To assess the association between use of etanercept and employment outcomes among patients with rheumatoid arthritis (RA).
To assess the association between use of etanercept and employment outcomes among patients with rheumatoid arthritis (RA).
In 1999, 497 RA patients of working ages (18–64 years) reported their employment status in the year of diagnosis and as of the study year, in structured telephone interviews. Of these, 238 had been in clinical trials of etanercept and were currently taking that medication, while 259 were members of an observational study and were not taking etanercept. We used regression techniques to estimate whether employment outcomes in 1999 (employed versus not and, among the employed, hours of work per week, weeks of work per year, and hours of work per year) among the 379 of the 497 patients who were employed at the time of diagnosis were associated with etanercept use, with and without adjustment for demographic characteristics, RA status, overall health status, and the nature of the job held at the time of diagnosis.
At the time of diagnosis, 75% of RA patients from the observational study who did not take etanercept and 77% of those who did take the medication were employed. By 1999, among those employed at diagnosis, 55% of the former group and 71% of the latter were employed (difference 16 percentage points). After adjustment for demographics, overall health status, duration of RA, RA status, and occupation and industry, the difference widened to 20 percentage points. Among all who were employed at the time of diagnosis, those from the etanercept clinical trials worked an average of 5.4 more hours per week in 1999; after adjustment, the etanercept group worked 7.4 more hours per week.
Among all persons who were employed at the time of RA diagnosis, having been in the etanercept clinical trials was associated with higher employment rates in 1999 and a greater number of hours per week of work in that year, suggesting that a randomized trial to establish the relationship between treatment and employment outcomes is now warranted.
Numerous individual studies (1–17) and several reviews (18–20) have established that work loss is common among patients with rheumatoid arthritis (RA). These studies span several decades (e.g., refs. 1, 3, 11, and17), early versus late disease (refs. 10, 11, and13–15 for early disease;1–9, 12, and16 for late disease), explanatory models, and clinical (1–7, 9–17) and population-based (8) sampling, but they show almost uniformly that between one-third and two-thirds of patients with RA have stopped working by the time of interview (18). Moreover, prospective studies demonstrate that between one-fourth and one-half of patients with RA will stop working within a decade of disease onset, and between one-half and 90% will stop working before age 65 years, the normal age of retirement (3, 7, 15, 16). Even those with disease of recent onset do not appear to be immune, with rates of work loss within the first decade of disease as high as 40% (10, 11, 13–15). As a result of the high work loss rates among patients with RA, indirect costs, associated with lost work or lost functionality in nonwork activities, appear to be even larger than the direct costs of RA associated with procurement of medical care services (21–31).
Despite the high rate of work loss experienced by patients with RA and the high indirect costs associated with such work loss, relatively few studies have addressed the impact of specific treatments on work loss costs (11). As of this writing, we could find only one published abstract about the association of biologic therapy in particular, and only one article and one abstract evaluating the impact of disease-modifying antirheumatic drug (DMARD) treatment in general, on employment outcomes. In the abstract about the effect of a biologic agent on employment, patients randomized to receive anakinra had ∼2 more “productive” days each month than those receiving placebo (“productive” days were defined as including work or domestic activity) (32). The article about the impact of DMARD therapy reported on a small clinical trial involving 83 patients; it showed that early treatment with a disease-modifying agent was associated with the maintenance of work capacity, but the study did not evaluate actual employment rates (11). Another abstract reported on a study that compared the impact of combinations of DMARDs versus monotherapy and showed that combination therapy resulted in fewer work disability days per person-year (12 versus 32), lower rates of work loss (20% versus 30%), and lower levels of disability benefits (€305 versus €897 per patient-year) (33).
