|Home | About | Journals | Submit | Contact Us | Français|
To track work loss among those employed and work entry among those not employed prospectively in a cohort of persons with SLE, assess risk factors for these outcomes, and compare rates of the outcomes to a matched national sample.
The present study analyzes four years of the Lupus Outcomes Study data, augmented by information on the local labor market from the Census Bureau and the Bureau of Labor Statistics. We use the Kaplan-Meier method to assess time from study initiation until work loss or work entry and Proportional Hazards regression to estimate factors affecting these outcomes. Finally, we compare rates of work loss and entry in the LOS to rates in the Survey of Income and Program Participation (SIPP).
At study initiation, 394 (51%) LOS participants were employed, of whom 92 (23.4%) experienced work loss. In multivariate analysis, advanced age, lower cognitive and physical functioning and higher reports of depressive symptomotology predicted work loss. In comparison to the SIPP sample, rates of work loss did not differ. Of the 376 LOS participants not employed, 76 (20.2%) experienced work entry. In multivariate analysis, less disease activity, fewer lung manifestations, better physical functioning, and shorter time since last employment predicted work entry. In comparison to SIPP, rates of work entry were only lower between ages 35–55.
Until age 55, low rates of employment among persons with SLE may be due to lower rates of work entry rather than higher rates of work loss. Beyond age 55, both high rates of work loss and low rates of work entry contribute to low rates of employment.
A series of studies using retrospective recall have established that persons with SLE have low employment rates (1–3, 4). For example, Partridge noted that almost half of persons with SLE stopped working within the first few years of disease (1). Mau (2) reported that, compared to an age- and gender-matched sample, persons with SLE with disease duration of a decade or longer were about a third less likely to remain employed. Using the same data source as the current paper, we noted that overall productivity among persons with SLE with an employment history declined by a third between the year of diagnosis and the most recent year, an average of about 13 years later, with most of the loss occurring among those who stopped working completely rather than among those who reduced their hours of work (3). Panopalis also used the same data source to describe the impact that memory impairment has on employment (4).
The foregoing studies established the basic parameters of the impact of SLE on employment. However, due to their retrospective nature, they could not fully explore the role of SLE-related factors on employment dynamics. The present study advances the study of employment dynamics among persons with SLE because of the prospective nature of the data.
This analysis uses data from the UCSF Lupus Outcomes Study (LOS), augmented with contextual data from the U.S. Bureau of the Census and the U.S. Bureau of Labor Statistics, to assess the impact of sociodemographics, health status, and work characteristics on work loss and work entry among persons with SLE. We also compare their rates of work loss and work entry to those of a matched national sample from the Survey of Income Program Participation (SIPP).
The LOS is an ongoing, longitudinal study of 957 individuals with a confirmed SLE diagnosis based on American College of Rheumatology criteria (5, 6). Details about enrollment and data collection have been reported previously (3) and are briefly summarized here. Participants were recruited through health care settings (34%) and non-clinical sources, including patient support groups and conferences, newsletters and websites; 73% are from California and the remainder from 38 other states. LOS interviews have been conducted annually by telephone since September, 2002. They include validated batteries capturing demographics and socioeconomic status, status of SLE, general health and functional status, mental and cognitive status, health care utilization, and employment. An average of 94% of eligible participants from each wave has been re-interviewed in the subsequent wave. Fourteen participants have withdrawn for health reasons, 38 have died, 75 have declined further participation, and 34 were lost to follow-up.
Employment rates were obtained from the 2000 U.S. Census at the block group level (areas covering 600–3,000 persons) and matched to the LOS participants’ home address through geocoding; details on this process have been published previously (7). County unemployment rates from 2003 were obtained from the Bureau of Labor Statistics, Local Area Unemployment Statistics file and matched to LOS participants’ county of residence. National longitudinal data on employment status were accessed from the U.S Census Bureau Survey of Income and Program Participation (SIPP) 2001 panel, the most recent available (8). The SIPP panel includes 36,700 households interviewed every four months from February 2001through January 2004. Employment status is available for every week during that period.
The present analysis incorporates the first four interviews of LOS, collected between September 2002 and February 2007. Of the 957 participants, 770 (80%) were included in the analyses after excluding 70 (7%) participants without follow-up interviews, 77 (8%) who were 65 or over at baseline, one participant living outside the US, and 39 (4%) observations with missing data. The final sample was subset into those employed (n=394) and not employed (n=376) at baseline. All analyses were performed separately by employment status group.
