Insomnia and the risk of depression: a meta-analysis of prospective cohort studies

Background Observational studies suggest that insomnia might be associated with an increased risk of depression with inconsistent results. This study aimed at conducting a meta-analysis of prospective cohort studies to evaluate the association between insomnia and the risk of depression. Methods Relevant cohort studies were comprehensively searched from the PubMed, Embase, Web of Science, and China National Knowledge Infrastructure databases (up to October 2014) and from the reference lists of retrieved articles. A random-effects model was used to calculate the pooled risk estimates and 95 % confidence intervals (CIs). The I 2 statistic was used to assess the heterogeneity and potential sources of heterogeneity were assessed with meta-regression. The potential publication bias was explored by using funnel plots, Egger’s test, and Duval and Tweedie trim-and-fill methods. Results Thirty-four cohort studies involving 172,077 participants were included in this meta-analysis with an average follow-up period of 60.4 months (ranging from 3.5 to 408). Statistical analysis suggested a positive relationship between insomnia and depression, the pooled RR was 2.27 (95 % CI: 1.89–2.71), and a high heterogeneity was observed (I 2 = 92.6 %, P < 0.001). Visual inspection of the funnel plot revealed some asymmetry. The Egger’s test identified evidence of substantial publication bias (P <0.05), but correction for this bias using trim-and-fill method did not alter the combined risk estimates. Conclusions This meta-analysis indicates that insomnia is significantly associated with an increased risk of depression, which has implications for the prevention of depression in non-depressed individuals with insomnia symptoms.


Background
Depression is a common mental disorder and is described as a continuum ranging from a few depressive symptoms to major depression [1]. It is one of the leading global burdens of disease (GBD) and is estimated to be one of the top three health concerns by 2020 [2,3]. Some evidence showed that 5.8 % of men and 9.5 % of women would experience a depressive episode in any given year for a lifetime [4]. Among older adults in Japan, depression is one of the most common diseases and is a leading cause of morbidity and mortality [5,6].
Insomnia is the subjective feeling of having difficulties initiating or maintaining sleep (DIS and DMS respectively, jointly referred to as DIMS) or of non-restorative sleep (NRS) [7,8]. Epidemiological studies have shown that 20 to 35 % of the general population report insomnia symptoms, and that 10 to 20 % have clinically significant insomnia syndrome [9][10][11][12][13]. Insomnia prevalence has been found to be associated with measurements of worse physical and mental health [14].
Both insomnia and depression are major public health problems. It has been reported that insomnia is associated with an increased risk of depression and/or anxiety disorders [9,15]. The identification of modifiable risk factors for depression has a greatly important implication for the primary prevention. Many observational studies have focused on whether insomnia has an influence on depression risk [9,[16][17][18][19][20][21][22][23]. In 2011, Baglioni et al. [24] performed a meta-analysis to investigate the association between insomnia and the risk of depression, and the results showed that an overall odds ratio (OR) for insomnia to predict depression of 2.60 (95 % confidence interval (CI):1.98-3.42). Since then, many new observational studies have emerged, and some of them had large sample sizes and long followup lengths. In addition, the previous review only conducted a subgroup analysis by different age groups of participants. The incidences and the risk factors for depression might vary with the definitions of depression and the exposure changes, or vary in samples from different gender, follow-up durations, and geographic regions. Including more studies and enlarging the sample size would be important for strengthening the reliability of describing the association between insomnia and depression risk. Therefore, we conducted an updated-analysis to further investigate the issue.

Search strategy
This meta-analysis was performed according to the checklist of the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines [25]. The systematic literature search was conducted by two investigators (L.Q.L. and C.M.W.) independently through the PubMed, Embase, Web of Science and China National Knowledge Infrastructure (CNKI) databases for pertinent studies published in English and Chinese from their inception to October 2014. The key words used as the search terms were the following: "insomnia", "sleep disorder", "sleep disturbance", "sleep problem", "sleep quality","sleep duration" in combination with "depression", "mental disorder", and "anxiety". The search was restricted to studies in humans. In addition, the reference lists of all identified relevant publications were reviewed.

Inclusion criteria and exclusion criteria
The eligibility of each study was assessed independently by two investigators (L.Q.L. and C.M.W.), and disagreements were resolved through consultation with the third investigator (Z.X.L.). Studies meeting the following criteria were included in the meta-analysis: (1) the main exposure of interest was insomnia and the outcome of interest was depression; (2) the study design was prospective cohort; (3) insomnia was characterized by DIS and/or DMS or NRS; (4) depression was measured by self-reported symptom scales, physician/clinician diagnosis, or structured clinical diagnostic interview [24]; and (5) the study reported a ratio-based measurement of association of insomnia with depression.
Studies were excluded if: (1) the study was not published as the full reports, such as case reports, commentaries, conference abstracts and letters to editors; (2) the study had a retrospective design; (3) participants with depression at baseline were not excluded for the analysis or the effect of symptoms of insomnia on predicting depression was not controlled for other depressive symptoms at baseline [24]; and (4) Both insomnia and depression acted as the exposure resulting in predicting the other disorder (such as anxiety). If duplicate publications from the same study were identified, we would include the result with the largest number of individuals from the study.