The present study was designed to investigate the association between receipt of etanercept therapy and employment outcomes. Specifically, we ascertained the work history of RA patients who were and those who were not participants in etanercept clinical trials, and then, for those who were employed at the time of disease diagnosis, we compared the 2 groups for the following employment outcomes as of the study year (1999): employed versus not employed, quantity of employment in terms of hours of work per week among all who were employed at the time of disease onset, and, among those who remained employed in 1999, quantity in terms of hours of work per week, weeks of work per year, and hours of work per year.
Since the analyses concerned employment outcomes, the study was limited to persons currently of working age (18–64 years). The RA patients in the present study were derived from 2 sources: etanercept users who had been participants in randomized clinical trials of this medication, and non–etanercept users from a separate long-term observational study. The data collection was approved by the University of California, San Francisco (UCSF) Institutional Review Board. In the etanercept users group, patients in both study arms were provided etanercept therapy after the trials, and were being followed up in an observational phase. Both etanercept users and non–etanercept users were administered identical batteries of questions about their RA, overall health status, demographic background, work history, and current employment status.
In the primary analysis, we compared patients who had been in the etanercept trials and were taking the medication versus those from the observational study who were not taking it. In a first sensitivity analysis, we conducted an intent-to-treat analysis comparing employment outcomes in the 2 groups, including in the etanercept group all those who were currently taking the medication and the 33 who were no longer doing so, and in the non-etanercept group the 23 who had begun taking the medication. In a second sensitivity analysis, we compared the employment of those who were and those who were not taking etanercept in 1999, regardless of their prior participation in the etanercept trials. In the final sensitivity analysis, we compared the employment outcomes of the 2 groups after incorporating a correction for potential selection bias in the characteristics of those who did and those who did not participate in the etanercept trials. We used logistic regression to estimate the employment rates of the 2 groups at latest interview among all who had been employed at the time of diagnosis, with and without adjustment for covariates known to affect employment status. We also used ordinary least-squares regression to estimate the hours of work per week among all who had been employed at the time of diagnosis and, among those who were still employed as of the latest interview, hours of work per week, weeks of work per year, and hours of work per year, again with and without adjustment for covariates.
The etanercept users were all adults with RA of ≥3 years' duration who had been in randomized clinical trials of this medication. Briefly, physicians at 70 sites had had patients who participated in 1 or more of the randomized trials of etanercept. In 1998 and 1999, physicians in 59 of the 70 sites (84%) agreed to have their patients participate in the observational phase of the study (Rheumatoid Arthritis Pharmacoeconomic and Outcomes Longitudinal Observational Study [RAPOLO]). There were 404 patients at the 59 sites who had been in 1 or more of the randomized trials and who had RA for ≥3 years; we successfully enrolled and conducted a structured telephone interview with 378 of the 404 (94%).
The non–etanercept users were participants in the UCSF RA Panel Study (RAPANEL), also with disease duration of ≥3 years. Briefly, RAPANEL participants were originally derived from random samples from northern California rheumatologists who maintained prospective lists of all adults with RA presenting to their offices during a 1-month period. Enrollment occurred in 1982–1983, 1989, and 1995. During these times, the physicians had listed 1,228 patients with RA on their enrollment logs, of whom we enrolled and conducted a structured telephone interview with 1,156 (94%). In each year of the RAPANEL, we have reinterviewed an average of 93% of the prior year's respondents, with approximately equal numbers lost to followup, declining further interview, or having died. In 1999, 540 were interviewed.
Patients were included in the study only if they were between the ages of 18 and 64 years in 1999 and had been working for pay or profit at the time of diagnosis of RA. Of the 378 patients in the RAPOLO, 238 were between the ages of 18 and 64 years in 1999. Of the 540 in the RAPANEL, 259 were within this age range in 1999. Of the 238 RAPOLO participants and the 259 RAPANEL who were between the ages of 18 and 64 years, 53 (22%) and 65 (25%), respectively, had not been working at the time of diagnosis.
In the primary analysis, we excluded the 33 patients in the RAPOLO group who did not report current use of etanercept and the 23 patients in the RAPANEL group who reported that they did currently take etanercept. (In sensitivity analyses, described below, this exclusion criterion was relaxed.)
