We use the UCSF Lupus Outcomes Study (LOS), a longitudinal cohort of persons with this condition, to estimate the time from incident specific organ manifestations, including thrombotic, neuropsychiatric, and musculoskeletal, and increases in disease activity until work loss among those employed at the outset of the LOS We also describe the sociodemographic, SLE, and employment characteristics associated with an elevated hazard of such work loss.
The LOS has included 1204 persons with SLE sampled from clinical environments (n=472, 39%), and such community-based resources as support groups, websites, and newsletters (n=732, 61%). The enrollment occurred between 2002 and 2009.
To be enrolled in the LOS, participants had to have a diagnosis of SLE confirmed by medical record review by a rheumatologist or nurse working under a rheumatologist’s supervision. The principal data collection for the LOS is an annual structured telephone interview that collects information on demographic and socioeconomic characteristics; status of SLE, including a validated measure of activity, the Systemic Lupus Activity Questionnaire (SLAQ) (9
); specific manifestations; general health and functional status; medications; a complete inventory of health care encounters in the twelve months prior to interview; kind and extent of health insurance; and employment status. Prior publications have listed the specific measures collected within each of these major categories of variables (3
For the present analysis, we omitted those among the 1204 LOS participants not working at their baseline LOS interview (n=658, 54.7%), those aged 65 or older (n=16, 1.3%), those with only one LOS interview due to death, decline of further participation, or loss-to-follow-up (n=43, 3.6%), and those with missing data on any variable used in the analysis (n=4, 0.3%), yielding a total of 483 participants.
Thrombotic events included deep vein thrombosis; pulmonary embolism; stroke; myocardial infarction; retinal vein thrombosis; other blood clots; or second or third trimester miscarriage, according to the criteria defined in previous LOS analyses(12
). LOS participants were classified as having an incident thrombotic manifestation if they had reported any of the above events since their prior interview. Thus, an incident event here was not necessarily the first thrombosis experienced by the participant. We did not have access to complete medical records to validate all self-reported thrombotic events. However, we were able to validate thrombotic events that occurred prior to the baseline interview using medical records originally obtained to determine study eligibility. Using these records to document self-reported events on the baseline interview, self-report of thrombosis had 97% specificity and 68% sensitivity with regard to the record data. The low sensitivity is likely due to the long recall period for these events (often more than 10 years prior to the baseline interview). In order to address whether a shorter recall period increased the sensitivity of self-reported events, we compared the interview data to the medical records for the 55 LOS participants for whom we had a record of a thrombotic event in the medical charts that was contemporaneous with the interviews. Among these participants, all but six accurately reported the thrombotic event found in the charts, yielding an improved sensitivity of 89%.
The definition of neuropsychiatric manifestations is based on new onset of clinically significant symptoms of depression, cognitive dysfunction, or seizures. We used a cut-point of 24 on the Center for Epidemiological Studies Depression Scale(CESD)(13
), which has been established as an appropriate indicator for this population(14
). Cognitive dysfunction was defined by a z-score of less than or equal to −1.5 compared to population normative data on one or more of three measures of cognitive functioning, including the Hopkins Verbal Learning Test-Learning Index, Hopkins Verbal Learning Test –Delayed Recall Index, or Controlled Oral Word Association (15
). Participants were classified as having an incident neuropsychiatric manifestation in an interview wave if they had one or more of the following: worsening depression defined as a CESD score of 24 or greater after at least one lower score; no cognitive dysfunction in previous interviews and presence of cognitive dysfunction in the current interview; or, report of a seizure in the year prior to the current interview.
Musculoskeletal manifestations were defined as a report of severe muscle pain, muscle weakness, pain or stiffness in joints, or swelling in joints in the three months prior to interview; these items derive from the SLAQ (9
). Participants were classified as having an incident musculoskeletal manifestation in an interview wave if they reported “no” or “moderate” symptoms from the 4 musculoskeletal conditions in the previous interview and reported “severe” symptoms in at least one of the conditions in the current
Renal manifestations were defined by the report of having a kidney biopsy or starting dialysis since the prior interview. Among the LOS participants in the study sample, i.e., those who were employed and under age 65 at baseline, 106 (22%) met this criterion prior to the start of the study. However, only 30 participants had an incident renal manifestation as defined here since the baseline interview year. As such, renal manifestations were not included in the present analysis because we lacked statistical power to estimate the relationship of these manifestations to employment.
Increases in Disease Activity
In the LOS, disease activity, as indicated above, was measured by the SLAQ self-report measure. An increase in disease activity was defined as an increase of 0.5 standard deviations of the mean SLAQ score, consistent with the definition proposed by Norman and colleagues for a clinically meaningful change (10
The annual telephone surveys include employment measures from the Current Population Survey (CPS), the source of the monthly employment statistics in the U.S. (18
). In the CPS, persons are defined as employed only if they report being employed, on temporary leave but with a job, or working for pay or profit in the week prior to interview. For those not working according to this definition, information is collected on the month in which the individual stopped working. For respondents for whom the precise date of work loss was not available, we assumed that work cessation occurred at the midpoint between consecutive interviews, a common approach in life table analyses.