We calculated the duration of time until first reported work loss among those employed at baseline. Employment was defined as either working, with a job but not working, or doing any work for pay or profit in the last week. Anyone who neither had a job in the last week nor did any work for pay or profit was defined as not employed. Only the first incident of work loss was considered in this analysis, whether it was temporary or permanent. We estimated the month and year of work loss, using the exact date when available (in approximately half the observations). For the remaining observations, which provided an interval during which participants last worked rather than an exact date, we estimated time until work loss using the method outlined by Lindsey and Ryan for interval censored data (9). We calculated the range of dates during which the work stoppage would have occurred, created variables to capture the beginning, midpoint, and endpoint of that interval, and ran all statistical analyses on each of the three variables. As the results did not differ appreciably, we report the results from the analysis using the midpoint of the interval.
Among those not working at baseline, we calculated duration of time until first work entry (i.e., first report of employment after baseline interview). We were able to calculate the exact time until work entry for approximately three-quarters of the participants. For the remainder, we estimated the interval until work entry using the method described above.
The framework for our analysis is based on a model of disability developed by Nagi (10) and extended by Verbrugge (11), which proposes a progression in the disablement process from underlying pathology (disease severity), to impairment (functional status) and, finally, to disability in major life activities (work loss). Yelin, et al (12) further extended the model of work disability to include societal and work-related factors. Following these models, potential predictors of work loss/entry were organized into five major sets: sociodemographic characteristics, neighborhood employment rate, disease severity, functional status, and characteristics of the job held at baseline (among those employed) or years since last employed (among those not employed). The specific variables within each set are listed in Tables 1 and and2.2. The models include numerous validated self-report health status measures: the MOS Cognitive Functioning Scale (13), SF-36 Physical Function Scale (14), and the Center for Epidemiologic Studies Depression Scale (15). Self-report of specific disease manifestations in LOS were validated using medical chart data, as outlined in Hersh (16). The measures of physical and cognitive demands at work derive from the Health and Retirement Survey (17). The measure of high demands and low control is from the Job Content Questionnaire (18). Traditional employment refers to full-time regular employment on the day-shift for a single employer; it has been used by these authors in previous analyses (19).
The enrollment and data collection protocol was approved by the UCSF Committee on Human Research.
We used the Kaplan-Meier method (20) to estimate the duration of time until work loss/entry among the employed and not employed samples, respectively. We then compared the employment patterns in the LOS sample to a general population sample, using SIPP data. To match the SIPP and LOS samples, we dropped men from both samples since the LOS sample is primarily female (90%), and charted the first three years of follow-up for the LOS since the SIPP is a 36-month cohort. It was not possible to precisely match the timeframes for the two samples, as the most recent SIPP cohort ended in 2004. However, the average unemployment rates were very similar for the years covered by the LOS and SIPP, 5.4% for the former and 5.5% for the latter.
The analyses were stratified by age group. We used the Kaplan-Meier procedure in SUDAAN (Research Triangle Park, NC) in order to account for the SIPP sampling design.
We used Cox proportional hazards regression (21) to estimate risk factors for work loss/entry among the LOS employed and not employed samples, respectively. We estimated regressions on each of the five sets of covariates: demographic characteristics, neighborhood employment rate, disease severity, functional status, and, for the work loss model, job characteristics. All variables in each set were included, unless there problems with collinearity, in which case we selected the variable with the most explanatory power. We then combined sets of covariates in two full models, one containing all covariates except functional status and one containing all covariates except disease severity. Collinearity between the measures of disease severity and function precluded putting them in the same model. In the foregoing analysis, we found no main effect of work characteristics on the risk of work loss. Accordingly, we re-ran the full models excluding the work characteristics; as there was no change in the overall fit of the model, the latter model is not reported here. We also tried alternative measures of certain key risk factors, for example, substituting the SF-12 Mental Component Score(22) for the CES-D. In no instance was there a compelling reason to report other than the principal analysis.
We used SAS 9.1 (SAS Institute, Cary, NC) Proc PHReg for the Cox proportional hazards regressions.
Tables 1 and and22 show the characteristics of those employed and not employed in the baseline year of the LOS, cross-classified by whether they subsequently left or entered work, respectively. Among the 394 persons with SLE employed at baseline, 92 (23.4%) left work over the ensuing three waves of data collection (Table 1). Those who left work did not differ significantly from those who did not in gender, ethnicity, marital status, or extent of education; they were, however, more likely to be 55 to 64 years of age. Persons with SLE who left work reported higher levels of SLE activity and CES-D scores (indicative of higher levels of depressive symptoms) and were more likely to report having one or more comorbid conditions. They also reported slightly, albeit significantly poorer cognitive status (by MOS scale), and poorer physical functioning (by SF-36 scale).
With the exception that persons with SLE who ultimately left work reported lower annual hours of employment in the baseline year (1457 versus 1707), they did not differ from those remained at work on any other work characteristic. Of note, persons with SLE who left work did not live in counties with higher unemployment rates or in local neighborhoods (as measured by the Census block group) with lower employment rates.