All study measures analyzed for this investigation were identical in the RAPOLO and RAPANEL interviews. The employment measures were derived from the Current Population Survey (CPS) (34) and its March supplement. The CPS is the source of the monthly unemployment statistics for the nation; the March supplement to the CPS is the source of data on employment over the course of the entire prior calendar year and includes questions about the work history of the respondents. The specific employment outcome measures included whether the individual was working at the time of diagnosis and whether the individual was working at the time of the latest interview (“currently”). For both the time of diagnosis and the current year, we also asked about hours of work per week and weeks of work per year. A third measure, hours of work per year, was the product of the latter 2 measures. The CPS was also the source of items about the occupation and industry of the respondent.
Measures of RA status included separate counts of major painful and swollen anatomic sites (35), global RA status, duration of disease since diagnosis, extent of morning stiffness in minutes and/or hours, extent of fatigue, presence or absence of involvement of the eyes or lungs, presence or absence of rheumatoid nodules, and Health Assessment Questionnaire (HAQ) score (36). In addition, in both studies there were standardized batteries of questions about medications ever and currently taken and about specific health care services utilized in the year prior to survey. These questions were derived from the National Health Interview Survey (NHIS) (37), but were modified for administration to patients with RA.
Measures of overall health status included a standard item indicating whether overall health status is excellent, very good, good, fair, or poor (38), and the number of comorbid conditions reported on a checklist derived from the NHIS, but modified to incorporate conditions frequently associated with RA.
Standard demographic items were derived from the CPS and NHIS, and included age, sex, race/ethnicity, marital status, size and composition of household, extent of formal education, and household income, by category.
We began by comparing the study populations from the RAPOLO and the RAPANEL in terms of demographic characteristics, overall health status, RA status, and extent of employment at diagnosis. For characteristics that were dichotomous (for example, proportion of each group in fair or poor health), the chi-square statistic was used to compare the 2 groups; for continuous variables, the t-test was used.
We then compared the overall employment status in the 2 groups across the 2 time frames, i.e., employed versus not employed at year of diagnosis and currently. This yielded 4 categories of outcome: persons continuously employed, those exiting employment, those entering employment, and those continuously not employed. Again, the chi-square statistic was used to compare the RAPOLO and RAPANEL groups.
In the primary analysis for the study, we used logistic regression to estimate the association between RAPOLO and RAPANEL membership and current employment status among persons employed at the time of diagnosis. Seven models were tested. In the first, only the treatment group was included (RAPOLO versus RAPANEL). In the second, the following demographic characteristics were also included in the model: sex, white versus nonwhite race, age, marital status (indicator variables for widowed, divorced or separated, and never married, with currently married as the reference), and formal level of education (indicator variables for less than high school, high school graduation, and some college, with college graduates or those with graduate training as the reference). In the third model, the 2 variables for overall health status were included: the indicator variable for fair or poor versus excellent, very good, or good health, and the number of comorbid conditions. In the fourth model, the duration of RA was added. The latter model provides a minimal control for RA status without including variables such as joint counts that could be related to etanercept treatment. In the fifth model, indicator variables for occupations and industries in the job held in the diagnosis year were added to the variables included in the fourth model. In the sixth model, the following RA status measures were added to the variables included in the fourth model: number of painful and number of swollen joints, global measure of RA status compared with 1 year earlier, extent of fatigue, extent of morning stiffness, and HAQ score. The final model included all the variables in the sixth model and the indicator variables for occupations and industries.
We report below the probability of employment for the RAPOLO and RAPANEL groups, the percentage difference between the two, and the 95% confidence intervals (95% CIs) for the differences. These were calculated using the method developed by Pasta et al (39). The variance associated with the adjusted differences presented was calculated using a Taylor series approximation (40).
