Gupta, Shaloo MS; Goren, Amir PhD; Kim, Edward MD; Gabriel, Susan MSc; Dupclay, Leon PharmD
Major depression affects nearly 1 in 10 US adults1 and is characterized by mood (eg, irritability, sadness, thoughts of suicide, feelings of worthlessness), cognitive (eg, inability to concentrate, difficulty making decisions), and physical (eg, fatigue, changes in appetite, changes in sleep) symptoms that last for at least 2 weeks, impair functioning2, and lead to an estimated $44 billion per year in lost productive time among US workers.3 Greenberg et al4 estimated the total US economic burden of depression in 2000 at $83 billion, with overall loss of workplace productivity due to depression accounting for 62% of the cost. The recommended goal of depression treatment is remission, defined by the absence of depressive symptoms5; however, approximately 50% to 70% of treated patients do not achieve conventional remission criteria.6 Several studies found that inadequate response to treatment is associated with high direct and indirect economic costs.7,8
Residual symptoms of depression are common among patients treated for depression; some may be independent comorbidities and others may be a medication side effect.9–11 Sleep disturbances are among the physical symptoms associated with depression and are highly prevalent among depressed patients, with up to 90% of patients complaining about sleep quality.12,13 Moreover, persistent insomnia is one of the more common residual symptoms of depression with incomplete treatment response.10,14,15 Sleep disturbances in depression may include difficulty falling asleep (initial insomnia), difficulty maintaining sleep or poor sleep quality (middle insomnia), waking up too early (terminal insomnia), prolonged sleep episodes at night or increased daytime sleep (hypersomnia), and circadian reversal.13,16,17 In patients with partial response to antidepressant treatment, 48% reported experiencing early insomnia, 53% middle insomnia, and 16% late insomnia.18 It is evident that sleep disturbances play an important role in depression and its treatment.
Although sleep disturbances are often comorbid with depression13,14 and are considered residual symptoms of depression,10 the relationship between depression treatment response and sleep disturbances has been investigated less frequently with respect to the degree of treatment response—that is, partial, not just complete, or nonresponse—and those studies have typically focused on nonrepresentative, clinical samples,18,19 using the Hamilton Rating Scale for Depression20 to evaluate treatment response.
Most notably, research is lacking with respect to the impact of inadequate response on health outcomes such as work productivity impairment and health care resource use, which can help inform payers and employers regarding the consequences of inadequate treatment. Remission from major depression has been associated with work impairment,21 but there is limited evidence regarding the link between antidepressants and work-related outcomes.22 This is a noteworthy gap in knowledge about depression treatment because current treatments resulting in partial response may help alleviate certain aspects of depression whereas specific symptoms such as sleep disturbances remain. The current study is intended to help address this gap, as well as to improve our understanding of the relationship among all these variables: depression, treatment response, sleep disturbance, and health outcomes. This analysis enables the specification of burden or costs associated with different degrees of failure to treat depression—not just cognitive or mood, but also sleep symptoms—more completely.
A previous study by Knoth et al7 provided initial evidence for the impact of partial response to depression treatment on health outcomes, showing that partial response and no response to antidepressants leads to greater resource use and lost work productivity. However, that study did not investigate sleep disturbances. Moreover, it did not differentiate between respondents whose treatment for depression had started recently (and most likely did not have a chance to take effect) and those whose treatment was well underway. Therefore, the burden of partial response to treatment may have been overestimated, given the number of newly treated patients.
The current study is intended to replicate the study of Knoth et al7 and to add an investigation into sleep disturbances. The current study provides a more conservative test of treatment response, by focusing on respondents who have been on their current treatment for at least 90 days. Also, it improves upon the previous study by providing more precise, sensitive measurement of the outcomes of interest, using nonlinear modeling for measures with nonlinear distributions (eg, productivity measures) and continuous rather than dichotomized measures of resource use (to provide levels rather than just likelihood of resource use).
