Employment and health status changes among women with fibromyalgia: A five-year study

Abstract

Objective

To assess changes in health status of women with fibromyalgia (FM) over 5 years and determine whether baseline employment status influences health outcomes adjusting for other baseline factors.

Methods

Two hundred eighty-seven women with FM were recruited from a national sample of rheumatologists and interviewed by phone at baseline and annually for 4 years. Data were collected on pain, fatigue, Center for Epidemiologic Studies Depression Scale and Modified Health Assessment Questionnaire (M-HAQ) scores, demographic characteristics, and employment status. At the end of the study, 211 participants remained. Data were analyzed using multilevel modeling techniques. Bootstrap methods adjusted for the cluster sampling.

Results

The participants' mean ± SD age was 47 ± 11 years, their mean ± SD education level was 14 ± 2 years, 90% were white, 50% employed, 64% married, and their median household income was ≥$50,000. Mean ± SD scores at baseline were 57.2 ± 24 for pain, 75.4 ± 22 for fatigue, 22.9 ± 13 for depression, and 0.73 ± 0.5 for the M-HAQ. Multilevel modeling indicated that all health status measures declined significantly over time except for pain. Rates of change varied from −1.22 for fatigue to −0.03 for the M-HAQ. Except for pain, patients who were employed at baseline had better health status over time. The employment and time interaction was not significant, indicating that health status changed at the same rate regardless of employment status. Other significant factors were age and income.

Conclusion

Employed women with FM have better health status at baseline and maintain that advantage over time. Employment does not seem to provide a protective health benefit.

INTRODUCTION

Fibromyalgia (FM) is a highly prevalent rheumatic disorder affecting 5–7% (1) of the US population, mostly women (2–4). FM is defined as widespread pain accompanied by 11 or more of 18 specific tender point sites (5). Other symptoms include fatigue, sleeplessness, and stiffness. The etiology of FM has not been confirmed, but several theories suggest that FM is a pain amplification disorder (6–8) resulting from the dysregulation of central pain processing. Relatively little is known about the progression of FM over time. The few longitudinal studies of FM indicate that total remissions are rare. Findings are inconsistent about long-term prognosis; some studies demonstrate general improvement in symptoms and functional status over time, but others show no improvement, or worsening of condition (9–17).

We previously reported on the cross-sectional relationships between employment and health status among women with FM (18). Based on community studies of women and general health status, we hypothesized that employment would provide a health benefit for women with FM (19–29). Employed women with FM reported better health status than those who were not employed. Although the results were suggestive, it was unclear whether employment provided a health benefit or whether the women with FM were employed because they were healthier. The purpose of this study was to assess health status changes among women with FM over a 5-year period, and to determine whether baseline employment status influenced health outcomes after adjusting for demographic factors.

PATIENTS AND METHODS

Patients.

A 2-step methodology was used to recruit women with FM. First, a national sample of rheumatologists was randomly selected from the American College of Rheumatology (ACR) membership. Physicians who were listed as Fellows were sent letters about the study, and we followed up with telephone calls to arrange participation. Of 427 asked to participate, 118 agreed, representing a 28% response rate. Physicians or office staff then asked patients meeting the ACR criteria for FM (5) if they were interested in the study. Patients completed cards indicating their interest, and the physician's office staff returned the cards to the University of Connecticut in prestamped envelopes. Three hundred twenty-four female patients who met the ACR criteria and had no other chronic health conditions were referred. University staff contacted these patients by telephone, and 287 (89%) agreed to participate. Participants were interviewed by phone at baseline and completed 4 followup interviews. Participants were paid $25 for each interview. The final sample for this analysis consisted of 241 participants (84% of the original sample) who had at least 2 observations used to calculate slopes. Although multilevel modeling techniques are well suited for analysis on data with only 1 observation, when too many participants have too few observations, estimation problems arise (30). Consequently, 46 participants (16% of the original sample) who had only 1 observation of data were excluded, leaving 241 participants who had at least 2 interviews for the analyses. Two hundred eleven (74%) participants completed all 5 interviews.

This study was approved by the University of Connecticut Health Center Institutional Review Board.

Measures.

The study included 4 measures of health status that assessed the major symptoms of FM, employment status, disease duration, and sociodemographic characteristics.

