Depression costs US employers tens of

Transcription

1 JOEM Volume 50, Number 7, July Comparative and Interactive Effects of Depression Relative to Other Health Problems on Work Performance in the Workforce of a Large Employer Ronald Kessler, PhD Leigh Ann White, PhD Howard Birnbaum, PhD Ying Qiu, PhD Yohanne Kidolezi, BA David Mallett, MA Ralph Swindle, PhD Objectives: To present data on the comparative and interactive workplace costs of depression relative to other health problems in the workforce of a large employer. Methods: The World Health Organization Health and Work Performance Questionnaire was used to assess self-reported health problems and work performance. Survey data were linked to medical-pharmacy claims data. Regression analysis was used to assess comparative effects of depression in the absence and presence of comorbidities on Health and Work Performance Questionnaire measures of work performance. Results: Depression had the largest individuallevel effect on work performance of any condition examined. Several comorbid conditions exacerbated the effect of depression, but had no effects in the absence of depression. Conclusions: Depression is a strong predictor of decrements in work performance. Other conditions that often co-occur with depression, including anxiety and fatigue-sleep disturbance, exacerbate the adverse effect of depression. (J Occup Environ Med. 2008;50: ) From the Department of Health Care Policy (Dr Kessler), Harvard Medical School, Boston, Mass; Analysis Group, Inc. (Drs White, Birnbaum, Qui, Mr Kidolezi), Boston, Mass; Ingenix, Inc. (Mr Mallett), Rocky Hill Conn; and Eli Lilly and Company (Dr Swindle), Indianapolis, Ind. Disclosure: One of the authors, Ralph Swindle, is an employee of Eli Lilly and Company. Address correspondence to: Ronald Kessler, PhD, Department of Health Care Policy, Harvard Medical School, 180 Longwood Avenue, Boston, MA 02115; kessler@hcp.med.harvard.edu. Copyright 2008 by American College of Occupational and Environmental Medicine DOI: /JOM.0b013e318169ccba Depression costs US employers tens of billions of dollars each year in lost work productivity. 1,2 Indeed, comparative cost-of-illness studies show that depression is among the most costly of all health problems to US employers in terms of productivity loss. 3 5 Although guideline-concordant depression treatment can be effective in reducing these workplace costs, 6 10 epidemiological studies show that only a minority of depressed workers receive treatment and that this treatment is often of poor quality. 11,12 Experimental studies show that depression screening and disease management programs can significantly increase the proportion of cases that receive treatment as well as improve the quality of care, but the cost-effectiveness of such interventions is still a matter of uncertainty. 16 Cost-effectiveness presumably could be increased by targeting interventions to the subset of depressed workers who are most likely to have adverse workplace effects, assuming equal effectiveness of the intervention across the range of this risk gradient, but no principled rules exist for such targeting. One important differentiating characteristic that might be useful in targeting interventions is comorbidity of depression with other physical and mental disorders. Epidemiological studies document significant comorbidity of depression with a wide variety of other disorders 17,18 and significant synergistic effects of de-

2 810 Depression and Work Performance Kessler et al pression with comorbid disorders in predicting various aspects of work impairment and performance Based on these results, it might be useful to target depressed workers with particular types of comorbidity for special outreach and intervention programs. To do this, though, information would be needed about the particular constellations of comorbid conditions that magnify the adverse effects of depression on work productivity. The current report presents data from a health and productivity survey carried out in a large information technology firm aimed at generating data of this sort. Methods Sample An attempt was made to administer the World Health Organization Health and Work Performance Questionnaire (HPQ) 22,23 in the fourth quarter of 2005 to all of the approximately 20,000 employees with at least 11 months of medical-pharmacy enrollment from October 1, 2004 through September 30, 2005 who worked for a large national information technology firm. The HPQ was selected for use in this survey because it is the most widely used instrument in the literature on health and