Data extraction
The following information was extracted for each study: name of the first author, publication year, study name, source of the participants, geographic region, gender, mean age of the participants at baseline, insomnia measurement, definition of insomnia based on DSM-IV-TR criteria [8], depression measurement, length of the follow-up period, number of the follow-up assessments, sample size, the OR, relative risk (RR) or hazard ratio (HR) with 95 % CI, and covariates that were adjusted in the multivariable analysis.

Quality assessment
Two investigators (C.M.W. and Y.G.) independently fulfilled the quality assessment using the Newcastle-Ottawa Scale [26], which is a validated scale for non-randomized studies in meta-analysis. The Newcastle-Ottawa Scale is a nine-point scale that allocates points on the basis of the process of selection of the cohort study and measurement of exposure (0-4 points), the comparability of cohorts (0-2 points) and the identification of the outcome and adequacy of follow-up (0-3 points). We assigned scores of 0-3, 4-6, and 7-9 for the low, moderate, and high quality of studies, respectively.

Statistical analysis
The RR was considered as the common measure of the association between insomnia and depression. The HR was considered to be equivalent to RR, and the OR was transformed into the RR. OR was corrected according to the following formula: RR ¼ [27]. In the cohort study, P 0 indicated the incidence of the outcome of interest in the non-exposed group. We preferentially pooled multivariable adjusted risk estimates where such estimates were reported. Where adjusted analysis was not available, we pooled the unadjusted estimates. The RRs for the associations between insomnia and the risks of depression were pooled using the fixed-effects model where heterogeneity was not detected, or the randomeffects model was used otherwise.
For further confirmation and assessment of the association between insomnia and the risk of depression, subgroup analysis was carried out to explore the sources of potential heterogeneity and examine the robustness of the primary results. The differences among subgroups were tested by meta-regression analysis (using STATA 'metareg' command). In sensitivity analysis, we conducted a leaveone-out analysis [28] to observe the magnitude of influence of each study on the pooled RR.
Statistical heterogeneity among studies was evaluated with the Q and I 2 statistics. For the Q statistic, statistical significance was set at P < 0.1 and for the I 2 , the values of 25 %, 50 % and 75 % respectively denoted cut-off points for low, moderate and high degrees of heterogeneity [29]. Potential publication bias was evaluated with a funnel plot and the Egger's test [30]. The Duval and Tweedie nonparametric trim-and-fill methods [31] were performed to further assess the potential publication bias. All statistical analyses were performed with STATA statistical software (version 12.0; College Station, TX, USA). All reported probabilities (P values) were twosided, with a significance level of 0.05 except where otherwise specified. Figure 1 presents the process of this study selection. The search strategy identified 4,802 articles, in which 4,185 articles from the PubMed, 355 articles from the Embase, 226 from the Web of Science, and 36 from the CNKI were retrieved. Of these, based on abstracts or titles, the majorities were excluded after the first screening because they were reviews, case reports, or not relevant to our analysis. After full-text review of the remaining 89 studies, 55 studies were excluded for the reasons shown in Fig. 1. Of note, all of the 21 studies included in the previous review were eligible according to the criteria in our research except two duplicated studies [7,16] used the same samples as the other two studies [20,32], and articles with longest follow-up and more detailed information were retained. Thus, 34 eligible cohort studies were finally included in this meta-analysis.

Quantitative synthesis
The results from the random-effects model combining the RRs for depression in relation to insomnia are shown in Fig. 2. Twenty-six studies suggested a significant positive relationship between insomnia and depression, while the other studies did not. The pooled RR of depression was 2.27 (95 % CI: 1.89-2.71) among populations with insomnia, and a high heterogeneity was observed among studies (I 2 = 92.6 %, P < 0.001).

Subgroup analysis
Subgroup analysis was conducted by mean age at baseline, sex, study location, insomnia definition, type of depression measurement, follow-up duration, sample size, study quality, publication year (before 2010 vs. after 2010), and whether age, socioeconomic status, smoking status, alcohol intake, body mass index (BMI) were controlled or not in models. Insomnia was significantly associated with an increased risk of depression in all subgroups, with the exception of populations from Australia (RR = 1.79, 95 % CI = 0.90-3.54, I 2 = 95.9 %, P < 0.001). However, moderate to high heterogeneities were observed. No interactions  between insomnia and stratification variables in relation to depression risk were observed (all P values for interactions > 0.05; Table 3).

Sensitivity analysis
Sensitivity analyses were used to identify the potential sources of heterogeneity in association between insomnia and the risk of depression. This helped to examine the influence of various exclusions on the combined RR and test the stability of the quantitative synthesis results.
In the leave-one-out analysis by omitting one study in turn, the overall combined RR did not change substantially, with a range from 2.07 (95 % CI:

Publication bias
The visual inspection of the funnel plot identified substantial asymmetry (Fig. 3). The Egger's test identified evidence of substantial publication bias (P < 0.05). A sensitivity analysis using the trim-and-fill method was performed with 16 imputed studies, which produced a symmetrical funnel plot (Fig. 4). Using the trim-and-fill method, the RR was 1.40 (95 % CI, 1.16-1.69; P < 0.001).
Correction for potential publication bias thus did not alter the significant association.