As a sensitivity analysis for the latter estimations, we completed an intent-to-treat analysis in which the RAPOLO group included the 33 patients who were no longer taking the medication and in which the RAPANEL group included the 23 patients who had begun taking it. As a second sensitivity analysis, we compared the employment of those who were and those who were not taking etanercept in 1999, regardless of whether they were in the RAPOLO or RAPANEL groups. As a final sensitivity analysis, we estimated a corrective factor (lambda) for selection bias using the method developed by Heckman (41). The value of lambda then was added to the model with the study group (RAPOLO versus RAPANEL). The purpose was to assess whether differences between the RAPOLO and RAPANEL groups affected the association between RAPOLO versus RAPANEL group status and employment outcome.
In the final set of analyses, we estimated the quantity of current employment in terms of usual hours of work per week among members of the RAPOLO and RAPANEL groups who had been employed at the time of diagnosis. In both groups, those who were currently employed also reported their weeks of work in the 12 months prior to interview, permitting the calculation of yet a third measure: hours of work per year. The calculation of hours of work per week among all who were employed at the time of diagnosis incorporates the total cessation of employment as well as any change in the hours of work among those who were employed. The calculation of hours of work per week, weeks of work per year, and hours of work per year among the employed shows the association of membership in the RAPOLO and RAPANEL groups and the quantity of work among those who were able to maintain employment. In all of these estimations, we used ordinary least-squares regression to estimate the association of RAPOLO versus RAPANEL membership on quantity of work, with and without adjustment for the demographic characteristics, duration of RA, RA status, overall health status, and occupation and industry covariates listed above.
Table 1 compares the RAPOLO group (etanercept users) and the RAPANEL group (non–etanercept users) in terms of demographic characteristics, overall health and RA status, and employment status in the diagnosis year. With the exception of the percent nonwhite (8.4% of the RAPOLO group versus 18.2% of the RAPANEL group), the 2 groups did not differ in demographic characteristics including age, proportion female, marital status, or various levels of educational attainment. A significantly smaller proportion of the RAPOLO members reported being in fair or poor health (17.2% versus 62.2%), but the RAPOLO members also reported a significantly greater mean number of comorbid conditions (0.76 versus 0.41). The RAPOLO members reported slightly, albeit significantly, shorter disease duration (14.0 years versus 17.2 years) and fewer major anatomic sites with painful joints (2.7 versus 3.2). They were also more likely to report improvement in their RA compared with the prior year (58.8% versus 20.1%). The RAPANEL members were less likely to report having an hour or more of daily morning stiffness (21.6% versus 59.2%) and had a better mean self-reported fatigue score (3.4 versus 2.1). The 2 groups did not differ significantly in the number of major anatomic sites with swollen joints or in HAQ score. Finally, among those employed at the time of diagnosis, the 2 groups did not differ significantly in the number of work hours per week, weeks of work per year, or hours of work per year.