Data on health status measures included pain, fatigue, functional status, and depressive symptoms. Pain on the day of the interview and fatigue in the past week were both measured on analog scales. Participants were asked, “On a scale of 0–100, with 0 being no pain at all and 100 being the most pain possible, how much pain do you feel today?” Fatigue was measured by asking participants, “For the following question, indicate on a scale from 0–100 how you have been feeling in the past week. To what degree have you experienced fatigue, from 0 (not at all) to 100 (a great deal)?” The Modified Health Assessment Questionnaire (M-HAQ) (31) assessed functional status. The scale consists of 8 items assessing difficulty, satisfaction, and change in the previous 6 months. Each item is rated 0–3 points, with higher scores indicating worse function. The Center for Epidemiologic Studies Depression Scale (CES-D) is a 20-item scale assessing frequency of depressive symptoms (32). Scale scores vary from 0 to 60, with higher scores indicating more depressive symptoms. Mean community scores are ∼9, and 16 is considered indicative of clinical depression.

Disease duration was reported by the participant as the number of years since diagnosis. Sociodemographic characteristics included age measured in years; education measured in years completed; family income grouped into 3 categories: <$30,000, $30,000–49,999, or ≥$50,000; race as white or nonwhite; and marital status defined as married or not married. Employment was measured by 1 self-report item: “Are you employed for pay outside the home?”

Changes over time were the critical outcomes assessed in these growth models. Each interview was scheduled to take place at approximately the 1-year anniversary date of the previous interview. In order to more precisely evaluate the effects of time, because individuals might vary on the time between interviews, the time between interviews was calculated for each individual time point by the number of days between interviews.

Statistical analysis.

Data were analyzed using full maximum likelihood in SAS PROC MIXED (SAS Institute, Cary, NC). Time was measured as the actual amount of time that elapsed between interviews. Multilevel modeling can not only handle unstructured time but also produce more precise estimated growth rates and reduce the estimated variance components (33). Demographic variables and disease duration, which were used as control variables in the modeling, were centered around their grand mean in order to aid in final model interpretation. The employment status variable was left uncentered in order to interpret the effects of those who were employed and those who were not.

Modeling.

Our approach to the modeling strategy was to first assess the unconditional growth models estimating unadjusted rates of change in the health status measures. We then investigated conditional growth models that looked at the fixed effects of baseline employment status on health status measures over time, adjusting for demographic characteristics and disease duration. Each control variable was entered into the model along with its interaction with time. If the interaction was a trend (P < 0.10), both the main effect and the interaction were retained in the model. If the interaction was not significant (P > 0.10), the interaction term was removed from the model and the main effect was examined for significance level. If the main effect was nonsignificant, the variable was removed from the model. Once all the terms that were significant or trending toward significance were examined, we determined whether the variables trending toward significance were confounders. In order to be considered a confounder in the model, the effect of the employment status estimate must have changed >20% and changed in significance. If these criteria were met, the variable was retained in the final model. Finally, the deviance statistic between the final model and the conditional growth model with the employment status variable were compared, using a chi-square test to determine whether the model adjusting for the covariates was a better model fit. Bootstrapping methodology adjusted for the cluster sampling design of selecting patients within physician practices.

RESULTS

Description of the sample.

There was a good retention rate, with 241 (84%) participants completing at least 2 interviews and 211 (74%) participants completing all 5 interviews. The only significant difference between those who remained in the study and those who dropped out was that dropouts were more likely to be nonwhite. The demographic characteristics of the sample at the first interview are shown in Table 1. Women were an average of 47 years old, and most were married, white, and had more than a high school education. The sample was fairly affluent, with 40% reporting a family income of ≥$50,000. Approximately half were employed outside the home for pay.

Table 1. Baseline demographic characteristics of participants with fibromyalgia*

	Participants (n = 241)
* Values are the mean ± SD unless otherwise indicated. M-HAQ = Modified Health Assessment Questionnaire; CES-D = Center for Epidemiologic Studies Depression Scale.
Age, years	47 ± 10.8
Married, %	63.5
Non-Hispanic white, %	90.5
Education, years	14 ± 2.5
Income, %
<$30,000	29.9
$30,000–49,999 (referent)	29.9
≥$50,000	40.2
Currently employed, %	50.2
Disease duration, years	4.9 ± 4.4
Reported pain today score	57.2 ± 24.2
Reported fatigue score	75.4 ± 21.8
M-HAQ score	0.73 ± 0.46
CES-D score	22.9 ± 13.4

The average disease duration was 4.9 years, although 13% of the participants had been diagnosed for <1 year. Participants reported high levels of symptoms. The mean ± SD pain score was 57.2 ± 24.2 on a scale of 0–100. Fatigue was quite high at a mean ± SD score of 75 ± 21.8, and the distribution of scores was skewed toward the high end of the scale. Participants had high levels of functional disability measured by the M-HAQ, with a mean ± SD of 0.73 ± 0.46, similar to women with rheumatoid arthritis. Finally, participants had high CES-D scores, far exceeding the score of 16 indicative of clinical depression, with a mean ± SD of 22.9 ± 13.4. A score of 22 is approximately that of people hospitalized for depression.

Unconditional growth models: changes in health status over time.

Slopes were generated with mixed model methods to estimate the unadjusted rate of change over the observation period. Pain did not change significantly over time (slope = −0.299). There was significant improvement in fatigue (slope = −1.02; P < 0.01), the M-HAQ (slope = −0.027; P < 0.001), and depressive symptoms (slope = −0.52; P < 0.01). These slopes can be interpreted to indicate that for approximately each year of observation, fatigue decreased by 1.02 points, function improved by 0.027 points, and depressive symptoms improved by 0.52 points. Although these are relatively small incremental changes, they are highly significant and could be clinically meaningful. A 2.08 point decline in CES-D score and 0.108 average decline in M-HAQ score could be important improvements in health status for these participants (Figure 1).

Figure 1. Trajectories of A, Modified Health Assessment Questionnaire (MHAQ), B, Center for Epidemiologic Studies Depression Scale (CESD), and C, fatigue over time.

Multivariate modeling.

Before conducting multivariate modeling, the data were examined for within-person correlations over time, linearity, and distribution of the dependent measures to assess the need for transformation. Random effects for time were included in the CES-D and M-HAQ models but not in the pain and fatigue models. Covariance parameters for the random effect of time were zero and nonsignificant and, consequently, were excluded from further models with these measures. In addition, the deviance statistics from the model with the random effect for time and the model without were compared using a chi-square test with 2 df. The test suggested that a model without random effects was a better fit for both the pain and fatigue models. The year 1 correlations with subsequent years among each health status measure are shown in Table 2. Correlations were significant and strong across the time points, demonstrating a decreasing magnitude across the time points, suggesting that alternative covariance structures should be examined. Consequently, the final models were tested with alternative covariance structures: unstructured, compound symmetry, heterogeneous compound symmetry, autoregressive, heterogeneous autoregressive, and Toeplitz. Because none of the covariance structures added substantially to the fit of the model, the standard covariance structure was maintained and used for the final models.

Table 2. Scores for pain, the M-HAQ, the CES-D, and fatigue in year 1, and Pearson's correlation coefficients between measures in year 1 and years 2–5*

Year	Pain	r	M-HAQ	r	CES-D	r	Fatigue	r
* Values are the mean ± SD. See Table 1 for definitions.
1	57.2 ± 24.20	–	0.73 ± 0.46	–	22.9 ± 13.44	–	75.4 ± 21.75	–
2	55.3 ± 25.32	0.39	0.67 ± 0.43	0.68	21.2 ± 12.95	0.67	71.7 ± 23.71	0.48
3	56.1 ± 24.39	0.42	0.68 ± 0.47	0.71	20.8 ± 12.71	0.65	74.4 ± 22.94	0.40
4	54.4 ± 24.33	0.33	0.64 ± 0.44	0.66	21.5 ± 13.19	0.62	72.5 ± 23.33	0.54
5	56.6 ± 26.24	0.39	0.61 ± 0.43	0.55	20.5 ± 12.78	0.57	70.3 ± 23.71	0.40

Individual plots of measures over time were visually inspected to identify the shape of the data. Linear trends were suspected, but additional tests of models were conducted with quadratic and cubic terms for time. In all models, these terms were not significant and were dropped.

In both the CES-D and pain, all time points and differences between time points were skewed. Normality assumptions for the multilevel models were met once the square root transformation for the CES-D measure and the log₁₀ transformation for pain were created.

The final results of the hierarchial linear modeling analyses for pain (using log transformation), fatigue, the M-HAQ, and the CES-D (square root transformation) are presented in Table 3. The first row of the table presents the intercept for each health status measure. Because of the techniques we used, the intercept can be interpreted as the average score on each measure at baseline after adjusting for the sociodemographic characteristics. The second row shows the fixed effects of employment status at the initial observation. Employed women had significantly lower fatigue and M-HAQ scores compared with women who were not employed. Employed women had, on average, a fatigue score 7 points lower than those who were not employed. Likewise, for the M-HAQ, women who were employed had scores, on average, that were 0.21 points lower than those who were not employed. The effects of employment on the CES-D were significant prior to adjusting for the cluster sampling, but the effects were reduced to a trend (P < 0.10) in the final analyses.

Table 3. Final results of the HLM analyses for pain, fatigue, the M-HAQ, and the CES-D*

	Parameter	Pain	Fatigue	M-HAQ	CES-D
* Pain measure transformed using log10, CES-D transformed using square root. HLM = hierarchial linear modeling; NS = not significant. See Table 1 for additional definitions. † P < 0.001. ‡ P < 0.10. § P < 0.05. ¶ P < 0.01. # Due to rounding, values for the M-HAQ look like 0. ** Model did not include the random effect; variance components for rate of change not estimated. †† Covariance between initial status and rate of change not calculated because the random effect for rate of change was excluded.
Fixed effects
Initial status, π0i
Intercept	γ00	1.72 (0.03)†	81.5 (1.9)†	0.81 (0.04)†	4.83 (0.2)†
Employment	γ01	−0.02 (0.04)	−7.05 (2.15)†	−0.21 (0.05)†	−0.34 (0.18)‡
Education	γ02	−0.01 (0.01)‡
Age	γ04	−0.005 (0.001)§	−0.52 (0.14)†		−0.03 (0.01)¶
Income
<$30,000	γ05	0.03 (0.05)	2.05 (2.39)	0.06 (0.06)	0.21 (0.19)
≥$50,000	γ06	−0.08 (0.04)§	−9.58 (2.79)¶	−0.17 (0.05)¶	−0.61 (0.22)¶
Nonwhite	γ08			0.18 (0.10)‡
Rate of change, π2i
Intercept	γ10	−0.01 (0.01)	−1.57 (0.61)§	−0.03 (0.01)†	−0.09 (0.03)¶
Employment	γ11	NS	NS	NS	NS
Disease duration	γ13	0.02 (0.01)§	NS	0.01 (0.01)‡	0.06 (0.02)§
Age	γ14		0.10 (0.05)‡
Nonwhite	γ18		3.66 (1.62)§		0.18 (0.09)§
Variance component
Level 1
Within-person	σ^2ε	0.08 (0.004)†	297.2 (13.2)†	0.06 (0.00)†	0.762 (0.04)†
Level 2
Initial status	σ	0.04 (0.006)†	192.8 (23.66)†	0.12 (0.02)†	1.37 (0.17)†
Rate of change#	σ	**	**	0.00 (0.00)¶	0.052 (0.02)¶
Covariance	σ	††	††	−0.01 (0.00)	−0.09 (0.04)§

Age and income also were significantly related to health status. Younger women tended to report higher levels of pain, fatigue, and depressive symptoms. Women in the highest income group had lower pain, fatigue, M-HAQ, and CES-D scores compared with those in the middle income group.

The results related to changes in health status over time are presented in the second half of Table 3. The intercept row indicates the rate of change in each health status measure, adjusting for the fixed effects and other interaction effects. Pain did not change significantly over time. The other 3 measures improved significantly over time (shown by the significant negative intercepts). Fatigue score had an average ± SE decline of 1.22 ± 0.61 points per year, the M-HAQ score had an average ± SE decline of 0.03 ± 0.01 points per year, and the CES-D score had an average ± SE decline of 0.49 ± 0.9 points per year.

There were no significant time by employment interactions, indicating that employed women had better health status at the start of the study and maintained that advantage over 5 years. However, there were several significant time interactions with other covariates, including disease duration, age, and race. Those with the longest disease duration tended to experience increasing pain, and those who were diagnosed more recently tended to have declining pain. A similar pattern existed for CES-D score, time, and disease duration. Those with the longest disease duration tended to experience increasing depressive symptoms, and those who were diagnosed more recently tended to have declining levels of depression. Younger women tended to have declining fatigue (P < 0.10), and older women tended to have increasing fatigue. Nonwhite women reported increasing levels of fatigue and depressive symptoms, and white women tended to have decreasing levels of fatigue and depressive symptoms.

The results on the variance components and the goodness-of-fit statistics are presented in the final sections of Table 3. The level 2 statistics on initial status, rate of change, and covariance provide information on the slopes and intercepts for each measure. Analysis of the rate of change and the covariance components shows that there were no random slopes identified for pain and fatigue. This indicates that the women in this study all had common trajectory over time and no significant random effects were associated with within-person variation. In contrast, for the M-HAQ and the CES-D, slopes did vary randomly, indicating that each woman had a unique experience over time for these health status measures. Furthermore, the covariance measure demonstrates that the slopes not only varied randomly, but that the slope varied as a function of the initial scores. Where women started out at the baseline measure influenced the trajectory of their disease course.

