Background
Numerous studies have investigated the association between sleep duration and mortality [
1‐
3], but evidence linking insomnia to mortality is less clear. Whereas some studies have reported a significant association between insomnia and mortality [
4‐
8], others have not [
9‐
16]. To further complicate matters, some studies have even found a
protective effect of insomnia on mortality [
17,
18].
The inconsistent results could reflect heterogeneous study designs, inclusion of different covariates and different operationalizations of insomnia. For example, a recent study of older adults in Taiwan found that the association between insomnia and mortality depended on the insomnia definition [
19]. Some studies have investigated insomnia without accounting for sleep duration or use of hypnotic medication, both factors themselves associated with increased mortality, closely linked to insomnia and therefore potential confounders. For example, Finn et al. [
4] found a hazard ratio (HR) of insomnia for mortality of 2.1, but neither sleep duration nor use of hypnotics were taken into account. Similarly, a study by Vgontzas et al. [
7] found a significant effect of insomnia on mortality, but without adjusting for use of hypnotic medication.
Another possible explanation for the inconsistent results is that the association between insomnia and mortality is constrained to certain groups. The Penn State Cohort demonstrated that insomnia was associated with mortality in men sleeping less than 6 hours, whereas no significant effects of insomnia were found for men with longer sleep duration, or for women [
7]. Mallon et al. [
6] also found a gender specific effect, although the only other study controlling for additional sleep covariates did not [
5].
The ambiguous results in the existing literature concerning the insomnia–mortality association therefore require further examination, taking into account possible subgroup differences, important confounders and correlates of insomnia such as hypnotic use and sleep duration. In addition, insomnia should be defined according to formal diagnostic criteria. Based on these considerations, the aims of the current study were: 1) to examine the independent effect of insomnia defined according to the insomnia criteria found in the 5
th edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-V) [
20] on mortality risk after adjusting for several confounders, including sleep covariates, in a community-based sample from Norway (the Hordaland Health Study), and 2) to examine whether the potential effect of insomnia on mortality depended on gender and sleep duration.
Discussion
The aim of this study was to investigate the effect of insomnia on mortality in a middle-age population, after taking into account the effect on mortality of many possible confounders including sleep duration, use of sleep medications and symptoms of OSA. In short, insomnia increased the risk of death 3-fold in the fully adjusted analyses. In exploratory analyses the effect of insomnia was appeared especially strong for men, and among insomniacs who also reported short sleep duration.
The link between sleep and mortality has been extensively studied over the last decades. Whereas the evidence of harmful effects of short and long sleep duration (see [
1‐
3] for review) and use of hypnotics [
36] is ample, the literature on any association between insomnia and mortality has been much more inconsistent. An important issue when studying any consequence of sleep, is to differentiate between distinct types of sleep problems. For example, although there is a significant overlap between those who report insomnia and those with short sleep duration [
37], it is also important to distinguish between these two groups. In addition to the diagnostic criteria of difficulty falling asleep and maintaining asleep, associated with daytime impairment or distress, a diagnosis of insomnia also requires that the sleep disturbance must occur despite adequate opportunity and circumstances for sleep [
38]. As such, voluntary or enforced (e.g. post-partum) sleep loss, characteristic for behaviorally induced insufficient sleep syndrome, should not be misclassified as insomnia, although such patients normally report sound sleep [
39]. The distinction between sleep deprivation and insomnia is important when studying their respective consequences. For example, whereas sleep deprivation consistently has been linked to reduced neuropsychological performance [
40] although with marked intra-individual variability, the effect of insomnia on this outcome is less clear, and at best more subtle and qualitatively different from sleep loss [
41]. Furthermore, patients with advanced or delayed sleep phase syndrome normally complain of early morning awakening and sleep onset difficulties, respectively, and it is not uncommon that studies misattribute such symptoms to insomnia rather than to a circadian rhythm sleep disorder [
42].
