5–1.95< 0.001
Physical function, HAQ (0–3)1.961.2–3.30.01
Self employed (yes)0.20.08–0.50.001
Ergonomic modification of work station (yes)0.40.2–0.80.015
Commuting difficulty (0–6)1.31.05–1.60.015
Comfort telling coworkers about RA (1–7)0.80.7–0.90.001
Importance of work to person with RA (1–4)0.450.2–0.850.014
Family support toward work (1–7)0.80.6–0.980.029

Because a number of previous studies have shown that greater job autonomy is associated with a reduced risk of WD, we looked at the effect of job autonomy using both the decision authority subscale of the Karasek model (39) and the scale developed by Yelin et al (30). When using summary scores for these scales, job autonomy was not a significant predictor in the multivariate analysis. When using individual items from the scales, only 1 of the 9 items from Yelin's scale was significant. Compared with those who were unable to decide when to take a day off from work, those who could decide with (OR 0.15) and without (OR 0.27) asking permission from a supervisor, were less likely to be work disabled (P = 0.01). In contrast to previous studies (2, 28), control over pace of work and over activities were not significant.

In the multivariate analysis, having received accommodations from the employer, as assessed by the employer help scale, was not significantly associated with WD. To allow comparison of results with the study of Allaire et al (21), we divided the items into 2 subscales, according to whether the type of accommodation provided by the employer resulted in reduced work productivity. Allaire et al suggested that help that reduces productivity may have the unwanted effect of increasing the risk of WD. We found no effect for the subscale of items that do not reduce work productivity (i.e., having received accommodations such as learning new skills, changing the job to something the worker can do, changing arrival and departure time, providing special transportation, and obtaining special equipment). However, individuals who had received more of the types of accommodations that result in reduced productivity (such as getting someone to help do the job, getting more breaks and rests, or shortening the work day), were more likely to be work disabled (P = 0.043).

In terms of the psychological and social variables, assessed as a secondary objective of the study, perceived self efficacy, sense of coherence, and general social support were not selected as significant determinants of WD in the multivariate regression model.


This study of determinants of WD in RA provides a comprehensive assessment of work-related factors. In addition to confirming the importance of work-related factors in determining one's risk of becoming work disabled from RA, this study has identified novel and potentially modifiable risk factors. Our results suggest that self-employed individuals are at a lower risk of WD; that adapting the workstation to accommodate for RA reduces the risk of WD; that greater importance of work to the individual and greater support for continued employment from family lowers the risk of WD; and finally, that difficulty commuting increases the risk of WD.

By identifying potentially modifiable factors, this study provides valuable information to vocational counselors and other arthritis health professionals, such as physicians, social workers, and occupational therapists, involved in advising patients about how to remain employed. In addition to trying to reduce the physical demand of the job and to improve job autonomy, as recommended previously from the arthritis WD literature, professionals need also to assess the work environment for ergonomic problems that may need adaptation, address problems with commuting to and from work, pay attention to support from family members, and explore the option of self employment as an alternative form of employment for those who express an interest in it. In our sample, only 8% had seen a vocational counselor, indicating that increased awareness of and referral to vocational services are necessary. This study also provides an argument in support of policies at the public level and at the level of employers for worksite ergonomic modifications and transportation assistance for people with physical disabilities, such as RA.

Our results are consistent with the literature. Although the rates of WD observed in our sample are among the lowest reported, they are similar to those of Wolfe and Hawley, who described WD rates due to RA of 22% at 5 years and 32% at 10 years (5). Other studies report higher WD rates, ranging from 39% to 50% 10 years after disease onset (2–4, 6, 7). The lower rate we observed may reflect the community-based nature of our sample. Furthermore, WD rates are difficult to compare across studies, because of differences in definitions of work disability, differences in the age and sex distribution of the samples and because of differences in social context affecting availability of work disability benefits.

All studies, including ours, that have looked at WD rates from RA over time, have demonstrated that WD starts early after disease onset (2–9). Some have found a particularly rapid decline in workforce participation within the first 2–3 years after onset of arthritis (2–4). We report a steady decline over time starting at disease onset. This emphasizes the need for early intervention. Just as the trends in medical treatment of inflammatory arthritis have changed, the time has come to invert the traditional pyramid for the management of WD.

Of the work-related factors influencing one's risk of WD, physical demand and job autonomy have been the most extensively studied. Greater physical demand is consistently associated with a greater risk of work loss (1–6, 21, 24, 25, 28, 31). Our study confirmed the association but it was not statistically significant in the multivariate model. We speculate this may be due to correlation between physical demand and other characteristics of a job that were selected as significant determinants of WD. Greater physical demand was mildly correlated with greater commuting difficulty, greater pain, less support from family toward continued employment, and less likelihood of having received ergonomic modifications, (Pearson's correlation coefficient varying from 0.13 to 0.25, P < 0.01). However no interactions were found between physical demand and the variables selected in the model.

Low job autonomy has been associated with increased risk of WD in a number of studies (2, 26, 28, 30), but not in some recent ones (21, 24, 29, 33), including ours. It is possible that changes in the labor market or in the overall level of autonomy in jobs over time have reduced the effect of autonomy in the more recent studies. We found the ultimate form of job autonomy, self employment, to be a very strong predictor of WD, with individuals who were not self employed being 5 times more likely to be work disabled. This had been assessed infrequently in previous studies. A protective effect of self employment was noted in only 1 other study (30) and a lack of association found in 2 others (2, 24). Conceptually, self employment offers a number of advantages to workers with RA by providing them a greater degree of flexibility to plan their work day around their arthritis and by allowing them to choose jobs better adapted to the limitations of their disease, including working from home. Difficulty in commuting to work had been identified as a barrier to work in qualitative research (47, 48); its independent effect on risk of WD has now been confirmed in 2 studies, including ours (21).