Among the 376 persons with SLE who were not working at the time of the baseline interview, 76 (20.2%) entered work over the ensuing three waves of data collection (Table 2). Those who entered work were younger and less likely to be married but did not differ in gender, ethnicity, or education. In terms of health characteristics, those entering work had shorter durations of disease, lower disease activity, fewer comorbid conditions and lung manifestations, but not thrombotic events or renal manifestations. Not surprisingly, they reported substantially better physical functioning at baseline. Although a significantly shorter amount of time had passed since they last worked, they did not differ from those who did not enter employment in their county unemployment rate or in their local neighborhood employment rate.
Figure 1 shows results of the time until work loss among those employed at baseline. Among persons with SLE of all working ages employed at baseline, more than 10% stopped working by 12 months after the baseline interview and just over 20% by 36 months. The percentage of individuals experiencing work loss was remarkably similar between persons with SLE of all working age and the national sample from SIPP of persons these ages.
Time until work loss was similar among persons with SLE and the SIPP sample for persons 35 to 54 years of age, the prime working ages, with about 20% of each sample leaving work by 36 months. Among workers between 18 and 34 years of age, the LOS and SIPP samples had almost exactly the same probability of work loss until 12 months after interview; after that point, however, persons with SLE did not sustain additional work loss in contrast to the SIPP sample
Rates of work loss among persons with SLE who were between 55 and 64 were similar to the SIPP national sample. At 36 months, a third of both groups had stopped working.
Focusing just on those with SLE, the probability of work loss increased sharply at age 55 or greater. Just under 20% of those between 18 and 54 experienced work loss by 36 months, but more than 30% of those between 55 and 64 did.
Figure 2 shows the results of the analysis of time until work entry among the two samples. Among persons of all working ages in LOS, just under 10% had entered work by 12 months and just over 20% by 36 months. In the full SIPP sample, about a quarter had entered employment by 12 months and nearly 40% had done so by 36 months. At that point, those with SLE were just under half as likely as those from the SIPP sample to have entered employment.
The difference in rates of work entry between persons with SLE and the national sample was most pronounced among persons 35 to 54 years of age. In that age range, only about 15% of persons with SLE and about 20% from the SIPP had experienced work entry by 12 months after initial interview. Among those between 18 and 34, job entry patterns were similar, and at 36 months, the rates were close to 45 and 55% for the LOS and SIPP samples, respectively. Very few of either group between the ages of 55 and 64 experienced work entry; even at 36 months after initial interview, only about 10% of either group had experienced work entry.
Focusing just on those with SLE, the likelihood of work entry declined markedly after age 35. About 45% of those between ages 18 and 34 had entered work by 36 months, but only about 15% of those 35 to 54 and only about 10% of those 55 to 64 did so.
Table 3 shows the impact of various sets of risk factors for work loss individually and in combination among persons with SLE employed as of the baseline interview. With the exception of neighborhood employment rate and work characteristics, each set of risk factors was associated with the risk of work loss when analyzed alone. With respect to individual variables within sets, lower age, higher levels of cognitive and physical functioning, and higher levels of cognitive job demands were associated with a lower risk of work loss (the hazard ratio for cognitive and physical function is estimated per point on a 0 to 100 scale), while lower education and higher levels of CES-D score (estimated per point on a 0 to 60 scale) was associated with higher risk. When combining the sets in the full model with severity measures, only older age and higher CES-D remained predictive of work loss. In the full model with functional measures, older age, never having been married, and lower cognitive and physical functioning were associated with work loss. None of the individual work characteristics was associated with work loss in the full models.
Table 4 summarizes the results of the analysis of the various sets of factors affecting work entry individually and in combination. When analyzing the sets of factors separately, each set was significantly associated with the risk of work entry, with the exception of the neighborhood employment rate. With respect to individual variables, being in the youngest age group, never having been married, having shorter duration of SLE, lower levels of disease activity, higher levels of physical functioning, and a shorter period of time since last regular employment were associated with an increased rate of work entry. In the multivariate analyses, age and marital status were not associated with the risk of work entry in either the “Severity” or “Function” models, while years since last regular work were associated with work entry in both models. Lower disease severity and fewer lung manifestations were associated with reduced risk of work entry in the “Severity” model while higher physical function was predictive of work entry in the “Function” model.
It has been established that persons with SLE have substantially lower employment rates than individuals of similar age. Using the same data source as the present study, for example, we previously observed that by little more than a decade after onset, only about half of working age adults with SLE were employed (3). The productivity of such adults had declined by about a third, mostly as a result of complete cessation of work rather than reduced hours of those who remained on the job.