We were able to identify 17 studies [
4‐
19,
43] investigating the effect of insomnia on mortality, of which 5 studies found a harmful effect [
4‐
8]. Among these five, only two studies controlled for both sleep duration, use of hypnotic medication and OSA, in addition to other somatic and psychiatric (depression) confounders [
5,
6]. So far no study has investigated the relationship between DSM-V based insomnia criteria and mortality. Our multivariable adjusted findings show that insomnia increased the risk of mortality 3.3 times compared to good sleepers, an effect somewhat larger than found by Hublin et al. [
44]. In agreement with the latter study, gender stratified analyses in the current study suggested that this association was stronger in men: Male insomniacs were 4.7 times more likely to die than non-insomniac males in the crude analyses, while a lower, but still significant effect was observed for women (HR = 2.0). We also found a possible, although not statistically significant, interaction effect of insomnia and short sleep duration on mortality. Participants reporting both insomnia and sleeping less than 6.5 hours had 2.8 times higher risk of dying during follow-up than non-insomniacs. A similar effect of the deleterious combination of short sleep duration and insomnia was also observed by Vgontzas et al.[
7] in a polysomnographic study of insomniacs.
The findings from the current study suggest that the insomnia-mortality association may be confined to subgroups: males and short sleepers. This may be part of the explanation why many previous studies have not found an independent effect of insomnia on mortality. Inconsistent operationalization of insomnia is another possible reason for the mixed findings. In the current study, we used a definition of insomnia, which resembles both The Research Diagnostic Criteria (RDC) for insomnia [
38] as well as the DSM-V criteria [
45]. A similar definition was used in the Hublin et al. study in which a somewhat smaller effect was observed. In contrast, some of the studies failing to find an effect of insomnia have relied on less-than-optimal definitions, often resorting to a single item, to assess insomnia. However, both Lack et al. [
12] and Philips et al. [
14] used well-defined operationalizations of insomnia, but nevertheless found no association between insomnia and deaths. Thus, operationalization issues may not explain all the observed discrepancies. To sum up, although the literature on insomnia predicting death is mixed, the current study clearly shows a strong link between well-characterized insomnia and mortality even after adjusting for confounders.
The big question is then: Why does insomnia predict premature death? Individual covariates in the current study only partly explained the association, and multiple adjustment actually increased the hazard ratio. For example, we found that adjusting for smoking, alcohol use and physical exercise reduced the effect of insomnia on mortality only to some extent (from HR = 2.7 to HR = 2.6). To our surprise, adjusting for depression and somatic diagnoses (hypertension, myocardial infarction, angina, stroke, or diabetes) did not indicate that these comorbid conditions are important pathways. However, the lack of mediation by somatic diagnosis may relate to the study sample being relatively young, and thereby healthier than older adults. With cancer being by far the most common cause in this middle age group many such somatic diagnoses and risk factors would be unlikely to be a mediator. The gender differences in the current study led us to speculate that there might be certain occupational aspects linking insomnia to mortality, but we found no evidence of insomnia being linked to accidents in hazardous occupations [
46], although were limited by the limited number of cause specific outcomes. As such, the question as to why insomnia may increase the risk of death remains open to further examination, especially in terms of exploring more specific the causes of deaths.
There are important methodological considerations to the present study that should be noted. First, the restriction of the sample to participants aged 40-45 limits the generalizability of the findings to other age groups. Second, insomnia was assessed by self-report rather than based on a clinical diagnosis. However, the questionnaire used in our study, the Karolinska Sleep Questionnaire, is relatively analogous to the criteria for insomnia as specified by the DSM-V [
20], and the prevalence found in the present study (5.6%) is identical to a previous report using a stringent DSM-IV definition of insomnia [
47], suggesting a reasonable estimate of insomnia disorder prevalence in the current study. However, we only measured insomnia on one occasion, and as sleep problems commonly fluctuate over time, it would be better to have repeated measurements over a long time period in order to identify chronic insomniacs. An important addition to the DSM-IV and DSM-V from previous versions was the inclusion of a clinical significance criterion to almost half of all the categories (including insomnia), which required that symptoms cause clinically significant distress or impairment in social, occupational, or other important areas of functioning. A similar impact on family, social or occupational life is also required in the latest ICD-10 criteria for insomnia. Still, the definition of insomnia in the current study operationalized the daytime criterion through self-reported tiredness or sleepiness only, and there are other domains of daily functioning that could be affected beyond these items.