Discussion
The results of this meta-analysis of 34 prospective cohort studies showed that insomnia was significantly associated with an increased risk of depression. The pooled estimates (RR = 2.27; 95 % CI: 1.89-2.71) indicated that participants with insomnia, compared to those free of it, experienced more than two-fold risk to develop depression. Furthermore, the association remained significant in most subgroup analyses.

Comparison with previous study
Our findings were approximately consistent with those from the meta-analysis by Baglioni et al. in 2011 [24], which also showed that sleep difficulty was significantly associated with depression. The results of this current meta-analysis generally concur and further complement the findings of previous review in several important aspects. The present meta-analysis included 15 new prospective cohort studies with larger sample sizes and many more cases, which significantly enhanced the statistical power to detect potential association between insomnia and depression risk. Additionally, the previous review did not investigate any subgroups other than age. More importantly, compared with the previous review, the OR was corrected to more approximately the true RR in the present study, therefore our risk estimate is more accurate and a reliable. Of note, the associations differed among populations of different ethnic backgrounds were investigated in the present study. The current meta-analysis showed that the increased risk was more pronounced for participants from the United States than for European participants. However, no statistically significant association was observed in Abbreviations: BMI body mass index, dur duration criterion, day daytime consequences criterion, F female, M male, NA not available, SES socioeconomic status, Sd sleep difficulties criterion *Study population truly or somewhat representative of a community or population-based study defined as general population, and study population was sampled from a special population (such as population from a company, register patients, data from the health insurance company or health examination organization or pregnant), which defined as non-general populations Australian populations, which might result from the limited number of included studies (two studies comprising 13,323 participants). In order to make the finding generalize to other populations, more studies are warranted to be conducted in other populations from Asia, Africa and South America. There were several possible biological mechanisms through which insomnia in general may increase the risk of depression. Sleep disturbance may play a key role in the development of depression. Experimental studies showed that sleep loss may result in cognitive and affective alterations that lead to depression risk [55]. Alternatively, sleep disturbance impaired emotional regulation and stability [59] and may alter neural processes that may result in the symptomatology of depression [60]. Secondly, sustained arousal and chronic activation of hyperactivity of the hypothalamic-pituitary-adrenal (HPA) axis, the major neuroendocrine mediator of stress response, have been suggested as playing a vital role in the development of depression in insomniacs with objective short sleep duration [14,61]. Finally, other proposed mechanisms by which insomnia might  increase the risk of depression included increasing levels of inflammatory markers, such as C-reactive protein and interleukin-6 (IL-6) [62-64], which indicated low-level systemic inflammation was a predictor of depression development [65].
Long-term, double-blind, randomized controlled trials provided the best evidence on the effect between insomnia and depression. Recently, the study by Gosling et al. [66] showed that an internet-based insomnia intervention would indeed reduce the risk of depression. The role of insomnia treatment in modulating subsequent risk of depression needs to be studied further.

Strengths and limitations
Our review is very valuable and crucial though it is an updated meta-analysis. First, our review added more than 3 times as many participants as the previous review, which provided stronger and more sufficient evidence. Second, the prospective nature of the included studies avoided the influence of recall and selection bias. Third, more studies from additional areas other than the North America and Europe were included, which increased the generalizability. Fourth, we did stratified analyses to explore whether the results were influenced by some confounding factors, and the consistent results from the sensitivity and subgroup analyses indicated that our findings were reliable and robust.
There are also some limitations in this meta-analysis. Firstly, the accuracy of our results might be influenced by the differences of the measurement criteria of insomnia and depression. However, no significant differences among groups were observed for the type of insomnia and depression measurement in this study. Secondly, we were unable to independently summarize the evidence of individual types of insomnia symptoms on depression risk due to no sufficient information in the original studies. Thirdly, although we extracted the most fully adjusted risk estimates, the adjusted confounders varied among the included studies. Some important confounding factors that might influence the relationship between insomnia and depression risk were gender, age, smoking, education, alcohol or drug abuse, other somatic or psychiatric disease, medication status, and social status. These important confounders were not fully adjusted in some of the included studies, which might influence the accuracy of the results. Finally, publication bias were detected, however, we used trim-and-fill method to correct the bias, which did not alter the significant positive association between insomnia and depression risk.
Based on our findings, we suggest that future research in this field is warranted, especially the long-term prospective cohort studies about the association between individual insomnia symptoms and depression. In addition, more interventional studies are necessary to explore the underlying mechanisms that link insomnia and depression.

Conclusions
In conclusion, this meta-analysis supports the hypothesis that insomnia is associated with an increased risk of depression. Considering the increasing prevalence of insomnia worldwide and the heavy burdens of depression, the results of our study provide practical and valuable clues for the prevention of depression and the study of its etiology.