Association between Depression and Insomnia Subtypes: A Longitudinal Study on the Elderly in Japan

Abstract

Study Objective:

To examine the association between depression and three subtypes of insomnia, namely, difficulty initiating sleep (DIS), early morning awakening (EMA), and difficulty maintaining sleep (DMS).

Design:

Cross-sectional and longitudinal study.

Setting:

Community dwellers in Japan.

Participants:

Nationally representative samples of adults aged 65 and over (total N = 4,997) were selected by a multistage stratified random sampling method in 1999 and were interviewed face-to-face in 1999, 2001, 2003, and 2006. Those who responded to the 3rd survey conducted in 2003 and the 4th survey conducted in 2006 were used in this study.

Measurement and Results:

Depression was evaluated according to the 11-item short form of the CES-D scale at 2 points in time. Insomnia subtypes were assessed by self-reported measures. A logistic regression was employed to examine the association between insomnia subtypes and the presence of depression, controlling for relevant factors. A cross-sectional analysis based on the 2003 data demonstrated statistically significant odds ratios (ORs) for DIS and EMA. In the longitudinal study, DIS at the time of the 3rd survey was found to be significantly related to the presence of depression at the time of the 4th survey, with an odds ratio (95%CI) of 1.592 (1.012 to 2.504). EMA (OR 1.070; 95% CI, 0.664 to 1.723) and DMS (OR 1.215; 95% CI, 0.860 to 1.716), however, were not found to be significantly related to the presence of depression.

Conclusion:

The longitudinal study revealed a statistically significant relationship, controlling for other relevant factors, between DIS and the presence of depression three years later, but not between EMA or DMS and depression. Based on our findings, we recommend that the association between insomnia subtypes and depression be studied longitudinally in clinical settings.

Citation:

Yokoyama E; Kaneita Y; Saito Y; Uchiyama M; Matsuzaki Y; Tamaki T; Munezawa T; Ohida T. Association between depression and insomnia subtypes: a longitudinal study on the elderly in Japan. SLEEP 2010;33(12):1693-1702.

DEPRESSION IS SUSPECTED TO BE STRONGLY ASSOCIATED WITH SLEEPING DIFFICULTIES. THE WHO STUDY ON GLOBAL BURDEN OF DISEASES PREDICTED depression would be the second greatest Burden of Disease in 2020 in developed countries.¹ These countries therefore urgently need to address the issue of depression. Depression is prevalent among the aged, and it is known that its prevalence rises after the age of 50 among Japanese people.²

It is reported that insomnia can be a precursor or risk factor in depression and that depression could result in insomnia. Thus, the two diseases apparently have a bidirectional relationship.³ Furthermore, insomnia is listed as a major diagnostic feature of depression in the Diagnostic and Statistical Manual of Mental Disorders, 4th Ed (DSM-IV), attesting to the close association of the two diseases.⁴

This problem has attracted many researchers, and various epidemiological studies have thus far been carried out. In their pioneering longitudinal study, Ford and Kamerow⁵ reported that those who complained of insomnia at the baseline showed a high risk of developing major depression after one year (odds ratio [OR]: 39.8), while those relieved of insomnia by the time of a second survey showed a much lower risk (OR: 1.6). Breslau et al.⁶ observed young adults longitudinally for three years and found that those with a history of insomnia at the baseline had a relative risk of 4.0 of developing major depression by a second examination. Chang et al.⁷ conducted a long-term cohort study in which they followed college graduates up to 45 years. They reported that insomnia in young people could result in the risk of developing depression for at least 30 years. All these studies have confirmed the importance of insomnia as a risk factor of depression and the need for early detection and treatment of insomnia. Riemann and Voderholzer⁸ reviewed eight longitudinal studies (including the three studies mentioned above) conducted before 2000, which examined the relationship between depression and insomnia. In their review, they listed two other studies dealing with that relationship among those aged ≥ 65 years.^9,10 More recently, epidemiological longitudinal studies of the association between insomnia and depression were conducted in various populations: among those ≥ 18 years old in the UK,¹¹ those ≥ 30 years old in Norway,¹² young adults in Switzerland,¹³ the general population in Sweden,¹⁴ and those ≥ 65 years old in South Korea.¹⁵ All of these studies, again, indicate the effect of insomnia on depression for different populations. These reports, however, did not examine the relationship of depression to the different insomnia subtypes: difficulty initiating sleep (DIS), early morning awakening (EMA), and difficulty maintaining sleep (DMS).

Hartz et al.¹⁶ examined risk factors for insomnia subtypes by using a cross-sectional survey and concluded that every insomnia subtype is strongly associated with depressive symptoms. Rodin et al.¹⁷ stated EMA was more closely related to depression than three other sleep problems, including difficulty falling asleep. Although Rodin et al. collected longitudinal data to examine the relationship between sleep disturbances and depression, their analyses of the relationship were basically cross-sectional in nature. Their analyses suggested that “in the same subject, a decrease in depression is associated with a corresponding decrease in early morning awakening.”

Nevertheless, two recent large-scale epidemiological studies in Japan yielded quite different results concerning the relationship between depression and subtypes of insomnia. Using the CES-D to elucidate the relationship of depression to insomnia, Kaneita et al. evaluated 24,686 people (aged ≥ 20 years) throughout Japan for depression.² Sukegawa et al. studied 2,023 residents aged ≥ 65 years in one city using the Geriatric Depression Scale and Pittsburgh Sleep Quality Index.¹⁸ These studies, both of which used questionnaire surveys based on self-rated responses, reported that the closest connection to depression was found not with EMA or DMS, but with DIS. Being cross-sectional studies, however, they could not adequately explore the causal relationship between insomnia subtypes and the presence of depression. We therefore, employed a longitudinal survey on elderly Japanese to examine the temporal association between insomnia subtypes and the presence of depression three years later.