work productivity and one of the few instruments to include validated measures of absenteeism and work performance. The 7320 survey respondents who completed the HPQ represented 37.3% of eligible workers in the firm in the fourth quarter of A propensity score adjustment weight 24 was used to adjust the sample for potential nonresponse bias based on a comparison of basic sociodemographic data and medicalpharmacy claims data for survey respondents versus nonrespondents. The hope in making this adjustment was that the estimates of prevalence and associations found in the weighted sample would more closely represent those in the total workforce than would estimates based on unweighted data. Work Performance The HPQ includes validated selfreport assessments of absenteeism in the 30 days before interview and performance while on the job in the 30 days before interview. Blinded validation studies have documented significant associations (r 0.61 to 0.87) of HPQ absenteeism reports with employer payroll records 23 as well as significant associations of HPQ performance reports with supervisor ratings (0.52) 22 and other administrative indicators (0.58 to 0.72) 23 of job performance. As the firm studied here had an integrated benefits policy that provided a fixed number of paid sickness absence and/or vacation days each year, these days had to be used or lost by the end of each year. This policy led to virtually all workers taking all of their days of paid absentee or vacation leave, resulting in very little variation in absenteeism among respondents. Based on this low variation in absenteeism, we focused on variation in job performance as the outcome of primary interest. HPQ ratings of job performance begin with priming questions followed by a series of summary questions that ask respondents to rate the performance of the average employee on their job, their own typical performance, and their performance in the past 30 days on a self-anchoring visual analog scale ranging from 0 (least productive) to 10 (most productive). A relative performance rating was constructed by dividing the employee s rating of their own performance by their rating of the performance of the average worker. Although this ratio is undefined in the special case where the denominator equals zero, no such cases were observed in this sample. Extreme scores were truncated to have a lower bound of 0.5 and an upper bound of 2.0 based on the observation in previous HPQ validation studies that the small numbers of respondents with scores outside this range do not differ meaningfully from those with extreme scores within the range on objective measures of work performance. This truncated version of the outcome measure was then rescaled to be in the range between 0 (the lowest relative performance) and 100 (the highest relative performance). The resulting transformed measure, which has a mean of 51.8 and a standard deviation of 1.8, is the outcome used here. Health Problems Health problems were assessed with a chronic conditions checklist of the sort used in the US National Health Interview Survey. 25 Such checklists have been shown to yield more complete and accurate reports of health problems than estimates derived from responses to openended questions. 26 The conditions included in the HPQ were selected from the much larger National Health Interview Survey list to focus on conditions that are relatively common among working people and that have been found in previous research to be meaningfully related to days out of role. 23 Methodological studies have documented good concordance between reports of treated condition in such checklists and medical records, although the HPQ separately assesses treated and untreated conditions in order to go beyond the assessment available in medical claims records and to allow for evaluation of the effects of untreated conditions. Nine broadly-defined classes of conditions were assessed in the HPQ checklist: cancer, cardio-metabolic (heart disease, hypercholesterolemia, hypertension), diabetes, digestive (irritable bowel syndrome, chronic heartburn or gastro-esophageal reflux disease, ulcer), energy-sleep (chronic fatigue, chronic sleep problems), mental (anxiety, depression, other emotional problems, substance abuse), musculoskeletal (arthritis, chronic back pain, chronic neck pain), respiratory (asthma, allergic rhinitis, chronic bronchitis or emphy-