Our study was the first to specifically look at the effect of ergonomic modifications at work on the risk of WD from RA. We found that individuals whose workstations had been adapted were 2.6 times less likely to be work disabled than those who's workstations had not. Modification of the work equipment was also significant in the univariate analysis, but not in the multivariate regression model. However, there was a moderate degree of correlation between modification of the workstation and of the work equipment (r = 0.42, P < 0.001). Modifications of the work organization and of the physical environment, other components of ergonomic assessments, were not significant. Although not specifically assessing ergonomic modifications, a study by Chorus et al (23) looked at adjusting furniture and tools as 1 item in a scale of accommodations at the workplace to help individuals with RA. They found that individuals who had received at least 1 form of accommodation were 2.5 times less likely to be work disabled (OR 0.4, 95% confidence interval 0.3–0.6). Other accommodations studied by Chorus et al included shortening work hours, slowing pace of work, changing tasks, and being allowed to manage work. This study was the only one to find a protective effect of receiving job accommodations from the employer to help workers perform their job. Our study, and 2 others performed in the US, using the same scale to assess job accommodations (the employer help scale of the SDA), did not find a significant association for the overall effect of the scale, after controlling for disease severity in the multivariate analysis (21, 49). This difference in results may relate to differences in the accessibility of such modifications in the Netherlands and North America due to different social and labor contexts, or may be due to the use of different measurement scales. The proportion of workers having received accommodations was surprisingly low in all 3 studies performed in North America, suggesting that self selection may account for the lack of association. Those experiencing difficulties with work were more likely to have been offered accommodations. Therefore, receiving the accommodations may act as an early marker of decreased ability to work, negating a potential beneficial effect on risk of WD. Alternatively, the accommodations may have been received too late to avert WD.

Our study was also the first to assess the effect of psychosocial work characteristics on the risk of WD in RA. We used the Karasek model, which postulates that high work demand (physical and psychological) and poor control over work (i.e., little ability to make decisions and to develop new skills) is associated with greater job strain. This relationship is accentuated by poor social support at work (from supervisors and coworkers) and poor job satisfaction (41). Greater job strain has been associated with poorer subjective overall health, greater perception of pain, and musculoskeletal symptoms in work-related musculoskeletal diseases (50, 51). We had hypothesized that adverse work conditions and greater job strain would be associated with a greater risk of WD. Although this was found in the univariate analysis, the associations were not significant in the multivariate regression model. A study of WD in mixed musculoskeletal conditions, using data from a US national survey, also failed to find any association between psychosocial work characteristics and risk of WD (52). In that study, job characteristics were also evaluated according to the Karasek model, but were estimated through a matching process from the occupational title rather than measured directly. Coworker support was reported as a significant factor in 1 study (21). The significance of our finding that individuals who are more comfortable telling their coworkers about having RA are more likely to be work disabled is unclear. It may not indicate that telling your coworkers has a negative impact, but rather it may be that comfort telling coworkers is associated with another significant factor that was not measured in our study, or that telling your coworkers is a consequence of having difficulty performing your job and thus an early marker of WD.

Finally, greater importance of work to the individual and greater support for continued employment from family were associated with lower risk of WD. Although helpful in allowing one to evaluate the risk of WD, these factors may not be easily modifiable. Furthermore, one should interpret these associations with caution in view of the cross-sectional nature of the study. It is possible that individuals readjusted the relative importance of work over other spheres of activities in life as a consequence of becoming work disabled, and similarly, that families became less supportive of continued employment as individuals began to experience more difficulties at work. Prospective studies are needed to understand the dynamics.

Some potential limitations of our study deserve comment. The cross-sectional nature of our study precludes the ability to assess the temporality criteria of causality for the associations observed. Data were obtained from self report through a questionnaire, rather than by direct measurement or observation of the work conditions and characteristics. Differences in recall could have occurred between working and work-disabled individuals. The timing of assessment of the factors was different for working and nonworking individuals. Working individuals reported factors at the time of survey, whereas nonworking individuals were asked to recall factors at the time of work cessation. Although we specifically asked people whether they had stopped working due to their RA, attribution of the cause for stopping work can be difficult and imperfect. Given the low number of people who had stopped working for reasons other than RA (n = 52; 9% of the sample), and that the impact of the factors identified was similar when the dependent variable was coded as “working” versus “nonworking,” this likely did not influence the results. Finally, the response rate of the survey was 52%. Differences between responders and nonresponders may limit the generalizability of the results, although we did not observe any sociodemographic differences between responders and nonresponders.

In conclusion, this comprehensive assessment of work-related factors determining the risk of WD has identified novel and potentially modifiable risk factors. Our results suggest that self employment, ergonomic modification of the workstation to accommodate RA, importance of work to the individual, and family support towards continued employment protect individuals from WD; and that difficulty commuting increases the risk of WD. These results emphasize the need for vocational counseling and for implementing policies at the public level and at the level of employers to help people with RA maintain their employment. In addition, our work confirms that WD starts early after RA onset and continues at a steady rate, indicating a need for early intervention.