PARTICIPANTS AND METHODS

Selection of Participants

Nihon University has been conducting a nationally representative longitudinal survey on elderly Japanese since 1999 (Nihon University Japanese Longitudinal Study of Aging: NUJLSOA).¹⁹ Six thousand seven hundred elderly Japanese aged ≥ 65 years were sampled by a multistage stratified sampling method in conformity with the population composition at the time. Four thousand nine hundred ninety-seven participants gave informed consent (74.6% retrieval). They were visited and interviewed by well-trained interviewers. Participants aged ≥ 75 years were oversampled by a factor of 2 so that the data for this age group could be analyzed in detail. All the numbers reported in the results are weighted values.

The survey was repeated on the same cohort in 2001, 2003, and 2006, with sample refreshing in 2001 and 2003 for those aged 65 and 66. The present study was based on the 2003 data (number of participants excluding proxy respondents: 4,028) consisting of 1,765 males (45.0%) and 2,263 females (55.0%). All of these participants were included in our cross-sectional studies. Of those, 3,065 without depression in 2003 were included in our longitudinal analyses. We excluded those who had died by the time the 4th survey was conducted in 2006 and those whose responses to the interview were in any way incomplete. The present study employs surveys conducted in 2003 and 2006 because questions on sleep disturbances were only first introduced in the survey conducted in 2003 as a module and kept in the survey conducted in 2006.

Measures

The relationship between the presence of depression and insomnia subtypes was examined by controlling for sleep related, sociodemographic, and health-related factors in the multivariate analyses in our study. Depression, a dependent variable of our study, was evaluated by means of the 11-item short form of the Center for Epidemiological Studies Depression (CES-D) Scale, which had been proposed by Kohout et al.²⁰ from the 20-item standard form CES-D, developed by Radloff.²¹ The time frame of the CES-D is over the previous one week. Shima et al.²² have confirmed the reliability and validity of the Japanese version of the 20-item standard form, for which they proposed 16 as a reasonable cut-off point. The cut-off point, reliability, and validity of the Japanese version of the 11-item CES-D scale used in this study has been examined and reported by Yokoyama et al.²³ The cut-off point for the 11-item CES-D scale was estimated at 7, equivalent to the cut-off point of 16 for the Japanese version of the standard 20-item CES-D scale. In the present study, those with a score ≥ 7 were defined as having depression.

Sleeping problems, including three insomnia symptoms, were measured by self-reported responses to 5 questions utilized by previous studies.^24–26 The five questions used in the survey are the following:

• 1) Difficulty initiating sleep (DIS): “Do you have difficulty falling asleep at night?”

• 2) Early morning awakening (EMA): “Do you wake up too early in the morning and have difficulty getting back to sleep?”

• 3) Difficulty maintaining sleep (DMS): “Do you wake up during the night after you have gone to sleep?”

• 4) Excessive daytime sleepiness (EDS): “Do you feel excessively sleepy during the daytime?”²⁷

• 5) Discomfort feeling in the legs (DFL): “Is your sleep interrupted by an itchiness (creeping sensation) or burning sensation in your legs after you go to bed at night?”^28,29

There are five response categories for these questions: 1, never; 2, seldom; 3, sometimes; 4, often; and 5, always. Those who responded 1 to 3 to the questions were regarded as not having the particular sleeping problem, and those who responded 4 and 5 as having the particular sleeping problem. Insomnia (ANY as a variable name in the tables) was defined, according to a previous study,²⁵ as the condition that resulted in the participant reporting the presence of at least one of the symptoms: DIS, DMS, and EMA.

In addition, two more sleep related variables were controlled for in this study. Subjective sleep sufficiency was examined with the question: “When you wake up in the morning, how often do you feel that you have had a good night's rest?” Response categories to this question are the same as those for insomnia symptoms. Those who responded “never” and “seldom” were regarded as having had insufficient sleep, and those who chose one of the three other responses as having had sufficient sleep. This question is the flip side of the question on non-restorative sleep.

Sleep duration was evaluated by the self-reported response to the question: “How many hours do you sleep a day on average?” The responses were classified into 5 categories (< 6 h; ≥ 6 to < 7 h; ≥ 7 to < 8 h; ≥ 8 to < 9 h; ≥ 9 h) and analyzed. Kaneita et al.² showed the effect of sleep duration on depression as taking on a U-shape, with the bottom at about 7 hours.

Sociodemographic factors including age, gender, educational history, and present place of residence (rural vs. urban dwellers) were controlled for as in most of the previous studies examining the relationship between depression and insomnia. This is because studies have shown that the prevalence of depression and insomnia increases as age increases, and tends to be higher for females.^2,15,16 Many previous studies of the relationship between insomnia and depression have also controlled for educational history^2,15,16 and current place of residence.¹⁵

For analyses, age was treated as a categorical variable. Four dummy variables were created for each 5-year age group: that is 70-74, 75-79, 80-84, and ≥ 85. Those aged 65-69 were the reference age group. Educational history was classified into two groups (junior high school or less, and high school or more). The present place of residence was subdivided into two groups, urban and rural, the former denoting cities and towns, and the latter referring to areas engaged in farming, forestry, and fishing.