3 JOEM Volume 50, Number 7, July sema), and other (migraine, other pain conditions, obesity, urinary/ bladder problems) conditions. To focus on conditions that were sufficiently common for interactive effects to be studied with precision, analysis was limited to condition that were reported by at least 5% of respondents. Other Measures Information on a variety of sociodemographic characteristics used as control variables in the analyses was obtained from the HPQ survey and administrative records, including age, sex, education, broad occupational category (executive, administrator, or senior manager; professional; technical support; sales; clerical and administrative support; service; precision production and crafts; operator and laborer), marital status, and number of children. These variables are known to be related to depression in the general population, 11 justifying their inclusion as controls in the current analysis. We also controlled for overall disease burden to make sure that any effects of comorbidity found in the analysis were not merely indirect indicators of the effects of overall disease burden. The information used to create these control variables was obtained from the medicalpharmacy claims data file obtained from the firm. This data file was used to calculate the Charlson Comorbidity Index (CCI) of treated conditions, a widely-used measure of global disease burden, 30 and to create a dichotomous measure of inpatient treatment for any reason in the 6 months prior to the survey. A dichotomy was also created for disability leave at any time in the 6 months before the survey. A continuous measure was created for total medical-pharmacy costs for services provided to each respondent for any reason in the 6 months before the survey. In addition, we constructed variables to describe the disease burden of family members based on the assumption that illness in the household might adversely influence the work performance of the focal respondent. These various measures of individual and family disease burden were used as control variables in the analysis. Analysis Methods As noted above in the subsection on the measurement of health problems, we focused on conditions that were reported by at least 5% of respondents. Linear regression analysis was used to regress the HPQ measure of work performance on a dummy predictor variable for one of these conditions at a time, controlling for measures of sociodemographics and general indicators of overall illness burden, to obtain gross estimates (ie, estimates obtained by considering only one condition at a time) of the associations of each condition with work performance. We focused on gross associations because the high levels of comorbidity found among the conditions raised the possibility that analysis of interactions in models that considered many conditions at once might yield insignificant results because of multicollinearity. Once models of gross effects were estimated, the model for the effect of depression was expanded to evaluate the statistical significance of interactions between depression and each of the other conditions in a series of models that added to the basic depression model the main effect of one other comorbid condition and the interaction between depression TABLE 1 Unweighted and Weighted Sample Characteristics (n 7320) Unweighted Weighted* Est (SE) Est (SE) I. Sociodemographics Age (% in each category) (0.5) 25.8 (0.5) (0.6) 46.2 (0.6) (0.5) 23.4 (0.5) (0.2) 4.6 (0.2) Gender (% female) 23.4 (0.5) 21.1 (0.5) Covered spouse (%) 77.3 (0.5) 74.0 (0.5) Years of eligibility prior to the survey (mean) 3.7 (0.0) 3.7 (0.0) Hourly worker (%) 2.1 (0.2) 2.7 (0.2) Child dependents* Any (%) 66.6 (0.6) 64.2 (0.6) Average number (mean) 1.3 (0.0) 1.2 (0.0) Education (% in each category) Some post-high school grad 11.1 (0.4) 11.2 (0.4) College grad 37.7 (0.6) 38.2 (0.6) Post-college grad 51.2 (0.6) 50.6 (0.6) Occupation (% in each category) Executive, administrator, or senior manager 3.3 (0.2) 3.2 (0.2) Professional 78.2 (0.5) 77.7 (0.5) Sales 15.2 (0.4) 15.5 (0.4) Other 3.4 (0.2) 3.6 (0.2) II. Health Charlson Comorbidity Index (mean) 0.5 (0.0) 0.5 (0.0) Number of medical claims diagnoses (mean) 1.2 (0.0) 1.2 (0.0) Recent hospitalization (%) 6.2 (0.3) 6.2 (0.3) Recent disability leave (%) 2.0 (0.2) 1.8 (0.2) Family health characteristics (%) Family member with recent hospitalization 16.2 (0.4) 14.7 (0.4) Family member with mental health claim 12.1 (0.4) 11.6 (0.4) Family member with CCI greater than (0.5) 20.7 (0.5) *Based on a propensity score weight to adjust for discrepancies between survey respondents and nonrespondents on characteristics recorded in administrative data files. In the 6 months before interview.