DISCUSSION

The results of studies of FM agree that patients with FM experience major psychosocial impacts associated with their condition. These studies have reported varying results on the prognosis of FM in terms of the major impacts of the disease: pain, fatigue, depressive symptoms, and functional status. The results of our study indicate that women with FM in this sample improved on fatigue, the M-HAQ, and the CES-D over the observation period. Although these improvements were relatively small, they could be clinically meaningful, particularly for the M-HAQ, which demonstrated a 4% reduction in disability per year. Pain did not change significantly.

A major factor that has been shown to provide a protective health benefit over time among women in community studies has been employment. Studies in the general literature on women's health show that employed women are not only healthier, but that their health status declines more slowly compared with women who are not employed. A goal of this study was to assess whether women with FM experience the same health advantages. As is found in most studies of women and health status, women in this study who were employed at the baseline interview reported better health status and continued to maintain that advantage over the observation period. There was no interaction between employment and time on the health outcomes studied, indicating that employment did not provide an increasing health advantage to women with FM over time. However, employed women with FM also did not experience worsening health outcomes, and they managed to maintain better health status over time. This finding suggests that women with FM can remain employed with no negative consequences to their condition, and probably should try to remain in the labor force as a strategy to maintain better health.

Several demographic characteristics affected both the initial status of the health status measures as well as disease trajectories. A negative coefficient indicated that younger women reported higher pain, fatigue, and CES-D scores. This finding was surprising because we would expect younger women to be healthier. This may signify better adjustment to FM among older women. Income also was negatively associated with all health status measures. This finding reflects the advantage of higher socioeconomic status among women with FM.

Disease duration and race/ethnicity significantly influenced disease trajectories as shown by the interaction of this variable with time for the CES-D and pain. Those with the shortest disease duration had the greatest improvement. This finding may illustrate a natural disease process in which some women improve over the short term and their condition resolves, but others experience intractable disease.

In this study, nonwhite women reported increasing levels of fatigue and depressive symptoms and white women tended to have decreasing levels of fatigue and depressive symptoms. These effects are independent of socioeconomic status, which was controlled for by family income. Many studies have demonstrated the existence of health disparities among racial and ethnic minority groups in the US (30) across a broad spectrum of diseases, as well as in health care access and in the quality of health care received. A recent review of patients with rheumatic diseases (34) cited several studies that demonstrated health disparities for these conditions. In that review, African Americans had higher standardized death rates from arthritis and other rheumatic conditions in the US from 1979 through 1998. African Americans also had a higher prevalence of osteoarthritis but fewer joint arthroplasties compared with whites. Therefore, it is not surprising that the current study found that being nonwhite was associated with worse health outcomes, measured by worse depression scores and greater fatigue, compared with being white. The number of nonwhites in our sample was relatively small, ∼10% of participants, but the impact of race/ethnicity must be fairly large to reach statistical significance. We investigated several potential mediating factors, including socioeconomic status, social support, and family factors, but these variables did not explain the differences in these health outcomes. As other investigators in the area of health disparities suggest, the underlying psychosocial and biologic factors contributing to these disparities should be investigated further.

An interesting finding was that there were no random effects demonstrated for the major symptoms of FM, pain and fatigue. This suggests that the participants experienced similar patterns of pain and fatigue over time, which could be related to an underlying biomedical mechanism of FM common across women with FM. In contrast, the impacts of the disease (disability and depression) do vary individually among women over time. The relationships between symptoms and outcomes and factors that mediate outcomes should be investigated.

A strength of this study is that it was a national sample drawn from rheumatology practices. However, there are several limitations as well. Because of the nature of the sample, the results are not generalizable beyond women being treated in rheumatology practices. There could be selection bias in that women with more serious disease or worse symptoms may have volunteered for the study. The participants had FM for varying lengths of time, and this study only provides a brief snapshot of the experiences of these women. As shown by our results on disease duration, this factor should be taken into consideration in future studies. Recently diagnosed patients may have very different experiences than those with long-term disease. Finally, nonwhite women had very different experiences of FM compared with white women. Our sample included a relatively small number of nonwhite women. This finding is suggestive, but should be viewed cautiously. The effects of race and ethnicity should be investigated further.