Three health-related factors were also controlled for in the present study. Previous studies have indicated a higher prevalence of depression among those with psychological stress,⁷ poor self-rated health,³⁰ and activities of daily living (ADL) difficulties.^31,32 Participants were asked if they had psychological stress in their daily life. This required either a “yes” or “no” answer.

Participants were asked about their self-rated health with the question: “How do you rate your present general health condition: 1 excellent; 2 good; 3 fair; 4 poor; 5 very poor?” Their responses were grouped into 3 categories: 1 and 2 as good, 3 as fair, and 4 and 5 as poor.³³ Two dummy variables were created for good and poor, and fair was treated as the reference group in the analyses.

For ADL, participants were asked if the following seven activities were difficult or not difficult to perform: bathing, eating, dressing, getting in and out of bed or chairs, walking inside the house, going outside, and using a toilet. Participants with at least one difficulty were classified as having difficulty.

The study procedure was approved by the Ethics Committee of Nihon University.

Statistical Analyses

First, we computed Pearson correlation coefficient among the three insomnia symptoms (i.e., DIS, EMA, and DMS) in order to examine the relationship among them. The correlation was computed for all the participants, including those without depression at the time of the 3rd survey conducted in 2003.

We examined statistically the association of insomnia with the presence of depression. Comparison was made by using the mean CES-D scores computed for those who were identified as having depression and those who were not. Comparisons of 2 categories were performed with the unpaired Student t-test, and comparisons of ≥ 3 categories were performed with one-way ANOVA and the Tukey method of multiple comparison.

The relationship between depression and insomnia symptoms was analyzed on the basis of univariate and multivariate logistic regressions. The cross-sectional study using data from the 3rd survey of NUJLSOA (2003) was conducted by a series of logistic regression analyses. The longitudinal study was conducted only on those without depression in 2003. Insomnia variables in 2003 were then regressed on the presence of depression in 2006, controlling for age, gender, educational history, place of residence, sleep duration, EDS, DFL, subjective sleep sufficiency, psychological stress, self-rated health, and ADL.

The 11-item CES-D scale includes the question on whether sleep was restless. In analyzing the relationship between depression and sleeping problems, a possible bias, especially for cross-sectional analyses, may be introduced by including the sleep-related question in the CES-D scale. We therefore conducted additional analyses with the 10-item CES-D scale, excluding this question. By applying a regression equation, following the method of Kohout et al.,²⁰ the cut-off point for the 10-item CES-D scale was estimated at 6.5. Because the cut-off points have to be an integer number, scores of 6 and 7 were used to evaluate the relationship between insomnia subtypes and the presence of depression. To evaluate possible biases, both the cross-sectional and longitudinal study were repeated for the presence of depression defined by the 10-item short form of the CES-D scale with the cut-off points of 6 and 7 in order, but excluding the sleep item question.

Statistical analyses were performed with SAS 9.1.³⁴ The statistical level of significance was P < 0.05 (two-sided tests).

RESULTS

Demographic Characteristics

Table 1 compares the age-bracketed proportional composition of elderly Japanese in the third survey conducted in 2003 and in the census data of 2000. The proportional composition at the baseline was similar to that in the census data in all age brackets except for the ≥ 85 bracket, where data used for this study were slightly lower. The average age and standard deviation (minimum to maximum) for the participants were 73.1 ± 6.4 (65 to 99) for 2003 and 75.7 ± 6.1 (67 to 103) for 2006.

Table 1.

Comparison of age groups between present study and census data

	Age (y)
	65-69	70-74	75-79	80-84	85+
Present study in 2003
Total (%)	32.8	25.2	19.5	13.9	8.7
Men (%)	34.7	27.8	19.0	11.7	6.8
Women (%)	31.2	23.1	19.8	15.6	10.3
Census data in 2000
Total (%)	32.3	26.8	18.9	11.8	10.1

Values given are percentage in age group. All the numbers listed in Table 1 and onward are weighted so they resemble the extracted values from the population composition and not integers.

Correlation among Insomnia Subtypes

Correlation coefficients among the 3 insomnia subtypes are shown in Table 2. Numbers in the upper triangle correspond to all participants and numbers in the lower triangle correspond to those without depression in the third survey conducted in 2003. The correlation was relatively higher for all the participants of the survey compared to those without depression. In all cases, the Pearson correlation coefficients are < 0.4, and the relationship among the 3 insomnia subtypes is considered to be relatively weak.

Table 2.

Pearson correlation coefficient among insomnia subtypes: 2003

Insomnia subtype	DIS	EMA	DMS
DIS	–	0.383*	0.309*
EMA	0.282*	–	0.358*
DMS	0.248*	0.321*	–

Numbers in the upper triangle are based on all participants (N ≅ 4,000). Numbers in the lower triangle are based on participants without depression (N ≅ 3,050).

DIS refers to difficulty initiating sleep; EMA, early morning awakening; DMS, difficulty maintaining sleep.

^*Statistically significant at 0.01 level.

Prevalence of Insomnia Symptoms and Depression

Table 3 compares participants' attributes. When estimated by means of the 11-item short form CES-D scale, the presence of depression among elderly Japanese (≥ 65 years) stood at 13.8%, as shown in the first line of Table 3. Overall prevalence of insomnia (ANY) for the sample was 27.8%. The prevalence of insomnia subtypes from highest to lowest is as follows: DMS (22.9%); EMA (11.5%); DIS (11.1%). Prevalence of sleep disturbances was 8.1% for EDS and 3.7% for DFL. Overall, 16.8% of elderly Japanese reported subjective insufficient sleep.