4 812 Depression and Work Performance Kessler et al and that other condition. Effects of depression on work performance were examined both in the presence and in the absence of the other condition, whereas effects of the other condition were examined both in the presence and in the absence of depression. A final regression model was then estimated that elaborated the predictors to include information about the separate and joint effects of depression in the presence absence of each logically possible combination of the other conditions that had significant two-way interactions with depression in predicting work performance. All these analyses were carried out with data weighted by propensity score adjustment. Statistical significance was consistently evaluated using 0.05-level two-sided tests. Results Sample Characteristics The modal respondent was a welleducated (88.8% college graduates), salaried (97.3%), middle-aged (46.2% in the age range 35 to 44), married (74.0%), male (78.9%), had one or more child dependents (64.2%), and had been hired at the company for an average of 3.7 years before interview (Table 1). Respondents had a mean of 1.2 treated conditions in the 6 months before the survey and a mean CCI of 0.5. The sample included small proportions of respondents who were either hospitalized (6.2%) or disabled (1.8%) at some time in the 6 months before the survey. Family health characteristics included 11.6% of respondents who had a family member with a mental health claim in the 6 months before interview, 14.7% with a family member who had been an inpatient over the same time period, and 20.7% with a family member having a CCI greater than 1.0. Comparison of unweighted and weighted data show that the unweighted sample of respondents is similar to the weighted sample on most characteristics, with the exception that the unweighted sample overrepresents females, married employees, employees with child dependents, salaried workers, and those that had a recent disability leave or had family members with recent health problems. The Gross Effects of Individual Conditions in Predicting Work Performance Sixteen conditions were sufficiently prevalent in the sample (ie, a prevalence of at least 5%) to be included in the analysis of gross effects, as well as subsequent analysis of interactions between depression (which had a prevalence of 8.3%) and the other conditions (Table 2). Six of these 16 were significantly related to the HPQ measure of work performance in the gross effects models that considered only one condition at a time: chronic fatigue, anxiety, depression, migraine, other headaches, and obesity. Metric regression coefficients are in the range from a high of 3.0 (depression) to a low of 1.6 (obesity). The proportion of the 16 coefficients that is statistically significant (37.5%) is much greater than expected by chance. Furthermore, all significant coefficients are negative, indicating that these conditions are consistently associated with decreases in work performance. Bivariate Interactions of Depression With Other Conditions in Predicting Work Performance The marginal effect of depression in predicting work performance in the absence of a comorbid condition was found to be statistically significant in all 15 equations for the interactions between depression and other conditions, with coefficients in the TABLE 2 Prevalence and Gross Effects of Commonly Occurring Conditions on the HPQ Measure of Work Performance* Prevalence Effect % (SE) (SE) I. Cardio-metabolic Hypercholesterolemia 20.0 (0.5) 0.1 (0.4) Hypertension 10.8 (0.4) 0.4 (0.6) II. Digestive Irritable bowel syndrome 5.0 (0.3) 1.2 (0.8) Chronic heartburn/gerd 8.1 (0.3) 0.1 (0.7) III. Energy-sleep Chronic fatigue 6.4 (0.3) 2.6 (0.8) Chronic sleep problems 8.6 (0.3) 1.1 (0.7) IV. Mental Anxiety 5.6 (0.3) 2.3 (0.8) Depression 8.3 (0.3) 3.0 (0.7) V. Musculoskeletal Arthritis 4.9 (0.3) 0.1 (0.8) Back pain 16.9 (0.4) 0.6 (0.4) Neck pain 14.9 (0.4) 0.1 (0.5) VI. Respiratory Asthma 8.6 (0.3) 0.0 (0.6) Allergic rhinitis 47.0 (0.6) 0.4 (0.3) VII. Other Migraine 7.7 (0.3) 1.7 (0.6) Other headaches 6.7 (0.3) 1.7 (0.7) Obesity 5.9 (0.3) 1.6 (0.7) *Based on 16 separate linear regression equations, in each of which one of the 16 conditions was used to predict job performance with controls for age, age-squared, gender, education, occupation, eligibility of a spouse, number of children, Charlson Comorbidity Index (CCI), inpatient stay in prior 6 months, disability leave in prior 6 months, and family member with CCI greater than 1.0. Significant at the 0.05 level, two-sided test.

