These results have to be interpreted with three limitations in mind. First, while duration of sleep problems was assessed over the past 12 months in order to study the extent to which these problems are persistent over the course of a year, concerns about recall bias in reports about days out of role and role functioning led us to assess impairment over the shorter recall period of the past 30 days. This lack of comparability in time frames is likely to have introduced a conservative bias into the estimated associations of sleep problems with role functioning. Second, both sleep problems and role impairment were assessed with self-report measures, rather than objective assessments, which might be contaminated by comorbid mental disorders. Although efforts were made to adjust for such bias through statistical methods, it would be useful to replicate the analysis in a dataset that included objective measures of sleep problems and role functioning. Third, we cannot exclude the possibility that unmeasured physical disorders account for the associations between sleep problems and role impairment. This third limitation is by far the most important of the three, as we know that sleep problems can be caused by a wide range of physical disorders and that the latter might have effects on role functioning by virtue of other pathways than sleep disruption. An examination of comorbid physical disorders that parallels our analysis of comorbid mental disorders would be needed to investigate this third limitation.
Within the context of these limitations, we conclude that self-reported sleep problems are highly prevalent, that they often persist throughout the year, that they often co-occur with DSM-IV mental disorders, and that they are associated with substantial self-reported role impairment that cannot be explained by comorbid mental disorders. As at least one of the self-reported measures of role impairment, days out of role, is known to be strongly related to objective measures (Kessler et al 2003
; Kessler et al 2004b
), we feel safe in concluding that self-reported sleep problems are significantly associated with true role impairments. Because of our failure to adjust for comorbid physical disorders, though, we cannot conclude that the associations of self-reported sleep problems with role impairment are due to causal effects of sleep problems. Sleep problems might, instead, be risk markers rather than causal risk factors for role impairment (Kraemer et al 1997
The finding that approximately one-third of respondents reported one or more of the four sleep problems assessed in the NCS-R is broadly consistent with the results of other epidemiological surveys (Ancoli-Israel and Roth 1999
; Grandner and Kripke 2004
; National Sleep Foundation 2005
). Our finding that NRS is the most common of the four sleep problems (25%, with the others in the range 16.4-19.9%) presumably reflects the fact that nonrestorative sleep it can occur as a result of DIS, DMS or EMA as well as in the absence of any of these three classic sleep problems. Furthermore, the fact that roughly one-third of people with NRS report neither DIS, DMS, nor EMA implies that nonrestorative sleep is sometimes indicative of poor sleep quality or continuity rather than short sleep duration. This is plausible in light of the fact that several sleep disorders (e.g., sleep apnea) have their primary effects on sleep quality and continuity rather than on sleep duration. At the same time, we found that sleep problems are highly inter-correlated, with roughly two-thirds of respondents who reported any having more than one. Furthermore, about one-third of respondents reported that their sleep problems persisted throughout the entire past 12 months. Chronicity is much more strongly associated with number than type of sleep problems.
The finding of modest socio-demographic correlates of sleep problems in the NCS-R is consistent with previous surveys (Roth and Roehrs 2003
), although our more fine-grained analysis of separate sleep problem that in previous surveys showed an interesting specification involving age. DMS and EMA are higher among respondents in the 45-59 age range than those either younger or older, while DIS and NRS are most common among the young. These results clearly suggest that the focus of previous reports on increasing age as a risk factor for insomnia (Griffiths and Peerson 2005
; Ohayon 2005
) needs to be revised to recognize the heterogeneity of the different sleep problems associated with insomnia. For example, young people are much more likely than elderly people to stay up late, leading to daytime sleepiness that is not associated with difficulties in getting to sleep or staying asleep, even though young people have a low prevalence of difficulty initiating sleep.
The finding that sleep problems are highly comorbid with mental disorders is not surprising, but we failed to replicate the finding of several previous epidemiological studies that major depression is more strongly related to a diagnosis of insomnia than are other mental disorders (Breslau et al 1996
; Ford and Kamerow 1989
; Ohayon and Roth 2003
). It is important to remember, though, that we examined sleep problems, not a DSM diagnosis of insomnia, while previous studies examined the latter. A DSM diagnosis of insomnia requires not only sleep problems but also daytime impairment associated with these problems. It might be, then, that sleep problems are associated with a wide range of mental disorders while the daytime impairment caused by sleep problems is associated more specifically with depression. Although exploration of this possibility goes beyond the bounds of the current report, it should be included in future investigations of sleep disturbance in depression.
The strength of the gross associations between sleep problems and role impairment is striking. This is especially true with regard to number of days out of role. Gross coefficients in predicting this outcome are in the range of 3.2 and 4.0 excess days per month associated with the individual sleep problems. These effects exceed those found in previous studies for most chronic physical and mental disorders in predicting days out of role (Kessler et al 2003
). This is consistent with previous reports of strong gross associations between insomnia and days out of role (Simon and Von Korff 1997
). Even though the size of these coefficients decreases substantially when controls are introduced for comorbid mental disorders, the fact that they remain statistically significant for all sleep problems other than EMA (with coefficients in the range 1.3-2.2 days) argues against the possibility that they are due to comorbid mental disorders. It is possible, though, that unmeasured comorbid physical disorders or systematic response bias explain part of the net associations.
It is important to recognize that the coefficients linking sleep problems to days out of role in the sub-sample of respondents without any DSM mental disorders are insignificant with the exception of the 2.0 excess days out of role per month associated with NRS. This finding is part of a larger pattern in the data that nonrestorative sleep is the sleep problem most consistently related to role impairment after controlling for comorbid mental disorders. This pattern is presumably due to the fact that nonrestorative sleep is the only type of sleep problem considered here that involves wake functioning rather than sleep functioning, implying that the effects of sleep problems on role impairment are strongly mediated by NRS.
It needs to be recalled, in light of the stronger associations of NRS than the other sleep problems with role impairment, that NRS was the only sleep problem assessed in the NCS-R with a dimensional scale. A dichotomization of this scale was created by selecting the cut-off point that maximized explained variance in role impairment. The other three sleep problems were assessed with simple dichotomies. Thus the relationship between NRS and role impairment could be spuriously elevated relative to the relationships involving the other sleep problems by the more precise assessment. As noted earlier in the paper in the section on measures, the NRS scale was dichotomized in order to parallel the dichotomous measures of the other three sleep problems. It might be that the finding that NRS is more strongly related to role impairment than the other sleep problems would be different if the other three problems had been measured dimensionally and dichotomized in a different way. It is relevant in this regard that DIS, DMS and EMA were all defined using conservative criteria. For example, in order to be classified as a DIS sufferer, a patient had to experience “two hours or longer in bed before falling asleep”, compared to the more widely used 30 minute sleep latency criterion used in other epidemiologic studies and clinical trials (Ohayon, 2005
). Future research should consequently examine the sensitivity of results to variation in cut-points on these dimensions.
The results reported here leave a number of issues unresolved that could be addressed in the NCS-R data, but go beyond the boundaries of this first report. We already noted the need to explore the differential associations of depression and other DSM-IV mental disorders with sleep problems versus DSM-IV insomnia. In addition, future research should investigate the joint effects of multivariate sleep problem profiles, the effects of chronic versus intermittent sleep problems, the role of comorbid physical disorders, and the extent to which the associations of sleep problems with role impairment are due to daytime sleepiness. In addition, the recent report of the NIH consensus panel on insomnia noted that more research is needed on the associations of sleep problems with work performance and disability (National Institutes of Health 2005
). All of these issues will be examined in ongoing analyses of the NCS-R data.