Table 3.

Sample distribution, prevalence of depression and mean CES-D scores by categories of sample attributes and self-reported responses, and results of statistical test for mean CES-D scores among categories in 2003

Baseline items in 2003	Percentage of category (%)	Prevalence of depression* (%)	11-item short forms of the CES-D score (mean ± SD)	Statistical test for mean CES-D scores#
Category
Total	100.0	13.8	4.1 ± 2.7
Age (year)
65-69	32.8	10.2	3.8a ± 2.7	P <= 0.001#1
70-74	25.2	13.9	4.2 ± 3.0
75-79	19.5	15.9	4.3b ± 2.9
80-84	13.9	19.3	4.4c ± 2.2
85+	8.7	15.9	4.4d ± 2.1
Gender
Women	55.0	16.0	4.2 ± 2.8	P = 0.006
Men	45.0	11.1	4.0 ± 2.5
Sleep duration (h)
< 6	10.7	21.2	4.7a ± 3.4	P < 0.001#2
≥ 6 to 7	19.7	13.1	4.0b ± 5.0
≥ 7 to 8	23.9	10.7	3.8c ± 2.5
≥ 8 to 9	29.8	12.0	3.9d ± 2.5
≥ 9	15.9	18.0	4.5e ± 2.7
Educational history
Junior high school	55.4	15.9	4.3 ± 2.6	P < 0.001
More than high school	44.6	11.3	3.9 ± 2.8
Place of residence
Urban	68.0	13.9	4.2 ± 2.4	P = 0.037
Rural	32.0	13.8	4.0 ± 2.8
Sleep disturbance
Any ins: Yes	27.8	21.0	4.8 ± 3.2	P < 0.001
No	72.2	10.7	3.8 ± 2.4
DIS: Yes	11.1	34.8	6.0 ± 3.7	P < 0.001
No	88.9	11.2	3.9 ± 2.4
EMA: Yes	11.5	29.1	5.7 ± 3.6	P < 0.001
No	88.5	11.9	3.9 ± 2.5
DMS:Yes	22.9	20.2	4.8 ± 3.2	P < 0.001
No	77.1	11.9	3.9 ± 2.5
EDS: Yes	8.1	28.1	5.2 ± 3.9	P < 0.001
No	91.9	12.6	4.0 ± 2.5
DFL: Yes	3.7	24.4	5.3 ± 3.5	P < 0.001
No	96.3	13.4	4.0 ± 2.6
Subjective sleep sufficiency
Insufficient	16.8	23.3	5.0 ± 3.2	P < 0.001
Sufficient	83.2	11.7	3.9 ± 2.5
Psychological stress
Yes	25.1	32.9	5.7 ± 3.5	P < 0.001
No	74.9	7.4	3.6 ± 2.1
Self-rated health
Good	31.9	5.6	3.3a ± 2.1	P < 0.001#3
Fair	40.3	10.1	3.9b ± 2.3
Poor	27.8	32.5	5.7c ± 3.3
Activity of daily living (ADL)
Any	12.7	35.9	6.0 ± 3.3	P < 0.001
None	87.3	11.8	3.9 ± 2.5

^*Calculation of prevalence (%) was conducted by 11-item Short Forms of CES-D (cut-off point = 7).

^#Comparisons of 2 categories were performed with unpaired Student t-test; and comparison of ≥ 3 categories were performed with one way ANOVA and Tukey method of multiple comparison

^#1Statistically significant pairs among the age groups were ^a and ^b, ^a and ^c, ^a and ^d, ^a and ^e.

^#2Statistically significant pairs among the sleep durations were ^a and ^b, ^a and ^c, ^d and ^e, ^c and ^e.

^#3Statistically significant pairs among the categories of self-rated health were ^a and ^b, ^a and ^c, ^b and ^c. Any ins refers to any of DIS, EMA or DMS; DIS, difficulty initiating sleep; EMA, early morning awakening; DMS, difficulty maintaining sleep; EDS, excessive daily sleepiness; DFL, discomfort feeling in the legs.

The prevalence of depression in the insomnia symptom groups was: DIS (34.8%); EMA (29.1%); and DMS (20.2%). Each insomnia symptom had significantly different average CES-D scores. DIS had the highest average “yes” responses. Elderly Japanese with EDS, DFL, and subjective insufficient sleep seemed to report a relatively higher prevalence of depression: 28.1%, 24.4%, and 23.3%, respectively.

Relationship between Insomnia Symptoms and Depression

Table 4 shows results of the cross-sectional and longitudinal study on the relationship between insomnia (defined as having any one of the insomnia subtypes) and depression in both univariate and multivariate analyses, thereby allowing a comparison of results from previously conducted studies. For multivariate analyses, sleep related, sociodemographic, and health-related factors were controlled. In the cross-sectional study, insomnia (ANY) was statistically significantly related to the presence of depression in both univariate (OR: 2.217 with 95% CI 1.832 to 2.682) and multivariate analyses (aOR: 1.272 with 95% CI 1.000 to 1.618). However, in the longitudinal study, insomnia (ANY) was significant only at 0.1 level in the multivariate analysis (aOR: 1.293 with 95% CI 0.959 to 1.744), while controlling for sleep related, sociodemographic, and health related factors.

Table 4.