5 JOEM Volume 50, Number 7, July TABLE 3 Interactive Effects of Depression With Other Chronic Conditions on the HPQ Measure of Work Performance* The Effect of Depression in the Presence of the Comorbid The Effect of the Comorbid Condition in the Presence of Depression Prevalence Effect of Depression 2 Effect of Comorbid Condition 2 Effect of Interaction 2 Condition % (SE) (SE) (SE) (SE) (SE) (SE) I. Cardio-metabolic Hypercholesterolemia 2.3 (0.2) 3.0 (0.8) 0.2 (0.4) 0.0 (1.7) 3.0 (1.5) 0.2 (1.6) Hypertension 1.3 (0.1) 3.2 (0.8) 0.4 (0.6) 1.0 (1.9) 2.2 (1.8) 1.3 (1.9) II. Digestive Irritable bowel syndrome 0.8 (0.1) 3.2 (0.7) 1.4 (0.9) 2.9 (2.7) 0.3 (2.6) 1.5 (2.5) Heartburn/GERD 1.5 (0.1) 3.3 (0.8) 0.1 (0.7) 1.9 (2.0) 1.4 (1.9) 1.8 (1.9) III. Energy-sleep Chronic fatigue 2.3 (0.2) 2.1 (0.8) 1.2 (0.9) 2.3 (1.9) 4.4 (1.7) 3.6 (1.7) Chronic sleep problems 2.6 (0.2) 2.0 (0.8) 0.4 (0.8) 3.4 (1.7) 5.5 (1.5) 3.1 (1.5) IV. Mental Anxiety 2.7 (0.2) 1.9 (0.9) 0.0 (1.2) 3.2 (1.8) 5.1 (1.6) 3.2 (1.4) V. Musculoskeletal Arthritis 1.1 (0.1) 2.8 (0.8) 0.8 (0.9) 2.1 (2.0) 4.9 (1.9) 1.3 (1.8) Back pain 2.7 (0.2) 3.4 (0.9) 0.6 (0.5) 1.5 (1.5) 1.9 (1.2) 0.9 (1.4) Neck pain 2.5 (0.2) 3.2 (0.8) 0.2 (0.5) 0.5 (1.6) 2.6 (1.4) 0.8 (1.5) VI. Respiratory Asthma 1.4 (0.1) 3.2 (0.8) 0.0 (0.7) 1.5 (2.1) 1.8 (1.9) 1.5 (2.0) Allergic rhinitis 4.8 (0.2) 2.3 (1.1) 0.6 (0.4) 1.3 (1.4) 3.6 (0.9) 0.7 (1.4) VII. Other Migraine 1.3 (0.1) 2.8 (0.8) 1.4 (0.7) 0.6 (2.0) 3.4 (1.8) 2.0 (1.9) Other headaches 1.5 (0.1) 2.7 (0.8) 1.2 (0.7) 0.6 (1.8) 3.4 (1.6) 1.8 (1.7) Obesity 1.3 (0.1) 2.6 (0.8) 0.7 (0.8) 2.3 (2.0) 4.8 (1.9) 3.0 (1.9) *Based on 15 separate linear regression equations, in each of which depression, one other condition, and the interaction between depression and the other condition were used to predict job performance with controls for age, age-squared, gender, education, occupation, number of children, eligibility of a spouse, Charlson Comorbidity Index (CCI), inpatient stay in prior 6 months, disability leave in prior 6 months, and family member with CCI greater than 1.0. The estimated effects of depression and the comorbid condition can be interpreted as the effects of each in the absence of the other, whereas the estimated effects of the interaction can be interpreted as the effects of adding the second condition among respondents with only one of the two. Significant at the 0.05 level, two-sided test. range between 1.9 (in the absence of anxiety) to 3.4 (in the absence of chronic back pain) (Table 3). The marginal effect of the other condition in the absence of depression, on the other hand, was significant at the 0.05 level in only one of these equations (migraine). Therefore migraine is the only one of 15 conditions found to have a unique adverse impact on job performance in the absence of depression. The interaction between depression and the other conditions was significant at the 0.05 level in only one of the 15 equations (chronic sleep problems) and of borderline significance (the 0.10 level) in one other (anxiety), with values between 3.2 and 3.4. We also evaluated the marginal effect of depression in predicting work performance in the presence of a comorbid condition and the marginal effect of the other condition in the presence of depression. These are lowpower tests because they are estimated in small subsamples. The marginal effect of depression was significant in 8 of these 15 tests. As the interaction between depression and the other condition was significant in only one of the 15, though, we cannot reject the hypothesis that the marginal effect of depression in the presence of each of the other 14 conditions is equal to the statistically significant effect of depression in the absence of the condition. The marginal effect of the other condition, in comparison, was significant in only three of the 15 comparisons. These three involved chronic fatigue, chronic sleep problems, and anxiety. Two additional observations are noteworthy about the associations of chronic fatigue, chronic sleep problems, and anxiety with depression. The first of these two can be observed in Table 3: that these three conditions have the strongest interactions with depression in predicting work performance. The second observation concerns comorbidity of these conditions with depression, which cannot be seen in the table. Simple correlations were estimated between depression and each of the other 15 conditions to examine basic patterns of comorbidity. Only three of the 15 conditions had Pearson correlations (r) greater than 0.20 with depression: chronic fatigue (r 0.26), chronic sleep problems (r 0.24), and anxiety (r 0.35).