Additional employment measures included job tenure (number of years at current job) and the type of industry in which the participant worked, categorized as government, including education; goods producing industries; professional, technical, scientific, and media services; retail, wholesale, and finance; and other service industry jobs.
Measures of Health Status and Sociodemographics
In addition to the specific SLE manifestations described above, health status measures included disease duration (years since diagnosis), a measure of general health (ranging from excellent to poor), and a global assessment of disease activity (on a 10 point scale). Sociodemographic variables included age, categorized as 18–34, 35–54, and 55 – 64; gender; race/ethnicity, dichotomized to white non-Hispanic vs. all others; and education, categorized as high school graduate or less, some college, and baccalaureate degree or greater.
Estimating the Onset of Incident Manifestations or Increases in Disease Activity
As mentioned, we hypothesize that among employed subjects the hazards for job loss would be higher among subjects who have an incident SLE-related manifestation or an increase in disease activity. Specifically, we examined whether having such events during the interview period immediately preceding job loss would increase the hazards of job loss. Since our data is collected longitudinally, we were able to assess at each interview if a patient reported having an incident SLE manifestation or increase in disease activity since their prior interview (on average 12 months). Because the reports of incident manifestations correspond to the prior 12 month interval and the report of job loss is an exact date, we assigned the manifestations variable from the interview immediately preceding the job loss date in the following way. First we calculated the midpoint between the dates of interview with first reported of job loss or censorship and and the prior interview. When the job loss date was before the midpoint of interview dates, the manifestation value from the prior interview was assigned; if the job loss date was at or after the midpoint interview dates, the value from the latter interview date was assigned. These assignments may have resulted in some cases being misclassified. For example, if an individual reported stopping work in the tenth month of the year prior to the interview that individual would be assigned the manifestation value from that same interview. It may be possible however, that the manifestation occurred during the eleventh or twelfth month of that interview period, after the time of job loss. However, results of a sensitivity analysis in which we assigned the manifestation data from the prior interview whenever job loss occurred at after or at the midpoint of the year did not differ appreciably from the results presented below. Because the neuropsychiatric manifestations and increase in disease activity measures were based on a change in scores from one year to the next rather than discrete events, we assigned the end of the year in which the change occurred as the point of incidence for these manifestations.
We calculated the time until first reported work loss among persons with SLE who met the criterion for employment in the baseline interview and subsequently developed incident manifestations or an incident increase in disease activity. Thus, we focused on the impact of specific incident manifestations or the number of such manifestations or increases in activity on work loss among those employed prior to the incidence of each of three manifestations or increased activity. In all analyses, we excluded those who were 65 years of age or older at baseline.
We first used the Kaplan-Meier method to compare the employment trajectories in LOS participants who experienced the SLE manifestations under study. We estimated the proportion working in each month following the incident manifestations or increase in disease activity. In these analyses, individuals who continued to be employed were censored observations, as were individuals who reached age 65 without work loss. For all observations, the length of time until work loss is measured not since study entry, but from the point of the first incident manifestation in each category. For this reason, not all observations are carried through for the entire time period, as reflected in , below.
Time Until Work Loss Among Persons with Incident Thrombosis, Musculoskeletal or Neuropsychiatric Manifestations or Increase in Disease Activity
We next estimated bivariable Cox proportional hazard regressions to ascertain the relative impact of the specific incident manifestations, number of manifestations, or increases in disease activity and other variables on the risk of work loss. Because we were interested in the immediate effect of an incident manifestation on employment status, we used a time-dependent approach (19
) in which the risk of job loss is evaluated based on the incident manifestation value from the interview immediately preceding the job loss or censorship. In multivariable Cox regressions, we also included sociodemographic measures (age, gender, race/ethnicity, and education), job characteristics (job tenure and industry), and disease duration. We did not include the general health status or global disease activity measures in these models, as we viewed them as potentially part of the pathway from manifestation to job loss and thus would be inappropriate to include in a multivariable model. To control for changing economic circumstances, we included the the year of baseline employment as a continuous variable
In the analysis of the impact of manifestations, the multivariable Cox regression included the specific time-dependent manifestations, i.e. thrombotic, musculoskeletal, and neuropsychiatric. In a separate Cox regression analysis, we substituted the number of the foregoing manifestations (with zero manifestations as the reference); this regression included the same variables listed above. The analysis of the impact of an increase in disease activity was done separately from the analysis of specific manifestations since there was considerable overlap between the SLAQ and the musculoskeletal manifestation measure. This analysis included only 342 participants, rather than 483, because the SLAQ score was only begun in the second wave of LOS data collection after its validity had been established.