Association between depression and ANY insomnia subtypes by univariate and multivariate analyses in the cross-sectional study and longitudinal study^a

	Univariate logistic regression analysis
Baseline items in 2003	Cross-sectional study in 2003			Longitudinal study from 2003 to 2006
Category	ORb	95% CI	P-value	ORb	95% CI	P-value
Sleep disturbance
ANY ins: Yes	2.217	1.832 to 2.682	< 0.001	1.460	1.111 to 1.918	0.007
No	1.000	referent		1.000	referent
	Multivariate logistic regression analysis
Baseline items in 2003	Cross-sectional study in 2003			Longitudinal study from 2003 to 2006
Category	aORc	95% CI	P-value	aORc	95% CI	P-value
Age (y)
65-69	1.000	referent		1.000	referent
70-74	1.358	1.005 to 1.836	0.046	1.252	0.894 to 1.753	0.191
75-79	1.626	1.183 to 2.235	0.003	1.020	0.688 to 1.511	0.922
80-84	2.008	1.399 to 2.883	0.000	1.394	0.876 to 2.218	0.161
85+	1.609	0.959 to 2.700	0.072	1.384	0.654 to 2.928	0.396
Gender
Women	1.361	1.076 to 1.722	< 0.001	1.298	0.979 to 1.720	0.070
Men	1.000	referent		1.000	referent
Educational history
Junior high school	0.864	0.742 to 1.005	0.059	0.924	0.777 to 1.098	0.368
More than high school	1.000	referent		1.000	referent
Place of residence
Urban	1.073	0.804 to 1.374	0.577	1.000	0.743 to 1.346	0.999
Rural	1.000	referent		1.000	referent
Sleep duration (h)
< 6	1.337	0.905 to 1.973	0.144	0.875	0.508 to 1.505	0.629
≥ 6 to 7	1.015	0.722 to 1.428	0.932	1.259	0.851 to 1.865	0.249
≥ 7 to 8	1.000	referent		1.000	referent
≥ 8 to 9	1.083	0.783 to 1.497	0.631	1.218	0.842 to 1.764	0.295
≥ 9	1.289	0.884 to 1.877	0.187	1.126	0.693 to 1.828	0.632
Sleep disturbance
ANY ins: Yes	1.272	1.000 to 1.618	< 0.001	1.293	0.959 to 1.744	0.092
No	1.000	referent		1.000	referent
EDS: Yes	1.418	1.008 to 1.995	0.045	0.850	0.497 to 1.453	0.551
No	1.000	referent		1.000	referent
DFL: Yes	0.853	0.506 to 1.439	0.551	1.096	0.494 to 2.429	0.822
No	1.000	referent
Subjective sleep sufficiency
Insufficient	1.867	1.426 to 2.444	< 0.001	1.045	0.711 to 1.536	0.822
Sufficient	1.000	referent		1.000	referent
Psychological stress
Yes	5.357	4.246 to 6.758	< 0.001	1.633	1.187 to 2.246	0.003
No	1.000	referent		1.000	referent
Self-rated health
Good	0.671	0.491 to 0.918	0.013	0.824	0.597 to 1.137	0.239
Fair	1.000	referent		1.000	referent
Poor	2.917	2.254 to 3.774	< 0.001	2.533	1.793 to 3.579	< 0.001
Activity of daily living (ADL)
Any	1.496	1.053 to 2.124	0.024	1.251	0.691 to 2.262	0.460
None	1.000	referent		1.000	referent

^aThe cross-sectional study was conducted on baseline subjects in 2003; the longitudinal study was conducted on subjects who participated in both the 2003 and 2006 surveys.

^bCrude odds ratio due to univariate logistic regression analysis.

^cAdjusted odds ratio due to multivariate logistic regression analysis. ANY ins refers to any of difficulty initiating sleep, early morning awakening, or difficulty maintaining sleep; EDS, excessive daytime sleepiness; DFL, discomfort feeling in the legs; 95%CI, 95% confidence interval.

Table 5 illustrates the results of the cross-sectional study involving all participants of the third survey. Using a univariate analysis, all insomnia symptoms were found to be significantly related to the presence of depression. The multivariate analysis, on the other hand, demonstrated that only DIS and EMA had a significant relationship to the presence of depression, after controlling for age, gender, educational history, place of residence, sleep duration, EDS, DFL, subjective sleep sufficiency, psychological stress, self-rated health, and ADL. The adjusted odds ratios (95% CI) were 2.020 (1.455 to 2.805) and 1.441 (1.025 to 2.026), respectively.

Table 5.

Association between depression and insomnia subtypes in the cross-sectional study^a