6 814 Depression and Work Performance Kessler et al Multivariate Interactions in Predicting Work Performance As chronic fatigue and chronic sleep problems are conceptually related, they were combined into a single either/or composite in the analysis of more complex interactions. The latter compared the effects of depression in the presence versus absence of the cross-classification of fatigue/sleep problems and anxiety. This cross-classification yielded four subgroups of workers with depression. The most common were those without either fatigue/sleep problems or anxiety (3.5% of the sample), followed by those with fatigue/sleep problems but not anxiety (2.0%), those with fatigue/sleep problems and anxiety (1.6%), and those with anxiety but not fatigue/sleep problems (1.2%). It should be noted that we did not examine the subgroups of workers with fatigue/sleep problems and/or anxiety in the absence of depression, as our earlier analyses showed that none of these other conditions was significantly related to work performance in the absence of depression. The four depression subgroups were consequently compared with a single control group of all workers without depression irrespective of whether or not they had fatigue/sleep problems and/or anxiety. The results of this analysis showed a clear dose-response relationship between depression and work performance as a function of the other conditions, with the least adverse effect ( 1.5) among workers with pure depression, larger adverse effects among depressed workers with either fatigue/sleep problems in the absence of anxiety ( 2.8) or anxiety in the absence or fatigue/sleep problems ( 3.6), and the largest adverse effects among depressed workers with both fatigue/sleep problems and anxiety ( 6.3). These coefficients are sizeable in substantive terms if we think of them as representing percentage reductions in absolute levels of performance. Based on a 236-day work year (ie, 5 days of work per week over 52 weeks with 24 days of combined paid sickness-vacation leave), a 1.5% reduction in performance is equivalent to roughly 3.5 lost days of work, whereas 2.8% and 3.6% reductions in performance are equivalent to 6.6 and 8.5 lost days of work, and a 6.3% reduction is equivalent to 14.9 lost days of work. This can be contrasted with the 1.4% performance reduction and 3.3 lost days due to migraine, the only other health condition found to have an effect independent of depression. Given the high salary of the workers in the firm studied here (median salary over $130,000), and taking into consideration a fringe benefit rate of over 30% of salary, the yearly human capital value of these performance decrements is in the range between $2550 (for workers with pure depression) and $10,710 (for depressed workers with both fatigue/sleep problems and anxiety), with a total human capital loss because of depression exceeding $8.4 million a year for this employer. Approximately 40% of that total is associated with the 1.6% of workers who have highly comorbid depression that is, with depression in the presence of both fatigue/sleep problems and anxiety, these highly comorbid cases make up only 19% of all workers with depression. Discussion The results reported here are limited by the fact that the assessment of work performance is based on selfreports rather than on objective assessments. This is an inevitable limitation in the analysis of large workforces, though, as few employers maintain administrative databases with assessments of the work performance of all employees. As noted above in the subsection on measures, this limitation is mitigated somewhat by the fact that the HPQ assessment of work performance has been found in methodological studies to correlate significantly with administrative records of work performance in a number of work settings. 22,23 Another limitation is that the relatively low response rate in the HPQ survey might introduce sample selection bias into the results. This limitation is mitigated somewhat by the fact that we had access to some administrative data for the entire workforce that was used to weight the HPQ sample to be representative of the entire workforce on a range of sociodemographic and health indicators. Given the initial aim of investigating subtyping distinctions among workers with depression, an additional limitation is that we focused only on interactions involving comorbid conditions and did not investigate interactions of depression with sociodemographic characteristics. This is an important limitation in light of evidence that the association between depression and work performance might vary as a function of worker age, sex, and occupation. 2,31 Within the context of these limitations, the finding that depression had the single largest adverse effect on work performance of any commonly occurring condition in the firm is consistent with much previous research showing that depression is among the most impairing of all chronic conditions. 4,5,32 The only other condition in the study with an independent effect apart from depression was migraine. The finding that depression was highly comorbid with a number of other conditions is also consistent with previous research. 17,18 The finding that the effect of depression on work performance is modified by the existence of comorbid conditions, finally, is broadly consistent with the results of several previous studies, although the latter studies did not investigate all the comorbid conditions we did here. 19,20 We are aware of only one previous US study that examined the marginal effects of chronic physical disorders in the presence versus absence of depression in predicting work performance. 33 That study used data from a nationally representative household