One objective of this study was to assess the relationship between sleep disturbance and inadequate response to depression treatment, using data from a representative, self-report survey of adults in the United States, the National Health and Wellness Survey (NHWS), which is comparable to other governmental sources such as the National Health Interview Survey, the National Health and Nutrition Examination Survey, and the Medical Expenditure Panel Survey.23 A second objective was to measure the association between inadequate treatment response and work productivity loss and health care resource use, using patient-reported outcomes. Finally, an exploratory third objective was to investigate whether sleep disturbances might mediate the relationship between treatment response and provider visits, or alternatively, whether inadequate response to treatment mediates the relationship between sleep disturbances and provider visits. This last objective begins to shed light on the possible causal connection between treatment response and sleep disturbances, which is currently not clearly indicated in the literature.
Data were analyzed from the 2010 NHWS (n = 75,000). NHWS is an annual, cross-sectional, self-administered Internet survey given to a sample of adults (18 years and older) to gather information about demographics, disease status, and health outcomes. Survey respondents were identified through a Web-based opt-in consumer panel. Panel members were recruited via opt-in e-mails, e-newsletters, online banners, and panel partner coregistration. Panelists explicitly volunteered for panel membership and were registered through unique e-mail addresses, and each completed a demographic profile. Stratified random sampling was implemented within the panel so that the final NHWS sample mimicked the demographic composition of the adult US population. The study was reviewed and approved by Essex IRB (Lebanon, NJ).
The study sample comprised 2010 US NHWS patients who reported having been diagnosed with depression by a health professional and having taken a depression medication for at least 90 days (based on patients' taking of any single antidepressant for at least 3 months and having used an antidepressant for at least 30 days in the past month). Patients were not excluded on the basis of diagnosis with other comorbidities, which may have included other psychiatric disorders.
The study sample was grouped according to self-reported treatment response status, defined on the basis of the response to a single item from the Short Form 12-Item Health Survey (SF-12v2): “How often in the past 4 weeks have you felt downhearted and depressed?” The item has a five-point Likert response scale (“all of the time” to “none of the time”). Patients who answered “all of the time” or “most of the time” were considered to be nonresponders. Patients who responded “some of the time” were considered partial responders, whereas those who answered “a little of the time” or “none of the time” were complete responders. This grouping follows the same approach used in a previous study with the corresponding SF-8 item: “During the past 4 weeks, how much have you been bothered by emotional problems (such as feeling anxious, depressed, or irritable)?”7 These items were chosen for their face validity as indicators of depression and its frequency, even in the face of concurrent treatment.
Patient demographics included age as a continuous variable, sex (female vs male), race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, other), education (some college education or more vs less than a college education), income (annual household income <$25,000, $25,000 to less than $50,000, $50,000 to less than $75,000, $75,000 or more, declined to answer), employment status (employed full-time, part-time, or self-employed vs not employed) and health insurance information (commercial: employer, spouse, or self-purchased; Medicaid/Medicare; VA/Tricare; uninsured; or not sure).
Patients' health history characteristics included body mass index (BMI) (underweight, normal, overweight, obese, or declined to answer), smoking behavior (currently a smoker vs nonsmoker), alcohol use (currently a drinker vs nondrinker), exercise behavior (currently exercise vs not), and the adjusted Charlson comorbidity index (CCI). The adjusted CCI24 mimics the CCI by summing the presence of the following conditions (weighted by each condition's contribution to risk of mortality): HIV/AIDS, metastatic tumor, lymphoma, leukemia, any tumor, moderate/severe renal disease, hemiplegia, diabetes, mild liver disease, ulcer disease, connective tissue disease, chronic pulmonary disease, dementia, cerebrovascular disease, peripheral vascular disease, myocardial infarction, and congestive heart failure. The adjusted CCI does not include diabetes with end organ damage and moderate/severe liver disease, both of which appear in the original measure. Higher CCI index scores indicate greater comorbid burden on the patient and prospective likelihood of mortality.
Sleep disturbances were indicated by the presence of any of the range of possible sleep symptoms, and they were defined specifically as patients' reporting of at least one of the following: insomnia, sleep difficulties, difficulty falling asleep, difficulty staying asleep, waking up too early, and difficulty staying awake. In addition, the following items were examined individually: difficulty falling asleep, difficulty staying asleep, and waking up too early.