	Logistic regression analyses
Baseline items in 2003	Univariate analysis			Multivariate analysis
Category	ORb	95% CI	P-value	aORc	95% CI	P-value
Age (year)
65-69	1.000	referent		1.000	referent
70-74	1.423	1.101 to 1.838	0.006	1.359	1.001 to 1.844	0.049
75-79	1.675	1.282 to 2.189	< 0.001	1.620	1.174 to 2.235	0.003
80-84	2.121	1.584 to 2.841	< 0.001	1.986	1.375 to 2.867	< 0.001
85+	1.673	1.117 to 2.508	0.013	1.516	0.889 to 2.583	0.126
Gender
Women	1.526	1.257 to 1.853	< 0.001	1.327	1.047 to 1.681	0.019
Men	1.000	referent		1.000	referent
Educational history
Junior high school	0.669	0.552 to 0.811	< 0.001	0.790	0.622 to 1.004	0.054
More than high school	1.000	referent		1.000	referent
Place of residence
Urban	0.987	0.807 to 1.208	0.898	1.066	0.830 to 1.370	0.616
Rural	1.000	referent		1.000	referent
Sleep duration (h)
< 6	2.238	1.628 to 3.075	< 0.001	1.162	0.776 to 1.742	0.466
≥ 6 to 7	1.261	0.939 to 1.693	0.124	0.944	0.667 to 1.337	0.747
≥ 7 to 8	1.000			1.000	referent
≥ 8 to 9	1.132	0.858 to 1.495	0.380	1.098	0.792 to 1.521	0.575
≥ 9	1.823	1.337 to 2.487	<0.001	1.349	0.923 to 1.972	0.122
Sleep disturbance
DIS: Yes	4.208	3.329 to 5.319	< 0.001	2.020	1.455 to 2.805	< 0.001
No	1.000	referent		1.000	referent
EMA: Yes	3.049	2.403 to 3.870	< 0.001	1.441	1.025 to 2.026	0.036
No	1.000	referent		1.000	referent
DMS: Yes	1.869	1.525 to 2.289	< 0.001	0.819	0.613 to 1.094	0.176
No	1.000	referent		1.000	referent
EDS: Yes	2.716	2.064 to 3.574	< 0.001	1.319	0.929 to 1.875	0.122
No	1.000	referent		1.000	referent
DFL: Yes	2.131	1.465 to 3.101	< 0.001	0.788	0.463 to 1.342	0.381
No	1.000	referent		1.000	referent
Subjective sleep sufficiency
Insufficient	2.290	1.839 to 2.852	< 0.001	1.758	1.335 to 2.315	< 0.001
Sufficient	1.000	referent		1.000	referent
Psychological stress
Yes	6.104	4.997 to 7.455	< 0.001	5.169	4.088 to 6.537	< 0.001
No	1.000	referent		1.000	referent
Self-rated health
Good	0.531	0.397 to 0.709	< 0.001	0.664	0.485 to 0.910	0.011
Fair	1.000	referent		1.000	referent
Poor	4.270	3.435 to 5.307	< 0.001	2.187	3.688 to 3.231	< 0.001
Activities of daily living (ADL)
Any	4.180	3.211 to 5.443	< 0.001	1.419	0.991 to 2.032	0.056
None	1.000	referent		1.000	referent

^aThe cross-sectional study was conducted on baseline subjects in 2003.

^bCrude odds ratio due to univariate logistic regression analysis.

^cAdjusted odds ratio due to multivariate logistic regression analysis. The logistic regression analysis was conducted as a response variable of CES-D (Score ≥ 7) by the short form. DIS refers to difficulty initiating sleep; EMA, early morning awakening; DMS, difficulty maintaining sleep; EDS, Exceesive daytime sleepiness; DFL, discomfort feeling in the legs; 95%CI, 95% confidence interval.

Table 6 represents the results of the longitudinal study of those without depression in 2003. The presence of depression in 2006 in the participants who had not been suffering from the disease at the time of the 3rd survey was regarded as the response variable and insomnia symptoms as the explanatory variables, after controlling for sleep related, sociodemographic, and health-related factors. DIS, EMA, and DMS were all shown to be significantly related to depression in the univariate analysis; but in the multivariate analysis DIS was the only symptom with a significant relationship to the presence of depression, with an adjusted odds ratio (95% CI) of 1.592 (1.012 to 2.504).

Table 6.

Association between depression and insomnia subtypes in the longitudinal study (2003-2006)^a for subjects without depression in 2003

	Logistic regression analyses
Baseline items in 2003	Univariate analysis			Multivariate analysis
Category	ORb	95% CI	P-value	aORc	95% CI	P-value
Age (y)
65-69	1.000	referent		1.000	referent
70-74	1.333	0.966 to 1.840	0.081	1.219	0.869 to 1.711	0.252
75-79	1.144	0.788 to 1.660	0.480	1.006	0.677 to 1.493	0.978
80-84	1.728	1.131 to 2.640	0.012	1.347	0.841 to 2.157	0.215
85+	1.657	0.851 to 3.224	0.137	1.344	0.634 to 2.849	0.441
Gender
Women	1.253	0.966 to 1.627	0.090	1.327	1.000 to 1.761	0.050
Men	1.000	referent		1.000	referent
Educational history
Junior high school	0.874	0.675 to 1.131	0.306	0.906	0.680 to 1.206	0.499
More than high school	1.000	referent		1.000	referent
Place of residence
Urban	0.948	0.721 to 1.246	0.700	0.968	0.718 to 1.305	0.832
Rural	1.000	referent		1.000	referent
Sleep duration (h)
< 6	1.147	0.694 to 1.894	0.593	0.849	0.487 to 1.481	0.564
≥ 6 to 7	1.378	0.948 to 2.004	0.093	1.227	0.824 to 1.828	0.314
≥ 7 to 8	1.000	referent		1.000	referent
≥ 8 to 9	1.246	0.878 to 1.770	0.219	1.218	0.839 to 1.767	0.300
≥ 9	1.319	0.840 to 2.073	0.229	1.175	0.722 to 1.912	0.517
Sleep disturbances
DIS: Yes	2.042	1.391 to 2.997	< 0.001	1.592	1.012 to 2.504	0.044
No	1.000	referent		1.000	referent
EMA: Yes	1.541	1.030 to 2.306	0.035	1.070	0.664 to 1.723	0.782
No	1.000	referent		1.000	referent
DMS: Yes	1.432	1.066 to 1.925	0.017	1.215	0.860 to 1.716	0.269
No	1.000	referent		1.000	referent
EDS: Yes	1.401	0.869 to 2.260	0.166	0.819	0.477 to 1.408	0.471
No	1.000	referent		1.000	referent
DFL:Yes	2.120	1.125 to 3.995	0.020	0.990	0.440 to 2.230	0.981
No	1.000	referent		1.000	referent
Subjective sleep sufficiency
Insufficient	1.093	0.760 to 1.573	0.631	1.051	0.713 to 1.547	0.803
Sufficient	1.000	referent		1.000	referent
Psychological stress
Yes	1.846	1.375 to 2.479	< 0.001	1.553	1.125 to 2.145	0.008
No	1.000	referent		1.000	referent
Self-rated health
Good	0.745	0.546 to 1.016	0.063	0.794	0.574 to 1.099	0.164
Fair	1.000	referent		1.000	referent
Poor	2.589	1.881 to 3.563	< 0.001	2.517	1.778 to 3.562	< 0.001
Activity of daily living (ADL)
Any	2.223	1.312 to 3.766	0.003	1.207	0.664 to 2.193	0.538
None	1.000	referent		1.000	referent

^aThe longitudinal study was conducted on subjects participating in the 2003 and 2006 surveys, with the former serving as the baseline.

^bCrude odds ratio due to univariate logistic regression analysis.

^cAdjusted odds ratio due to multivariate logistic regression analysis, the logistic regression analysis was conducted as a response variable of CES-D (score ≥ 7) by Shorter Form in 2006.

DIS refers to difficulty initiating sleep; EMA, early morning awakening; DMS, difficulty maintaining sleep; EDS, excessive daytime sleepiness; DFL, discomfort feeling in the legs; 95%CI, 95% confidence interval.

It is worth mentioning that the results indicate that gender, stress, and self-rated health among the control variables had statistically significant effects on the presence of depression. The effect of psychological stress was substantively about the same as DIS with an adjusted odds ratio (95% CI) of 1.553 (1.125 to 2.145). Those who reported their health as poor at the baseline survey were more likely to have depression in the following survey with an adjusted odds ratio (95% CI) of 2.517 (1.778 to 3.562).

In addition, we examined the relationship between the presence of depression and the insomnia subtypes by controlling for the baseline participants' CES-D scores that vary from 0 to 6 with other control variables. The result of this analysis (not shown) differed slightly from the result shown in Table 6. The adjusted odds ratios (95% CI) for DIS, EMA, and DMS were 1.509 (0.957 to 2.381), 1.023 (0.634 to 1.651), and 1.206 (0.853 to 1.706), respectively. Although the significance level of the effect of DIS on depression was slightly affected by controlling for the CES-D score at baseline, the general tendency of the effect of insomnia subtypes was not affected.

Comparison of the Sleep-Item-Excluded 10-Item CES-D Scale with the 11-Item CES-D Scale

The results of the cross-sectional and longitudinal analyses were re-examined based on the 10-item CES-D scale, which, excluding the sleep item, is the 11-item short form of CES-D with 2 different cut-off points (6 and 7). As shown in Table 7, the adjusted odds ratios in the cross-sectional analyses (95% CI) were 1.819 (1.327 to 2.493) for DIS with the cut-off point of 6, and 1.721 (1.214 to 2.439) with the cut-off point of 7. For EMA, the adjusted odds ratios (95% CI) were 1.262 (0.910 to 1.748) with the cut-off point of 6, and 1.476 (1.030 to 2.114) with the cut-off point of 7. In the longitudinal analyses, the adjusted odds ratios (95% CI) were 1.659 (1.047 to 2.629) for DIS with the cut-off point of 6, and 1.444 (0.936 to 2.227) with the cut-off point of 7. Although there are slight differences between the 10-item scale and the 11-item scale, the general tendency appears to be similar.

Table 7.

Comparison of adjusted odds ratios (95% CI) of 10-item CES-D scale and 11-item CES-D scale based on multivariate analyses of the association of insomnia subtypes with depression in cross-sectional and longitudinal study

Insomnia Subtypes	10-item CES-D Scale				11-item CES-D Scale*
	Cut-off point = 6		Cut-off point = 7		Cut-off point = 7
	Cross-sectional study
	aOR#	95%CI	aOR	95%CI	aOR	95%CI
DIS	1.819	(1.327 to 2.493)	1.721	(1.214 to 2.439)	2.020	(1.455 to 2.805)
EMA	1.262	(0.910 to 1.748)	1.476	(1.030 to 2.114)	1.441	(1.025 to 2.026)
DMS	0.874	(0.669 to 1.142)	0.899	(0.662 to 1220)	0.819	(0.613 to 1.094)
	Longitudinal study
	aOR	95%CI	aOR	95%CI	aOR	95%CI
DIS	1.659	(1.047 to 2.629)	1.444	(0.936 to 2.227)	1.592	(1.012 to 2.504)
EMA	1.242	(0.773 to 1.993)	1.153	(0.735 to 1.808)	1.070	(0.664 to 1.723)
DMS	1.304	(0.916 to 1.854)	1.263	(0.905 To 1.763)	1.215	(0.860 to 1.716)

^*Results were taken from Table 5 for the cross-sectional study and from Table 6 for the longitudinal study.