7 JOEM Volume 50, Number 7, July TABLE 4 Effects of Depression in Conjunction With Anxiety and Fatigue/Sleep Problems on the HPQ Measure of Work Performance* Conditions and Interactions Prevalence % (SE) Effects (SE) Human Capital Loss at the Level ofthe... Individual $ Firm $1,000s % No depression 91.7 (0.3) Depression only 3.5 (0.2) 1.5 (1.0) 2,550 1, Depression and fatigue/sleep problems only 2.0 (0.2) 2.8 (1.6) 4,760 1, Depression and anxiety only 1.2 (0.1) 3.6 (1.5) 6,120 1, Depression, fatigue/sleep problems, and anxiety 1.6 (0.1) 6.3 (1.5) 1,0710 3, Total 8,405 *Based on a single linear regression equations in which dummy predictor variables for subsets of respondents with depression in the absence presence of the other conditions were used to predict job performance with controls for age, age-squared, gender, education, occupation, number of children, eligibility of a spouse, Charlson Comorbidity Index (CCI), inpatient stay in prior 6 months, disability leave in prior 6 months, and family member with CCI greater than 1.0. Significant at the 0.05 level, two-sided test. survey to examine the interactive effects of a broad array of mental disorders in conjunction with five commonly occurring chronic physical conditions (arthritis, asthma, hypertension, ulcers) in predicting days out of role. Consistent with our current results, that earlier study found that depression (and anxiety disorders) had powerful adverse effects in the absence of the physical disorders and even more powerful effects in the presence of these disorders, but that the physical disorders did not have significant effects in the absence of mental disorders. The consistency of this finding across conditions in both the earlier study and the current study suggest that depression, possibly along with other mental disorders, is likely to represent a common pathway that mediates the impact of many physical conditions on work performance. Successful treatment of depression, then, might uniquely be capable of ameliorating adverse workplace effects of a broad range of other conditions that often are related to depression. With regard to the initial aim of investigating subtyping distinctions among workers with depression, we found that comorbid fatigue/sleep problems and anxiety were associated with a substantially increased adverse effect of depression on work performance in the firm studied. Importantly, these comorbid conditions were also highly correlated with depression. Highly comorbid depression, defined as depression in conjunction with both fatigue/sleep problems and anxiety, accounted for 40% of the estimated overall $8.4 million annual lost productivity associated with depression in the firm even though less than 20% of depressed workers has this type of depression. Although the interpretation of human capital cost data of this sort needs to be approached cautiously because of uncertainties regarding measurement and valuation of the lost productivity of individual workers, it is clear from these cost estimates that depression is a condition of importance to this particular firm. Given that the lost work performance is concentrated among workers with high comorbidity, information about this high comorbidity might be used to target a relatively small subset of depressed workers for special outreach and treatment efforts to increase the cost-effectiveness of expanded depression outreach and treatment programs. It is noteworthy that fatigue/sleep problems and anxiety might be considered components of the depression spectrum, in which case the interactions documented here would represent effects of depression severity markers rather than effects of comorbidity. Even if that is the case, though, the fact that these significant interactions exist means that information about the cooccurrence of depression with these comorbidities-markers can be used to target a subset of depressed workers for outreach in an effort to implement interventions that could have an especially high returnon-investment from the employer perspective. Acknowledgments Supported by Eli Lilly and Company. References 1. Greenberg PE, Kessler RC, Nells TL, Finkelstein SN, Berndt ER. Depression in the workplace: an economic perspective. In: Feighner JP, Boyer WF, eds. Selective Serotonin Re-uptake Inhibitors: Advances in Basic Research and Clinical Practice. 2nd ed. New York: John Wiley & Sons; 1996: Kessler RC, Frank RG. The impact of psychiatric disorders on work loss days. Psychol Med. 1997;27: Kessler RC, Greenberg PE, Mickelson KD, Meneades LM, Wang PS. The effects of chronic medical conditions on work loss and work cutback. J Occup Environ Med. 2001;43: Stewart WF, Ricci JA, Chee E, Hahn SR, Morganstein D. Cost of lost productive work time among US workers with depression. JAMA. 2003;289: Wang PS, Beck A, Berglund P, et al. Chronic medical conditions and work performance in the health and work performance questionnaire calibration sur-