Health care resource utilization in the preceding 6 months was assessed by number of health care provider visits, the number of emergency department visits, and the number of times patients were hospitalized. Patients were asked the following questions: “How many visits did you make to the following traditional health care provider(s) in the past 6 months? (eg, general practitioner, internist)”, “How many times have you been to the emergency department for your own medical condition in the past 6 months?” and “What is the total number of days you were hospitalized for your own medical condition in the past 6 months?”
Work productivity was assessed using the Work Productivity and Activity Impairment (WPAI) questionnaire.25 The WPAI scale is a validated scale used to measure lost work productivity and impairment in daily activities. Four subscales (absenteeism, presenteeism, overall work impairment, and activity impairment) are generated in the form of percentages, with higher values indicating greater impairment. Absenteeism represents the percentage of work time missed due to health in the past 7 days. Presenteeism represents the percentage of impairment while at work due to health in the past 7 days. Overall work impairment represents the total percentage of work time missed due to either absenteeism or presenteeism (since those measures are mutually exclusive). Activity impairment represents the percentage of impairment during daily activities. Only employed respondents provided data for absenteeism, presenteeism, and overall work impairment, whereas all respondents provided data for activity impairment.
Bivariate analyses were used to compare demographics, patient characteristics and health outcomes across the no response, partial response, and complete response depression treatment groups. Chi-squared tests were used for categorical variables, and analyses of variance for continuous variables.
Multivariable regression models adjusted for the demographics and health history variables specified earlier, including the following as covariates: age, sex, race/ethnicity, education, income, employment, health insurance, BMI, smoking behavior, alcohol use, exercise behavior and the adjusted CCI. These control variables were included, in part, on the basis of whether they were found to contribute to significant differences between the treatment response groups at the bivariate level (Table 1), in which case they proved to be potential confounders for regression analyses, especially given the likelihood of correlations between the covariates and the outcomes of interest. For example, lower BMI, reduction in smoking, and daily exercise such as walking have been found to reduce health care costs.26 Obesity can have a significant impact on work productivity and daily activities,27 and alcohol dependency can cause sleep problems.28 The literature also suggests that CCI predicts poor long-term health outcomes.24 Daily smoking,29 obesity,30 regular exercise,31 and alcohol consumption patterns32 have been found to predict depression or depression symptoms.
The independent association of treatment response (no response, partial response, and complete response) with sleep disturbances, work productivity, and health care resource utilization was assessed using regression models. A series of logistic regressions was used to predict from treatment response groups the presence versus absence of sleep disturbances, difficulty falling asleep, difficulty staying asleep, and difficulty waking up. Odds ratios were calculated for all logistic regressions. Generalized linear models were fitted to predict work productivity, activity impairment, and resource use from treatment response. As work productivity impairment and resource utilization are often highly skewed, the generalized linear models specified a negative binomial distribution, testing whether adjusted log counts (controlling for covariates) differed across groups. The antilogs of the regression estimates were calculated to yield rate ratio values. These indicate the number of times impairment was greater for a given group than the reference group. For all regression models, statistical significance was set at P < 0.05.
Mediation analyses were conducted to assess the possible order in which sleep disturbances and inadequate response to depression treatment may have caused health care resource utilization. Four mediation models were tested, in which one variable (the mediator) was hypothesized to explain the causal connection between another variable (the predictor) and an outcome: (a) sleep disturbances as a mediator between inadequate response to depression treatment and resource use and (b) inadequate response to depression treatment as a mediator between sleep disturbances and resource use. Inadequate response to treatment was operationalized as no response versus complete response and no response versus partial response to treatment. Resource use was operationalized as number of health care provider visits. The Sobel test was used to calculate the total and specific indirect effects, whereas bootstraps (×1000) were used for resampling and estimating 95% confidence intervals for indirect effects. Test for significance of the mediation followed the methods of Preacher and Hayes33 for estimating indirect effects. Ordinary least squares multiple regressions were used to estimate direct paths between predictors, mediators, and outcomes. Unstandardized path coefficients were reported.