The estimates in the latter study as well as others, however, should be considered as conservative estimates of the impact of SLE on employment. All of the studies of employment among persons with SLE use retrospective data collection among prevalence cohorts because establishing true incidence cohorts is difficult in rare conditions such as this one. It stands to reason that, even though mortality rates in SLE have declined over the years, some persons with SLE died before becoming eligible to enter prevalence cohorts. Since such persons likely had severe forms of the condition, it is likely that had they not died, they would have had a high probability of work loss.
In most severe chronic diseases, the initial goal of improved treatment is to reduce mortality, in effect to turn fatal conditions into chronic ones. Thereafter, the goal shifts to improving quality of life. In SLE, the transition from a fatal to chronic condition is well underway (23–24). The relatively low rates of employment would suggest that the transition to reducing the substantial impact of SLE on quality of life is incomplete.
The contribution of the present study is two-fold. First, because it is prospective in design, we were able to capture the impact of a full range of risk factors on employment, including those subject to substantial recall bias such as past levels of disease activity. The study also included sets of risk factors not frequently included in work disability studies such as the unemployment rate in the county of the respondents and the employment rate of their neighborhoods. Secondly, it is a given at this point that employment rates among persons with SLE are relatively low. The present study allows us to estimate the risk factors for transitions from current status that could result in improvement in the employment prospects of persons with SLE by decreasing rates of work loss and increasing rates of work entry.
As a result of transitions out of employment between the onset of SLE and the baseline year of the LOS, only about half were employed in the latter year. Over the ensuing thirty-six months, more than a fifth of all of those who were employed in the baseline year experienced work loss. Time until work loss was similar for those 18 through 34 and 35 through 54. However, time until work loss was much shorter for persons with SLE ages 55 through 64, about forty percent of whom had stopped working by 36 months after the initial LOS interview. With the caveat that persons with SLE started with lower employment rates, rates of work of loss among all working age persons were similar between those with SLE in the LOS and the national SIPP sample. However, the age patterns were different, with those ages 35 to 54 in the national sample having the lowest rates of work loss, and those in the youngest and oldest age groups having higher rates.
Rates of work entry were substantially lower among persons with SLE than in the national sample. This was true in every age group, but the differential impact was especially profound among those 35 to 54, the ages when most of us are acquiring the jobs we will hold for the longest amount of time during our working lives, often called “career jobs” (25–26). Together with the results for work loss, this suggests that persons with SLE take longer to gain the toehold in such jobs in the prime working ages, and then go on to sustain an earlier exit from employment, an effect accentuated by their relatively low employment rates to begin with.
Younger age and higher levels of cognitive functioning were associated with a lower risk of work loss among those employed, while higher levels of depressive symptoms were associated with a higher hazard ratio, suggesting that effective treatment for depression might reduce the speed with which persons with SLE exit the labor force. However, no single work characteristic predicted work loss, a finding at odds with many studies in rheumatoid arthritis and other chronic conditions (27–30). The severity of SLE may simply trump risk factors found to affect work outcomes in these other conditions. The strongest factor affecting the risk of work entry was the amount of time that had elapsed since the last regular job, indicating once again the importance of maintaining employment as the key strategy to improve employment outcomes in SLE. How to accomplish the latter goal remains elusive, however, since we could not identify kinds of jobs or working conditions that, in the work loss models, were associated with the retention of employment.
The principal limitation of the current study is that the data source is not a true incidence cohort. As a result, when the cohort was formed, many of the LOS participants had already left work and could not be expected to report on the characteristics of their disease in prior years that had led them to stop working. This also limits our ability to study how individuals and their employers accommodate the onset of the disease and subsequent periods of severe exacerbation. It may also explain why there was no association between work characteristics and work loss; persons at highest risk of work loss may have left employment prior to the formation of the cohort. Another limitation is that self-report of disease variables may have limited our ability to estimate their impact on work loss although we have validated respondent reports of disease manifestations (16).
Nevertheless, the results indicate that, at least until age 55, the employment problems of persons with SLE may be due more to their lower rates of job entry if out of the labor force than to high rates of work loss if employed. This result is broadly consistent with studies of other disease entities in showing that every effort must be made to help persons with health problems, in this case SLE, stay on the job despite their illness (18). Unfortunately, the relative dearth of strong risk factors for work loss from this study suggests that figuring out how to accomplish this goal will not be easy.
NIAMS P60-AR-053308, AHRQ/NIAMS R01-HS-013893, Arthritis Foundation, State of California Lupus Fund (Dr. Yelin), R01 AR44804 and M01-RR-00079 (UCSF General Clinical Research Center for Dr. Criswell), and Rosalind Russell Medical Research Center for Arthritis, UCSF
The authors gratefully acknowledge the contributions of Janet Stein, Stephen King, Jessica Spry, and Rosemary Prem.