|Characteristic||RAPANEL (non–etanercept users) (n = 259)||RAPOLO (etanercept users) (n = 238)||Total (n = 497)|
|Age, years†||51.0 ± 9.4||50.7 ± 7.8||50.8 ± 9.0|
|% less than high school||7.0||3.4‡||5.2|
|% high school||30.5||29.4‡||30.0|
|% some college||26.6||29.0‡||27.8|
|% graduate school or more||35.9||38.2‡||37.0|
|Overal health status|
|% in fair or poor health||62.2||17.2‡||40.6|
|No. of comorbid conditions||0.41 ± 0.7||0.76 ± 0.9‡||0.58 ± 0.8|
|Duration, years||17.2 ± 8.8||14.0 ± 8.3‡||15.7 ± 8.7|
|No. of painful joints||3.2 ± 3.3||2.7 ± 2.1‡||2.9 ± 2.8|
|No. of swollen joints||1.7 ± 2.1||1.8 ± 1.5||1.8 ± 1.8|
|% reporting ≥1 hour of morning stiffness||21.6||59.2‡||39.6|
|Fatigue score (1 = severe; 6 = none)||3.4 ± 1.2||2.1 ± 1.1†||2.8 ± 1.3|
|HAQ score||0.9 ± 0.7||1.0 ± 0.7||0.9 ± 0.7|
|% reporting RA improved in past year||20.1||58.8‡||38.6|
|Employed at diagnosis|
|Hours per week||40.3 ± 13.5||40.5 ± 16.5||40.4 ± 15.0|
|Weeks per year||48.1 ± 8.8||46.9 ± 9.4||47.5 ± 9.1|
|Hours per year||1,967 ± 770||1,958 ± 751||1,962 ± 760|
The RAPOLO and RAPANEL groups differed in the distribution of employment statuses across the diagnosis and current periods (Table 2). A larger proportion of RAPOLO members reported being employed at both time periods (55% versus 41%), and a smaller proportion of RAPOLO members who had been employed at diagnosis reported not being employed as of the latest interview (22% versus 34%). When the former 2 categories were combined, the 2 groups had almost exactly the same rate of employment at the time of diagnosis (77% among RAPOLO members, 75% among RAPANEL members). RAPOLO members who had not been employed at the time of diagnosis were less likely to report being employed currently (8% versus 15%), and were more likely to report not being employed at the time of diagnosis and currently (14% versus 10%).
|Employment status||RAPANEL (n = 259)||RAPOLO (n = 238)|
|Employed at diagnosis and latest interview||41||55|
|Employed at diagnosis, not employed at latest interview||34||22|
|Not employed at diagnosis, employed at latest interview||15||8|
|Not employed at diagnosis and latest interview||10||14|
Table 3, limited to those who were employed as of the time of diagnosis, presents the principal study results concerning the probability of current employment. On an unadjusted basis, RAPOLO members were significantly more likely to report being employed (71% versus 55% [difference 16 percentage points; 95% CI 7, 26]). Adjustment for various sets of covariates did not have much effect on the difference in employment rates among members of the 2 groups. Thus, when just demographic characteristics were added to the model, the difference between the RAPOLO and RAPANEL groups in employment rates was 17 percentage points (95% CI 8, 26); it was 20 percentage points (95% CI 9, 32) in models incorporating disease duration, RA and overall health status, and occupation and industry. Patients who were older and those with less than a high school education, longer durations of disease, and higher (poorer) HAQ scores had lower employment rates, while (counterintuitively) those with a larger number of painful joints actually had higher employment rates (data on specific variables affecting employment status not shown in table).
|Independent variables||RAPANEL (n = 194)||RAPOLO (n = 185)||Difference (95% CI)|
|Study group||0.55||0.71||−0.16 (−0.26, −0.07)†|
|Study group and demographics||0.55||0.72||−0.17 (−0.26, −0.08)†|
|Study group, demographics, overall health status||0.52||0.74||−0.22 (−0.32, −0.12)†|
|Study group, demographics, overall health status, duration of RA||0.54||0.72||−0.18 (−0.29, −0.09)†|
|Study group, demographics, overall health status, duration of RA, occupation and industry||0.53||0.72||−0.19 (−0.29, −0.09)†|
|Study group, demographics, overall health status, duration of RA, RA status||0.53||0.73||−0.20 (−0.31, −0.08)†|
|Study group, demographics, overall health status, duration of RA, RA status, occupation and industry||0.53||0.73||−0.20 (−0.32, −0.09)†|
In all of the sensitivity analyses (Table 4), the point estimates were consistent with the principal findings that RAPOLO members (etanercept users) were more likely than RAPANEL members (non–etanercept users) to report being employed currently, although after adjustment in the etanercept users versus non–etanercept users comparison, the difference did not meet the traditional criterion for statistical significance. Accordingly, when RAPOLO members not taking etanercept were included in the RAPOLO group and RAPANEL members taking the medication were included in the RAPANEL group (the intent-to-treat analysis), on an unadjusted basis, the employment rate of the RAPOLO members exceeded that of the RAPANEL members by 18 percentage points (95% CI 9, 27); after adjustment for demographic characteristics, overall health status, duration of RA, RA status, and occupation and industry, the employment rate among the RAPOLO group still exceeded that of the RAPANEL group by 10 percentage points (95% CI 1, 19). Similarly, when we compared the employment of patients who were and those who were not taking etanercept regardless of whether they had been in the RAPOLO or RAPANEL groups, on an unadjusted basis the employment rate of the former group exceeded that of the latter by 10 percentage points (95% CI 1, 19). After adjustment, the difference was 9 percentage points (95% CI −1, 18). In the final sensitivity analysis, in which a correction for selectivity bias as well as the RAPOLO versus RAPANEL group membership was included in the model, on an unadjusted basis the difference was 15 percentage points (95% CI 5, 24). After adjustment, the difference was 19 percentage points (95% CI 8, 29).
|Sensitivity analysis†||Unadjusted difference in employment rates (95% CI)||Adjusted difference in employment rates (95% CI)|
|Intent-to-treat analysis†||−0.18 (−0.27, −0.09)‡||−0.10 (−0.19, −0.01)‡|
|Etanercept vs. no etanercept||−0.10 (−0.19, −0.01)‡||−0.09 (−0.18, 0.01)|
|RAPANEL vs. RAPOLO with correction for selection bias||−0.15 (−0.24, −0.05)‡||−0.19 (−0.29, −0.08)‡|
When incorporating those who had stopped working and those who reported being employed as of the latest interview, on an unadjusted basis the RAPOLO members averaged 5.4 more hours of work per week than the RAPANEL members (95% CI 1.1, 9.7) (Table 5). On an adjusted basis, the RAPOLO members averaged 7.4 more hours per week (95% CI 2.6, 12.3), or approximately one-third more work time than the RAPANEL members.
|Current employment status, quantity of employment†||RAPANEL||RAPOLO||Difference (95% CI)|
|Patients currently employed and patients not currently employed|
|Mean hours per week|
|Unadjusted||21.4||26.8||−5.4 (−9.7, −1.1)‡|
|Adjusted||20.4||27.8||−7.4 (−12.3, −2.6)‡|
|Mean hours per week|
|Unadjusted||38.8||37.8||1.0 (−2.7, 4.6)|
|Adjusted||38.0||38.5||−0.5 (−4.0, 5.0)|
|Mean weeks per year|
|Unadjusted||47.8||47.8||0.0 (−2.3, 2.3)|
|Adjusted||46.8||48.6||−1.8 (−4.7, 1.2)|
|Mean hours per year|
|Unadjusted||1,916||1,876||40 (−156, 236)|
|Adjusted||1,839||1,940||−101 (−345, 142)|
Among patients who reported being employed currently, the 2 groups did not differ significantly or substantially in any of the 3 measures of the quantity of work (Table 5). On an unadjusted basis, they differed by ∼1 hour of work per week, by 0 weeks of work per year (after rounding), and by only 40 hours of work per year. After adjustment, the groups differed in the quantity of work one-half hour per week, <2 weeks per year, and only 101 hours of work per year.
Work loss is an extremely common occurrence among patients with RA, affecting a majority of those employed at the time of diagnosis, as documented in many studies (18–20). Moreover, indirect costs associated with the work loss appear to be the single largest item in cost of illness studies in RA, larger than costs of all ambulatory or hospital care, including total joint replacement (21–23). Updated to 2001 terms (42, 43), annual indirect costs in 2 studies using the standard prevalence-based human capital approach averaged $18,473 (21, 23). This is almost half of the median annual wages and salaries for US full-time workers in that year (42). Maintaining individuals' ability to remain at work after diagnosis is essential to reduce the prevalence of work disability, since the rate of return to work after cessation of employment is low (3).