A total of 6116 patients in the 2010 US NHWS database reported both a diagnosis of depression and taking antidepressant medication for at least 90 days. Among these patients, 2179 (35.6%) were classified as exhibiting “no response,” 2129 (34.8%) as “partial response,” and 1808 (29.6%) as “complete response.” The average age of respondents among the depression response groups ranged from 47.8 to 51.2 years. A higher proportion of nonresponders were male, not college educated, not currently employed, currently smoking, or obese, compared with partial responders and complete responders (Table 1).
Prevalence of Sleep Disturbances Among Depression Treatment Response Groups
Unadjusted bivariate results showed that 90.0% of nonresponders, 83.9% of partial responders, and 78.5% of complete responders reported sleep disturbances. Furthermore, nonresponders and partial responders reported experiencing a greater number of sleep disturbance symptoms (Table 2).
Logistic regression models revealed that nonresponders had more than twice the odds of reporting at least one sleep disturbance compared with complete responders (P < 0.001). Partial responders had 44% greater odds of reporting at least one sleep disturbance than complete responders (P < 0.001). Nonresponders had higher odds of reporting difficulty with falling asleep (94%), difficulty staying asleep (75%), and difficulty waking up (45%) than complete responders (P < 0.001 for all). Partial responders had higher odds of reporting difficulty with falling asleep (28%) and difficulty staying asleep (20%) than complete responders (P < 0.01 for both) (Table 3).
Effect of Inadequate Response to Treatment on Work Productivity
A smaller proportion of patients was currently employed with decreasing levels of treatment response; 44.4% of complete responders were employed, versus 43.1% of partial responders and 36.7% of the nonresponders (P < 0.001) (Table 1). Among patients who were employed, bivariate analyses revealed that nonresponders and partial responders reported missing more work (absenteeism) and experiencing greater impairment while working (presenteeism) than complete responders. The overall work impairment of nonresponders (49.2%) was more than twice the level of impairment of complete responders (23.8%). Among all depressed patients, nonresponders (62.7%) and partial responders (46.4%) also reported higher levels of activity impairment than complete responders (37.2%) (Table 4).
Adjusting for covariates, regression analyses showed that among employed patients, nonresponders reported 2.32 times as much absenteeism, 2.00 times as much presenteeism, and 1.95 times as much overall work impairment (all P < 0.001) as did complete responders. Partial responders reported 1.33 times as much presenteeism and 1.31 times as much overall work impairment (both P < 0.001) as did complete responders. The rate of absenteeism did not differ significantly between partial and complete responders. Among all depressed patients, nonresponse versus complete-response patients reported 1.64 times more activity impairment (P < 0.001). Patients exhibiting partial versus complete response reported 1.25 times greater activity impairment (P < 0.001) (Table 5).
Effect of Inadequate Response to Treatment on Health Care Resource Use
Unadjusted bivariate analyses showed that nonresponders reported an average of 10.3 visits to a health care provider in the past 6 months, which was higher than their partial responder (8.1 visits) and complete responder counterparts (7.6 visits). In addition, nonresponders and partial responders more often utilized the emergency department than complete responders (Table 4).
Adjusting for covariates, generalized linear regression analyses showed that nonresponders experienced 1.30 times as many health care provider visits (P < 0.001), 1.58 times as many emergency department visits (P < 0.001), and 1.42 times as many hospitalizations (P < 0.001) as did complete responders (Table 6). Partial responders experienced 1.20 times as many emergency department visits (P = 0.021) and 1.27 times as many hospitalizations (P = 0.018) as did complete responders. Partial responders and complete responders did not differ significantly in their number of health care provider visits.
Sleep Disturbances as Mediator Between Inadequate Response to Depression Treatment and Resource Use
A post hoc, exploratory analysis to test whether sleep disturbances mediated the relationship between inadequate response to depression treatment and number of provider visits (ie, resource use) was conducted using a Sobel test and bootstrapping. Notwithstanding the significant direct effect of no response versus complete treatment response on number of provider visits (β = 2.14, P <0.001), there was also a significant indirect effect of inadequate response on provider visits via sleep disturbances (β = 0.23; 95% CI, 0.14 to 0.33) (Fig. 1). This was a significant reduction from the total effect (β = 2.37, P <0.001; R2 = 0.071), suggesting partial mediation that may be due to inadequate response causing sleep disturbances, which in turn lead to more health care provider visits. Partial response versus complete response also had a significant, albeit weaker, indirect effect on the number of provider visits via sleep disturbances (β = 0.10; 95% CI, 0.05 to 0.16).
Inadequate Response to Depression Treatment as Mediator Between Sleep Disturbances and Resource Use
Whereas sleep disturbances partially mediated the relationship between inadequate response to depression treatment and number of provider visits (ie, resource use), it is possible that inadequate response mediates the relationship between sleep disturbances and provider visits. To test this possibility, a second set of models was conducted with inadequate response as the mediator. Again, though there was a significant direct effect of sleep disturbances on provider visits (β = 2.03, P <0.001), there was also a significant indirect effect of sleep disturbances on provider visits via no response versus complete treatment response (β = 0.43; 95% CI, 0.29 to 0.60) (Fig. 2). This was a significant reduction from the total effect (β = 2.46, P < 0.001; R2 = 0.071), suggesting partial mediation that may be due to sleep disturbances causing inadequate response, which in turn leads to more provider visits. The indirect effect of sleep disturbance on provider visits via partial response versus complete response (β = 0.03; 95% CI, −0.02 to 0.09) was not significant.
Although inconclusive, the stronger indirect effect in this model suggests inadequate treatment response (namely, no response vs complete response) rather than sleep disturbances as the most likely mediator of provider visits.
In this cross-sectional study, depressed patients with at least 90 days of antidepressant therapy reported high rates of sleep disturbance symptoms, consistent with previous findings.13,16,17 Patients with only partial or no response to antidepressant treatment exhibited a higher degree of sleep disturbance than complete responders. Residual depressive symptoms, including sleep disturbances, have been found to predict relapse.34 As some antidepressant therapies may impair sleep quality,35 it is unclear whether residual sleep disturbance is a side effect of treatment or a symptom of unresolved depressive illness. Therefore, clinicians may be faced with a challenge when evaluating treatment options. Treatments offering a more comprehensive management of depression are needed to deal effectively with the high prevalence of sleep disturbance symptoms. Physicians should anticipate the potential for sleep disturbances in depressed patients and suggest appropriate treatments to prevent likelihood of occurrence rather than responding to sleep problems on an ad hoc basis.
In addition, the current study demonstrates the potentially profound impact of inadequate response to treatment on work productivity and health care resource use. Consistent with previous research,7 partial response and nonresponse to treatment (vs complete response) were associated with greater likelihood of emergency department utilization and hospitalization (indicating considerably higher direct costs), as well as greater work productivity loss. In one claims database study, patients with inadequate response to at least two antidepressant regimens had approximately 40% greater direct health care costs than those who responded to earlier lines of therapy.8 Depressive patients with inadequate response to treatment experience a high degree of sleep disturbances. Depression with sleep disturbance is also associated with higher resource use. The association of inadequate treatment response with a variety of outcomes (e.g. productivity loss, resource utilization, and sleep disturbances) suggests the potential value of a simple (eg, single-item) measure by which patients can rate their response to treatment in clinical practice. Also, this increased resource use is costly to patients, payers, and employers and is a substantial burden to the health care system. Appropriate surveillance and management may lead to improved functional and economic outcomes by achieving more complete response to antidepressants.
Although results from the mediation models suggest that inadequate response may either drive sleep disturbances or vice versa, they are slightly more supportive of a model in which sleep disturbances predict negative response to treatment (specifically, no response vs complete response), which in turn leads to greater resource use. To explore the fullest possible range of effect of treatment response, further modeling and longitudinal studies are necessary to test this hypothesis more fully and to provide more precise tests of the potential mediation effects, accounting for the appropriate distributions and including all desired variables within a single model. Also, these results suggest that there is a need for additional studies that confirm the possible mediation of inadequate treatment response using various patient samples that are self-reported and clinically driven, and can therefore more clearly determine what proportion of the higher resource utilization and loss of work productivity are attributable specifically to inadequate treatment response or sleep disturbances. This additional research can help guide physicians in administering the most appropriate treatments with the best timingaccounting for whether reduction in sleep disturbances ought to be a primary initial goal or one of the main goals of antidepressant treatment and efforts to bring about successful remission. Additional research is needed to investigate the incremental cascading effects of inadequate treatment response when complete treatment response does not occur.
Several limitations of this study should be addressed. This study was a cross-sectional analysis based on a self-reported patient survey, limiting the ability to assess causal relationships. Although the models controlled for a variety of key variables such as patient demographic characteristics and comorbidities, it is possible that unaccounted-for variables may explain the relationships among the observed variables. As the NHWS is a self-reported online survey, data are susceptible to recall bias and selection bias. The WPAI metrics used to assess work productivity were recalled in the past 7 days and health care resource use was recalled for the past 6 months. A single item from the SF-12 was used to measure and categorize treatment response; given that this item has not been validated extensively with respect to this particular use, further research is needed to estimate its reliability and validity vis-a-vis treatment response and existing validated measures of depression. All data were self-reported; variables such as diagnoses, treatments, and sleep symptoms were not confirmed by clinicians, patient records, or administrative claims data, and are not validated measures. In addition, data to assess the severity of depression were not available, and differing severity would likely affect outcomes such as work productivity and health care resource use.
The mediation model output should be interpreted with caution, as (a) the models assessed only two of the three response categories (no response and complete response) and (b) due to the dichotomous nature of the mediators and the negative binomial distribution of the outcome variable, estimates are imprecise (the models assume that mediator and outcomes variables are continuous).
Patients with depression and inadequate response to treatment experience a high degree of sleep disturbance as well as higher health care resource use and work productivity loss. Appropriate monitoring of treatment response is important to optimize clinical and economic outcomes. Sleep disturbances continuing despite treatment may be a strong indicator of an incomplete treatment response.
The authors thank Lu Lu Kuang, PhD, for the background research and editorial assistance. The authors also thank Dr Safiya Abouzaid who provided comments about the various drafts of this paper.
1. Centers for Disease Control and Prevention. Current depression among adults–United States, 2006 and 2008. MMWR Morb Mortal Wkly Rep. 2010;59:1229–1235.
2. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders. 4th ed. Text revisions. Washington, DC: American Psychiatric Association; 2000.
3. Stewart WF, Ricci JA, Chee E, Hahn SR, Morganstein D. Cost of lost productive work time among US workers with depression. JAMA. 2003;289:3135–3144.
4. Greenberg PE, Kessler RC, Birnbaum HG, et al. The economic burden of depression in the United States: how did it change between 1990 and 2000? J Clin Psychiatry. 2003;64:1465–1475.
5. Koenig AM, Thase ME. First-line pharmacotherapies for depression - what is the best choice? Pol Arch Med Wewn. 2009;119:478–486.
6. Nierenberg AA, DeCecco LM. Definition of antidepressant treatment response, remission, nonresponse, partial response, and other relevant outcomes: a focus on treatment-resistant depression. J Clin Psychiatry. 2001;62:5–9.
7. Knoth RL, Bolge SC, Kim E, Tran QV. Effect of inadequate response to treatment in patients with depression. Am J Manag Care. 2010;16:e188–e196.
8. Gibson TB, Jing Y, Smith Carls G, et al. Cost burden of treatment resistance in patients with depression. Am J Manag Care. 2010;16:370–377.
9. Fava GA, Grandi S, Zielezny M, Canestrari R, Morphy MA. Cognitive behavioral treatment of residual symptoms in primary major depressive disorder. Am J Psychiatry. 1994;151:1295–1299.
10. Nierenberg AA, Keefe BR, Leslie VC, et al. Residual symptoms in depressed patients who respond acutely to fluoxetine. J Clin Psychiatry. 1999;60:221–225.
11. Menza M, Marin H, Opper RS. Residual symptoms in depression: can treatment be symptom-specific? J Clin Psychiatry. 2003;64:515–523.
12. Armitage R. The effects of antidepressants on sleep in patients with depression. Can J Psychiatry. 2000;45:803–809.
13. Tsuno N, Besset A, Ritchie K. Sleep and depression. J Clin Psychiatry. 2005;66:1254–1269.
14. Fava M. Daytime sleepiness and insomnia as correlates of depression. J Clin Psychiatry. 2004;65:27–32.
15. Nelson JC, Portera L, Leon AC. Residual symptoms in depressed patients after treatment with fluoxetine or reboxetine. J Clin Psychiatry. 2005;66:1409–1414.
16. Thase ME. Depression, sleep, and antidepressants. J Clin Psychiarty. 1998;59:55–65.
17. Benca RM, Obermeyer WH, Thisted RA, Gillin JC. Sleep and psychiatric disorders. A meta-analysis. Arch Gen Psychiatry. 1992;49:651–668.
18. Paykel ES. Remission and residual symptomatology in major depression. Psychopathology. 1998;31:5–14.
19. Fava M, Thase ME, DeBattista C. A multicenter, placebo-controlled study of modafinil augmentation in partial responders to selective serotonin reuptake inhibitors with persistent fatigue and sleepiness. J Clin Psychiatry. 2005;66:85–93.
20. Hamilton M. Rating depressive patients. J Clin Psychiatry. 1980;41:21–24.
21. Mintz J, Mintz LI, Arruda MJ, Hwang SS. Treatments of depression and the functional capacity to work. Arch Gen Psychiatry. 1992:49:761–768.
22. Greener MJ, Guest JF. Do antidepressants reduce the burden imposed by depression on employers? CNS Drugs. 2005;19:253–264.
23. DiBonaventura M, Wagner JS, Yuan Y, L'Italien G, Langley P, Kim WR. Humanistic and economic impacts of hepatitis C infection in the United States. J Med Econ. 2010;13:709–718.
24. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40:373–383.
25. Reilly MC, Zbrozek AS, Dukes EM. The validity and reproducibility of a work productivity and activity impairment instrument. Pharmacoeconomics.1993;4:353–365.
26. Leigh JP, Hubert HB, Romano PS. Lifestyle risk factors predict healthcare costs in an aging cohort. Am J Prev Med. 2005;29:379–387.
27. Rodbard HW, Fox KM, Grandy S; Shield Study Group. Impact of obesity on work productivity and role disability in individuals with and at risk for diabetes mellitus. Am J Health Promot. 2009;23:353–360.
28. Brower KJ. Alcohol's effects on sleep in alcoholics. Alcohol Res Health. 2001;25:110–125.
29. Colman I, Naicker K, Zeng Y, Ataullahjan A, Senthilselvan A, Patten SB. Predictors of long-term prognosis of depression. CMAJ. 2011.
30. Onyike CU, Crum RM, Lee HB, Lyketsos CG, Eaton WW. Is obesity associated with major depression? Results from the Third National Health and Nutrition Examination Survey. Am J Epidemiol. 2003;158:1139–1147.
31. De Moor MH, Boomsma DI, Stubbe JH, Willemsen G, de Geus EJ. Testing causality in the association between regular exercise and symptoms of anxiety and depression. Arch Gen Psychiatry. 2008;65:897–905.
32. Paljärvi T, Koskenvuo M, Poikolainen K, Kauhanen J, Sillanmäki L, Mäkelä P. Binge drinking and depressive symptoms: a 5-year population-based cohort study. Addiction. 2009;104:1168–1178.
33. Preacher KJ, Hayes AF. Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behav Res Methods. 2008;40:879–891.
34. Fava GA, Fabrri S, Sonino N. Residual symptoms in depression: an emerging therapeutic target. Prog Neuropsychopharmacol Bio Psychiatry. 2002;26:1019–1027.
35. Roth T. The relationship between psychiatric diseases and insomnia. Int J Clin Pract Suppl. 2001;3–8.