There are also other limitations of the present study related to sleep measurement issues: First, the operationalization of OSA was based on a brief self-report questionnaire. Although the prevalence estimate found in the present study (8.4%) is similar to that found in other epidemiological studies in similar age groups [
48‐
52], the use of self-reported symptoms to measure OSA remains problematic because persons are often unaware of their behavior during sleep. However, the Karolinska Sleep Questionnaire is based on sleep problems also reported by the person’s spouse or co-habitant and thus attempts to improve the validity of reported symptoms. Another source of confounding is that both the OSA and insomnia operationalizations included the same item assessing reduced daytime functioning (daytime/sleepiness). Second, the measure of sleep duration did not include reports of wake time after sleep onset (WASO). As such, there is a risk that we may have overestimated sleep duration, especially in persons with maintenance and early morning awakening insomnia. Third, self-reported use of sleep medications was limited to daily or nearly daily basis over the last year. This is a strict definition with individuals using medication 3-4 times per week (and less frequently) not being accounted for.
The list of possible confounding factors most likely did not capture all possible confounding effects from chronic somatic or psychiatric conditions, and there might be other relevant risk factors for mortality not assessed in the current study. Self-report instruments are prone to error and residual confounding cannot therefore be ruled out. Screening for psychiatric morbidity was limited to symptoms of anxiety and depression. In addition, some of the confounders were operationalized in a rather crude way. For example, the employed cut-off for alcohol consumption was based on weekly instead of daily intake. This puts some limitation on the adjustment for confounders concerning the association between insomnia and mortality. Also, the number of deaths was limited to 160, which prevented us from conducting further sub-analyses, for example on the interaction between insomnia and OSA and use of hypnotic medication, and also precluded us from investigation of the relationship between insomnia and specific causes of death. The small number of deaths in the insomnia group also represents a problem with regards to statistical power. A related issue concerns the long list of adjustment variables that were used in the regression analyses. Given the modest sample size, it is not unproblematic to include so many confounders, which may distort some of the statistical findings. For example, we cannot easily explain why the fully adjusted HRs in several instances were larger than the unadjusted. For this reason, we did not conduct fully adjusted analyses when examining subgroups (interaction with gender and sleep duration). Also, as the current study included measures of both insomnia, sleep medications, sleep duration and OSA, multicollinearity could have produced biased results, but testing for this in the current study showed this not to be the case (with tolerance values larger than 0.61 and VIFs less than 1.63 for all predictor variables) [
53]. This issue was also pointed to by Hublin et al. [
44] emphasizing how difficult it is to fully separate risks of insomnia from those of similar conditions and disorders. Therefore, the findings, and especially those from the fully adjusted analyses, should be interpreted with caution and repeated by studies based on larger samples. Also, given the age in our study population, several risk factors that were not present at baseline (obesity, sleep apnea, diabetes, etc.) may have emerged later during the course of follow-up, which may have influenced the outcome. Finally, the participation rate was not very high (63%) and generalizations of the results must be made with caution as both mortality rates [
54] and symptom levels [
55] are higher among non-participants.
There are several strengths of this study that deserves mention. First, the design included complete follow-up through the nationwide Cause of Death Registry, made possible through linkage using the personal identification number unique to each Norwegian resident. Also, the comprehensive list of confounders, including other sleep variables, allowed investigation of the unique effect of insomnia. In addition, the study had sufficient statistical power to conduct sub-analyses of the interaction effects of insomnia and gender and sleep duration.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
Authors BS and SO were responsible for conception and design of the study, and BS conducted the statistical analysis and drafted the manuscript. Author GT were involved in acquisition of data and obtained funding for the HUSK study. Authors SP, NG, BB, PS, GT, SO and RU gave critical revision of the manuscript for important intellectual content. Authors BS and SO had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. All authors read and approved the final manuscript.