8 816 Depression and Work Performance Kessler et al veys. J Occup Environ Med. 2003;45: Berndt ER, Finkelstein SN, Greenberg PE, et al. Workplace performance effects from chronic depression and its treatment. J Health Econ. 1998;17: Kocsis JH, Frances AJ, Voss C, Mason BJ, Mann JJ, Sweeney J. Imipramine and social-vocational adjustment in chronic depression. Am J Psychiatry. 1988;145: Mauskopf JA, Simeon GP, Miles MA, Westlund RE, Davidson JR. Functional status in depressed patients: the relationship to disease severity and disease resolution. J Clin Psychiatry. 1996;57 : Mintz J, Mintz LI, Arruda MJ, Hwang SS. Treatments of depression and the functional capacity to work. Arch Gen Psychiatry. 1992;49: Simon GE, Katon W, Rutter C, et al. Impact of improved depression treatment in primary care on daily functioning and disability. Psychol Med. 1998;28: Kessler RC, Berglund P, Demler O, et al. The epidemiology of major depressive disorder: results from the National Comorbidity Survey Replication (NCS-R). JAMA. 2003;289: Wang PS, Lane M, Olfson M, Pincus HA, Wells KB, Kessler RC. Twelvemonth use of mental health services in the United States: results from the National Comorbidity Survey Replication. Arch Gen Psychiatry. 2005;62: Rost K, Smith JL, Dickinson M. The effect of improving primary care depression management on employee absenteeism and productivity. A randomized trial. Med Care. 2004;42: Simon GE, Ludman EJ, Tutty S, Operskalski B, Von Korff M. Telephone psychotherapy and telephone care management for primary care patients starting antidepressant treatment: a randomized controlled trial. JAMA. 2004;292: Simon GE, VonKorff M, Rutter C, Wagner E. Randomised trial of monitoring, feedback, and management of care by telephone to improve treatment of depression in primary care. BMJ. 2000;320: Wang PS, Patrick A, Avorn J, et al. The costs and benefits of enhanced depression care to employers. Arch Gen Psychiatry. 2006;63: Katon W, Lin EH, Kroenke K. The association of depression and anxiety with medical symptom burden in patients with chronic medical illness. Gen Hosp Psychiatry. 2007;29: Patten SB. Long-term medical conditions and major depression in the Canadian population. Can J Psychiatry. 1999;44: Schmitz N, Wang J, Malla A, Lesage A. Joint effect of depression and chronic conditions on disability: results from a population-based study. Psychosom Med. 2007;69: Stein MB, Cox BJ, Afifi TO, Belik SL, Sareen J. Does co-morbid depressive illness magnify the impact of chronic physical illness? A population-based perspective. Psychol Med. 2006;36: Waghorn G, Chant D, Lloyd C. Labor force activity among Australians with musculoskeletal disorders comorbid with depression and anxiety disorders. J Occup Rehabil. 2006;16: Kessler RC, Ames M, Hymel PA, et al. Using the WHO Health and Work Performance Questionnaire (HPQ) to evaluate the indirect workplace costs of illness. J Occup Environ Med. 2004;46:S23 S Kessler RC, Barber C, Beck A, et al. The World Health Organization Health and Work Performance Questionnaire (HPQ). J Occup Environ Med. 2003;45: Glynn RJ, Schneeweiss S, Sturmer T. Indications for propensity scores and review of their use in pharmacoepidemiology. Basic Clin Pharmacol Toxicol. 2006;98: Schoenborn CA, Adams PF, Schiller JS. Summary health statistics for the U.S. population: National Health Interview Survey, Vital Health Stat : Knight M, Stewart-Brown S, Fletcher L. Estimating health needs: the impact of a checklist of conditions and quality of life measurement on health information derived from community surveys. J Public Health Med. 2001;23: Baker M, Stabile M, Deri C. What do self-reported, objective measures of health measure? J Hum Resour. 2004;39: Edwards WS, Winn DM, Kurlantzick V, et al. Evaluation of National Health Interview Survey Diagnostic Reporting. National Center for Health Statistics. Vital Health Stat ;120: Revicki DA, Rentz AM, Dubois D, et al. Gastroparesis cardinal symptom index (GCSI): development and validation of a patient reported assessment of severity of gastroparesis symptoms. Qual Life Res. 2004;13: Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol. 1992;45: Waghorn G, Chant D. Work performance among Australians with depression and anxiety disorders: a population level second order analysis. J Nerv Ment Dis. 2006;194: Merikangas KR, Ames M, Cui L, et al. The associations of mental and physical conditions with role disability in the US adult household population. Arch Gen Psychiatry. 2007;64: Kessler RC, Ormel J, Demler O, Stang PE. Comorbid mental disorders account for the role impairment of commonly occurring chronic physical disorders: results from the National Comorbidity Survey. J Occup Environ Med. 2003;45: