Stewart, Walter F. PhD; Wood, G. Craig MS; Bruce, Christa MPH; Buse, Dawn C. PhD; Runken, M. Chris Pharm D; Lipton, Richard B. MD
* Demonstrate familiarity with previous studies of the economic impact of migraine in the workplace, including the nature of the effect on lost productive time (LPT).
* Summarize the findings of the new longitudinal study in terms of how changes in days with migraine headache affect LPT.
* Discuss the study implications for treatment of migraine headaches to reduce the impact on LPT.
Studies of the impact of migraine in the workplace have repeatedly shown increased absenteeism,1–18 and reduced performance while at work (presenteeism).1 More recent findings show substantial variation in the indirect cost impact of migraine by frequency of headaches.1 Notably, however, all previous studies of indirect costs have been based on cross-sectional samples. Notably, with one exception,19 all previous studies have not been able to examine how costs change over time in relation to change in attack frequency.
Lost productive time (LPT) is a widely used proxy for indirect cost estimates. LPT is the sum of absenteeism and presenteeism, weighting time at work by the extent to which a health problem results in reduced effectiveness. For migraine1 and other symptomatic conditions,20,21 presenteeism accounts for the majority of the LPT. Estimates of LPT due to headache in persons with migraine range from 74 to 96 hours per year.1 Recent evidence from the American Migraine Prevalence and Prevention (AMPP) study indicates that those with more frequent episodic migraine account for a disproportionate share of LPT.1 In analysis of cross-sectional data, the association between attack frequency and LPT time is not linear.1
As headache frequency increases, previous findings suggest that the work impact per attack decreases in a nonlinear manner.1 This nonlinear effect could be due to less severe pain per attack as headache frequency increases or to other changes that co-occur with an increase in headache frequency (eg, seeking medical care, using preventive agents, developing depression) that could modify the impact of headache on work. Alternatively, this nonlinear effect could simply be an artifact caused by those with the most severe frequent headaches becoming unemployed (ie, no longer contributing to the impact of headache frequency on LPT).1 Finally, it is not known if the cross-sectional relationship between headache frequency and work impact is consistent with what would be observed longitudinally.
We used longitudinal data from the AMPP study20,21 to quantify the relationship between changes in days with headache (ie, headache-days) among those with migraine and related changes in work impact. To determine whether the previous findings of a nonlinear effect is an artifact of employment status, we completed separate analyses of the relation between change in headache frequency and change in (1) LPT among individuals who were employed in each of the 2 consecutive years; (2) employment status (ie, employed or not) between each of 2 consecutive years; and (3) overall work productivity status by using a composite measure that incorporates changes in LPT and unemployment status.
Data for this study, including measures of employment status and LPT, were based on survey responses to the 2005 through 2008 AMPP surveys among individuals 18 years of age and older.4,22 The AMPP was modeled on the methods of the American Migraine Studies I and II, described in detail elsewhere.23,24 In brief, a self-administered headache questionnaire was mailed in 2004 to a stratified random sample of 120,000 US households, drawn from a 400,000-household nationwide panel maintained by the Taylor Nelson Sofres/National Family Opinion, Inc. The panel is constructed to match the US Census on age, gender, household income, household size, and census region. A screening questionnaire was completed by the head of the household, who reported the total number of household members and the number of household members suffering from at least occasional self-defined severe headache. A total of 257,329 household members were contacted (132,674 females, 124,655 males) and data were obtained on 162,756 individuals, including 85,571 females (64% response rate) and 77,185 males (62% response rate).23 Each household member with severe headaches was asked to complete a clinically validated questionnaire25,26 to assess all headache features important to the detection of migraine (ie, taking the first edition of the International Classification of Headache Disorders as a reference).27 Subsequently, a second questionnaire was mailed to all individuals who met criteria for migraine or severe headache. The questionnaire collected detailed data on demographics, headache features, and related symptomology, impact of headache on functioning (ie, Migraine Disability Assessment [MIDAS]) including work impact, conditions, and employment status, among other measures. In 2005, a modified version of the AMPP questionnaire was sent to a random sample of adult (ie, ≥18 years of age) respondents who reported active (ie, past 12 months), severe headache in the screening survey. Of the 24,000 household members sent the questionnaire, 18,514 responded (77.1% response rate). From 2006 to 2008, the AMPP was mailed to the same population on an annual basis, with some modifications to components that were removed and added.
Migraine Case Definition
Respondents were asked to indicate the number of days they had headache in the preceding 3 months and the number of days with their “most severe type of headache” over the preceding year. Active migraine (ie, at least one severe headache in the previous year) cases were defined using ICHD-2 criteria27 (ie, headache with unilateral or pulsatile pain, and either nausea, vomiting, phonophobia with photophobia, or visual, or sensory aura before the headache). Migraine case status was defined for each year that an AMPP questionnaire was completed to allow for the possibility that individuals could remit or, more generally, transition from an active migraine to inactive migraine state.
The relation between change in headache-days and change in work impact, including employment status was examined using data from respondents who completed consecutive year AMPP surveys between 2005 and 2008 (ie, defined as “couplets”). An individual could contribute from one to three couplets, depending on their pattern of responses. Individuals were excluded from the analysis if they completed only two AMPP surveys separated by a year where an AMPP survey was not completed. Couplets were limited to respondents who met criteria for migraine as previously defined, had reported employed status on at least one of the five surveys, and had answered at least one of the work related MIDAS question in consecutive years.
Predictor and Outcomes Variables
Headache-days was the primary predictor variable of interest in the analysis. We used change in headache-days (ie, defined as the difference between the first and second of 2 consecutive years) from 1 year to the next as one of two primary predictor variables because it is approximately normally distributed compared to the highly skewed distribution (ie, mode of 0 or 1) of headache-days for which it is not possible to transform to a normal distribution. LPT was estimated from the MIDAS questionnaire. Two of the five MIDAS questions ask about number of days of missed work in the past 3 months due to headache and number of days while at work where productivity was reduced by half (ie, denoted half days) or more because of headache. We defined LPT as:
LPT = Missed workdays + 0.5 × (Workdays where productivity was reduced by half) The LPT formula is for the number of missed workday equivalents in the 3 months before the survey. We assumed that a workday reduced by half or more is equivalent to working for half of a usual work day, an estimate that was validated in a previous diary study.28 The minimum value for LPT is 0 and the maximum LPT value was set at 60 (ie, 20 working days per month for 3 months). Change in LPT was defined as LPT in the first year minus LPT in the second year of a couplet.
Employment status was defined as a binary variable. Individuals were defined as employed for pay (ie, excludes those on short or long term medical leave) if they reported actively working enough hours to qualify for full (≥35 hours) or part-time (20 to 34 hours) status. Individuals were defined as “unemployed” if they reported they were unemployed, on medical leave, or disabled.
Given that individuals can contribute more than one couplet to the analysis, we used Generalized Estimating Equations (GEE)29 to account for the effect of correlation among repeated measures on confidence intervals estimates. For continuous outcome measures (ie, change in LPT), generalized linear mixed models (GLIM) were used. GEE with the logit link function was used to model the dichotomous variables for employment status. Separate models were run for those who were employed or unemployed in the first year of a couplet given the possibility that the process of changing status depends on baseline status. Moreover, separate models were run when defining employment status at baseline as full-time only, full- or part-time, and unemployed. For all analyses, standard errors were estimated after imposing a correlation/covariance matrix structure to account for the effect of longitudinal repeated measures on confidence interval estimates.
In regression models, we included covariates to account for nonlinear (ie, squared and cubed terms for centered values of change in headache-days) relations between the primary dependent and independent variables. Baseline headache-days was included as a categorical covariate (ie, 0 to 3, 4 to 14, ≥15 headache-days per month) given that the primary predictor variable was represented as a measure of change and that the effect of change may be confounded by baseline headache-days itself. Preliminary analysis indicated that the 3 headache-days categories were sufficient to account for variance explained by a larger number of categories. Because change in headache-days may also be modified by baseline status, interaction terms between change in headache-days and baseline headache-days were modeled. Covariates for all regression models also included the following known confounders: sex, age (18 to 29, 30 to 39, 40 to 49, 50 to 64, ≥65 years of age), race (white, black, other), education (no high school diploma, high school diploma, or GED, some college or associates degree, bachelors degree, graduate degree), health insurance status (insured, not insured), and average pain intensity (ie, reported on a scale of 0 = “no pain” to 10 = “pain is as bad as it could be”).
SAS version 9.1 was used for all analyses (SAS Institute Inc, Cary, NC).
There were 12,791 respondents who met criteria for episodic or chronic migraine on one or more surveys and reported they were employed (ie, working full or part time) or unemployed (ie, unemployed, on medical leave, disabled) in at least one survey from 2005 to 2008. A total of 4697 respondents were excluded because they did not complete surveys in consecutive years, leaving 8094 respondents with 16,016 couplets. Of these, 993 couplets were excluded from the analysis because data on headache-days were missing on one or both years of a couplet.
Approximately 54% of the couplets (n = 8610) used in the analysis were from individuals (n = 2870) who contributed three couplets (Table 1). Compared to all potentially employable AMPP respondents with migraine (ie, Table 1, first and second columns), those who contributed one or more couplets to the analysis were similar on demographic factors (ie, sex, age, education, income), health insurance status, body mass index, and headache-related factors. Those who contributed one or more couplets were more likely to be employed full-time, especially for those contributing two or three couplets. Of those who contributed one or more couplets, 68% reported full-time employment during at least 1 year from 2005 to 2009; another 19% reported part-time employment. Of the remaining 13%, 10% reported being unemployed. The proportion reporting unemployment decreased in relation to the number of couplets contributed to the analysis.
TABLE 1-a. Profile o...Image Tools
LPT and Headache Days
TABLE 1-b. Profile o...Image Tools
In all analysis discussed herein, LPT is expressed in lost workday equivalents/3 months and headache frequency is expressed in headache-days/month. Among those who were employed during both years of a couplet, the mean lost workday equivalents/3 months was 1.46 days/3 months (Table 2) across all years; half of the respondents reported some LPT. The mean change in LPT/3 months was –0.11 days/3 months. The median change was 0. The change was less than or equal to 0.1 days/3 months for approximately 44% of the couplets; in 6.3% of the couplets, the change in LPT was 5 days or more/3 months. The mean headache-days/month was 3.6 and the interquartile range was 1 to 4 headache-days/month. The mean change in headache-days/month was –0.22.
Change in LPT had a nonlinear relationship with change in headache-days among those who were employed in both years of a couplet. The linear, quadratic, and cubed terms for change in headache-days were all significantly associated with LPT (Table 3, Model 1). Baseline headache-days was also associated with change in LPT (Table 3, Model 2), but the association with the latter did not influence the coefficients for linear and nonlinear covariates for change in headache-days. Interaction terms between baseline headache-days and change in headache-days were statistically significant and reduced the change in headache-days coefficients for the quadratic and cubed terms, as well as, the coefficients for baseline headache-days (Table 3, Model 3). The headache-day covariates in Model 3 were not influenced by the addition of other covariates (Table 3, Model 4), a finding consistent with the likelihood ratio tests indicating that Model 4 did not offer a superior fit to the data over Model 3. Model 3 indicates that the change in LPT is different by both the baseline level of headache-days and the level of change in headache-days. The graphical representation of Model 3 (Fig. 1) shows the nonlinear relation between change in headache-days and change in LPT, as the effect asymptotes at the higher negative or positive levels of change in headache-days. Figure 1 also shows that the effect of change in headache-days varies by number of headache-days at baseline and that the asymptotic effect is staggered in relation to change in headache-days.
LPT and Employment Status
GEE logistic regression analysis was completed to determine the relation between change in headache-days and transition from full- or part-time employment to unemployed status (Table 4). In a model with the linear term only for change in headache-days, the odds ratio for employment was 0.9858 per increase in each headache-day (95% CI = 0.9627–1.0095). The quadratic term was statistically significant (Model 2, Table 4), suggesting that the effect of change in headache-days on employment status is nonlinear. None of the other terms (Table 4, Models 3 and 4), including a cubed term for change in headache-days, baseline headache-days, or interaction terms between these two covariates, were statistically significant. This same pattern was observed when employment was more strictly defined as full-time work (ie, Table 4, lower panel). However, none of the odds ratio estimates are significantly different from 1.0.
For 1718 couplets, individuals reported their work status in the first year of the couplet as unemployed, on medical leave, or on medical disability. In the second year of the couplet, 26% were working full- or part-time. In logistic regression analysis (Table 5), only baseline headache-days was associated with full- or part-time employment in the second year of a couplet. Specifically, individuals who had 15 or more headache-days/month in the first year, had significantly lower odds of being gainfully employed in the second year compared to those with 0 to 3 headache-days/month. The statistically significant interaction term (Table 5) indicates that the effect of baseline headache-days on remaining unemployed in the second year is substantially greater for those with 15 headache-days or more/month.
Composite Effect of LPT and Employment
We developed a composite measure of work impact by including individuals who were employed in 1 year but not the other (added 967 couplets from 509 respondents). For these couplets, we assumed that transitioning from full-time to not employed (unemployed, medical leave, or disabled) was equivalent to a change in MIDAS LPT of 60 (ie, missed 60 working days in 3-month period). Similarly, we assumed that going from part-time to not employed was equivalent to a change in MIDAS LPT of 30. In addition, we assumed the same magnitude of change if the change was in the opposite direction. The value of 60 was chosen for full-time work because there are approximately 60 work days in 3 months; 30 was chosen for part-time work because part-time workers in the 2005 AMPP survey had approximately half as many work hours as full-time workers. The results of the GLIM regression (Table 6) indicate that change in LPT is strongly and linearly related to change in headache-days. Moreover, there are no apparent effects from baseline headache status or from interactions between baseline headache status and change in headache-days. This model indicates that a change in one headache-day is associated with a 0.24 change in LPT day equivalent.
It is well-established that migraine headaches are associated with lost productive time (LPT). Evidence is largely based on reports of recent headache experience (eg, past 3 months to past week) in one of two forms. Individuals first report on LPT and the related cause. Alternatively, individuals report on their headache experience and the associated LPT. When using either method, a majority of the LPT associated with migraine takes the form of presenteeism or reduced performance at work.1 While headache-days is known to change from year to year, it is not known what impact this change has on LPT or employment status. Using year-to-year AMPP survey data, we found that a change in one headache-day/month from 1 year to the next was associated with a change in LPT, but the effect diminishes as the change in headache-days is less than –15 or greater than 15. We also found that an increase in one headache-day from 1 year to the next was associated with decreased odds of employment in the second year of a couplet. When these two analyses are completed separately, change in work impact has a nonlinear relationship to change in headache-days. This nonlinear effect appears to be an artifact. When we used a composite measure of work impact (ie, for both change LPT and change in employment status), change is headache-days is linearly related to change in work impact.
In a previous study, we showed that LPT was similar among migraineurs with chronic migraine and those with 10 to 14 headache-days/month, but that the unemployment rate was higher among those with chronic migraine.22 We hypothesized that the nonlinear or asymptotic effect of change in headache-days on change in LPT (ie, Table 3; Fig. 1) may be explained by individuals transitioning from employed to unemployed status. Specifically, as headache frequency increases over time (ie, transitioning from high frequency headache to chronic migraine), individuals may initially experience higher rates of absence time and presenteeism. Those who continue to lose the most LPT as headache-days increase would logically be the ones who are at highest risk of unemployment; when these individuals transition to unemployed status they are no longer represented in the numerator or denominator statistic for estimating LPT. In our previous analysis,29 we suggested that a steady state may be maintained, where proportionately more of those with higher frequency and high impact headaches leave the workforce and then return as headache frequency declines.
We examined the above question in a single model (Table 6) to assess all work-related effects of change in headache. This model finding suggests that the overall effect of migraine on LPT may be relatively straightforward to understand when a combined measure of work impact and under-employment is used. We recognize, however, that such a combined measure represents the interests of employers (ie, LPT among employed workers and the cost of workers becoming unemployed) and society (ie, unemployment) that are logically and ordinarily represented as separate domains.
The above findings suggest that there may be a return on investment for improving the treatment of migraine headaches. Effective use of the optimal acute treatment and advice to use such treatments when the pain is mild30 yield a potentially valuable return in reduced work impact. Importantly, use of preventive pharmacologic treatments and empirically supported nonpharmacologic interventions among those with a high number of headache-days may offer a valuable return in either maintaining employment status or facilitating a return to work. Intervention with both acute and preventive agents is likely to yield meaningful reductions in work impact among the latter, given that previous analyses suggest that this group accounts for a disproportionate share of the headache-related work loss. In future studies, it will be important to examine whether use of acute and preventive agents together either prevents unemployment or increases the likelihood of return to work.
In our analysis, we estimated the work burden for all headache-days in persons with migraine, rather than migraine headache-days only. Evidence indicates that most or all headache that occurs among those with migraine responds to acute migraine-specific treatment31 regardless of the overt symptoms. This suggests that there is a common underlying pathophysiology for headaches among those with migraine. Moreover, treating migraine headaches while the pain is mild, following current recommendations, improves treatment response. As a consequence, the diagnostic threshold for a migraine attack may not be crossed because of this early treatment effect.30
Interpretation of results from this study may be limited by several factors. The findings rely on self-reported data for headache symptoms, headache-days, LPT, and employment status. Errors in self-report could lead to biased estimates, but the expected effect of such bias is likely to be toward a null finding or weaker association between LPT and headache-days, not toward an over-estimate of the strength of such associations. On the basis of previous work, ascertainment of migraine status using AMPP survey questions has proven to be valid.25,26 In previous analysis,22 we used the Work and Health Questionnaire and a 2-week recall period, which has been validated.32,33 However, in the current analysis, we were limited to the two MIDAS questions on work impact of headache and a 3-month recall period. Previous studies indicate that the MIDAS score and the specific questions on work impact are highly reliable34 when evaluated in test-retest studies and highly valid28 when compared to measures based on a 3-month diary study. It is possible that differences in productivity could be due in part to seasonal differences from the time of year that the surveys were distributed. Finally, we note that the AMPP study population is a volunteer cohort, selected to be representative of the US population with regard to selected demographics. Bias from selective participation is possible. However, prevalence estimates of migraine and other migraine features from AMPP are similar to estimates from previous studies using more rigorous sampling methods.35
The American Migraine Prevalence and Prevention Study is funded through a research grant to the National Headache Foundation from Ortho-McNeil Neurologics, Inc, Titusville, New Jersey, with additional funding provided by GlaxoSmithKline, Research Triangle Park, North Carolina.