Several observational studies have shown that, although medical factors are certainly correlated with employment outcomes, demographic characteristics and the nature of the job held may have a greater effect on the probability of staying at work (2–4, 7, 12, 18). The relatively small impact of medical factors on employment may explain why the addition of such factors did not narrow the difference between the RAPOLO and RAPANEL groups in the employment outcome. Of course, the relatively small impact of medical factors may change with the advent of more effective therapies for RA. One published study and 2 abstracts provide some evidence that DMARD treatment in general and a biologic agent (anakinra) in particular may have a positive effect on employment. However, we could find no evidence with respect to the impact of anti–tumor necrosis factor (anti-TNF) therapy on employment.
We report here the results of an observational study of employment outcomes among patients with RA, stratified by whether they had been participants in etanercept trials. We found that, among RA patients who had been employed at the time of diagnosis, those from the etanercept trials were more likely to be currently employed. On an unadjusted basis, the current employment rate among those from the etanercept trials was 16 percentage points higher (95% CI 7, 26); full adjustment did not narrow the difference (it remained 20 percentage points [95% CI 9, 32]). Although among those who maintained employment, the 2 groups did not differ in the quantity of their current work, when those who had left employment were included in the analysis, on an unadjusted basis those who had participated in the etanercept trials reported 5.4 more hours of work per week (95% CI 1.1, 9.7) and on an adjusted basis they reported 7.4 more hours of work per week (95% CI 2.6, 12.3).
In an observational study such as this one, it is entirely possible that differences in the characteristics of the 2 groups may account for the employment outcomes found. We used 4 separate techniques to attempt to reduce this bias. First, we used multivariate regression to adjust for differences in demographic characteristics, duration of RA, RA and overall health status, and the mix of occupation and industries in the job held at the time of diagnosis. As noted, after adjustment for these characteristics, the difference between those who had and those who had not participated in the etanercept trials remained.
Second, we performed an intent-to-treat analysis in which we included in the etanercept trials group those who had participated in the trials but who were no longer taking the medication, and in the non-trials group those who were actually taking the medication. The results of this sensitivity analysis, reported above, were consistent with the results of the principal analysis, although after adjustment the impact was smaller. Third, we compared employment according to etanercept use, regardless of the original group membership. On an unadjusted basis, the employment difference was 10 percentage points and statistically significant; after adjustment, it remained 9 percentage points, but was no longer significant. Of note, the difference was of the same magnitude (10 percentage points) as in the study published in abstract form in which a randomized trial was used to compare the impact of combinations of DMARDs versus monotherapy (33). Thus, it is possible that the magnitude of effect in the principal analysis of this study is in part the result of unmeasured differences in the characteristics of patients in the RAPOLO and RAPANEL groups, for example, if persons who choose to participate in a trial are more likely to experience better employment outcomes. However, in the final sensitivity analysis we estimated a correction for selection bias using the method of Heckman, and the point estimate of the difference remained substantial (15 percentage points without adjustment and 19 percentage points with).
It is also possible that the magnitude of the difference in the employment outcomes may be due to differences in employment that arose between disease diagnosis and the start of etanercept therapy. However, there were only slight differences between the 2 groups in their employment at the time of diagnosis, reducing the chance that changes between diagnosis and the initiation of etanercept therapy could account for the findings reported here.
Overall, the evidence from the base case and sensitivity analyses is consistent with an association between etanercept usage and better employment outcomes, albeit with the caveat that part of the effect may be due to unmeasured differences in the kinds of patients with RA in the RAPOLO and RAPANEL groups. Observational designs such as the one used in the present study cannot provide a definitive answer to the question of whether etanercept results in better employment outcomes. Given the cost of this medication in particular and anti-TNF therapy in general on the one hand, and the large magnitude of indirect costs associated with RA on the other, it would be prudent to implement randomized clinical trials of sufficient duration to test the employment effects of such agents in a more definitive manner. The results of the present study can be used to justify such trials.
The authors gratefully acknowledge the counsel of Mary Buatti, formerly of Wyeth.