^#aOR: adjusted odds ratio and 95%CI; 95% confidence interval. Logistic regression analysis was conducted by controlling for age, gender, educational history, place of residence, sleep duration, excessive day time sleepiness, discomfort feeling in the legs, subjective sleep sufficiency, psychological stress, self-rated health, and activities of daily living. DIS refers to difficulty initiating sleep; EMA, early morning awakening; DMS, difficulty maintaining sleep.

From the cross-sectional analyses, DIS and EMA appeared to be associated with the presence of depression. Only DIS, however, demonstrated a significant relationship to the development of depression with the cut-off point of 6 when the sleep item was removed from the 11-item short form of the CES-D Scale.

DISCUSSION

An important feature of this study is that it was designed longitudinally so that the relationship between insomnia symptoms and depression could be examined in temporal order. A second feature is that we differentiated among the diverse components of insomnia, namely, DIS, EMA, and DMS. The study indicated that DIS had a statistically significant relationship to developing depression three years later, while EMA and DMS did not. As far as we know, there have been no reports, based on longitudinal studies of elderly Japanese, on this relationship. Our findings thus provide an important reference point for further studies on the association between insomnia subtypes and depression.

A sizable literature supports insomnia as a risk factor for depression. Yet, insomnia is a composite pathological entity consisting of such various symptoms as DIS, EMA, and DMS, and it has not been established which of these contribute to the development of depression. On the basis of the survey results on the Japanese general population in 2000, Kaneita et al.² reported that DIS, EMA, and DMS were related independently to depression. Although their investigation enabled us to understand the significance and separate correlation between these symptoms and depression, because of the cross-sectional nature of the data, their study could not assess the direction of effect between each symptom and the presence of depression. Our current study takes us one step further towards understanding the relationship between the presence of depression and subtypes of insomnia.

Previous studies on depression using cross-sectional data have attached more importance to EMA than DIS.^3,35 Although in our own univariate and multivariate cross-sectional analyses a significant relationship was found between EMA and the presence of depression, no statistically significant relationship between them was found in our multivariate longitudinal analysis. In the present investigation we adjusted for these associated phenomena by employing multivariate analysis; otherwise it might have been possible for EMA to present a seemingly high correlation. To elucidate the relationship between insomnia symptoms and depression, it appears essential to design longitudinal studies with due attention to the adjustment of associative factors. As a result, we found a statistically significant relationship between DIS and the presence of depression, but not between EMA and depression. In addition, we found a statistically and substantively significant effect of psychological stress and poor self-rated health on the presence of depression. These relationships may need to be studied further using longitudinal data.

Our findings, moreover, may shed light on the bidirectional relationship between insomnia and depression. Buysse et al.,¹³ using a longitudinal cohort study of young adults, expressed the bidirectional relationship between insomnia and depression in temporal order without specifying particular insomnia subtypes. They stated that “insomnia predicted future MDE (major depressive episodes) and that MDE tended to predict future insomnia.” Based on our study, DIS seems to precede the presence of depression. Rodin et al.¹⁷ followed 196 subjects aged 62 and over for three years and indicated that depression is related to sleep disturbance, particularly EMA. EMA possibly arose as a result of depression. If one were to examine, using longitudinal data, the relationship between depression and insomnia as a composite measure that includes DIS and EMA, one might find a bidirectional relationship between depression and insomnia. If this relationship between DIS, depression, and EMA holds in the cross-sectional studies, we should observe the significant association both between DIS and depression, and depression and EMA, as observed in the present study. However, the results from the cross-sectional studies may be misleading. In the future, we plan to study the relationship between the presence of depression and EMA in the opposite direction, as examined by Rodin et al.¹⁷

Although our study produced different results from those of previous studies, it is quite difficult to compare studies on the relationship between insomnia and depression because the definition, criteria, and measures of insomnia and depression were different for each study. One plausible reason why we had different results is that our criteria for insomnia symptoms were less strict and, therefore, individuals with transient and minor sleep disturbances may have been included. More specifically, it may have caused a relative increase in sleep maintenance complaints among the elderly population. As a result, minor transient insomniacs may have diluted the clinical sample of sleep maintenance insomniacs, possibly reducing the specificity of this particular symptom for predicting later presence of depression.

Insomnia, which is defined rather loosely in our study, could be another concern for our longitudinal study. As mentioned before, questions on insomnia subtypes used in the study relied on self-reports without reference to time. Our findings were not particularly robust, and odds ratios in our results were relatively low compared to ones found in previous longitudinal studies. In order to examine the effect of changes in definitions of insomnia, we have run another model for a longitudinal study with more strict criteria for insomnia subtypes. We defined presence of insomnia subtypes only for those who answered “always” (we included both “often” and “always” in our present study because of obtaining stable estimates). Adjusted odds ratios (95% CI) for DIS increased from 1.592 (1.012 to 2.504) to 2.086 (1.091 to 3.989) with a significance level of 0.026. This indicates that to examine the relationship between DIS and depression in the longitudinal study may be warranted. In any case, given that epidemiological data on the relationship between insomnia subtypes and the development of depression are relatively scarce, it is hoped that many more longitudinal studies on this subject will be carried out and more knowledge accumulated.

Study Limitations

This study has a few limitations. First, our study participants were limited to those aged 65 and older. Studies with the general population should be conducted to see whether our conclusions apply to the population at large.

Second, we evaluated depression using only the CES-D scale. Neither the DSM-IV⁴ nor the ICD-10³⁶ are applicable to a community-based survey, and, therefore, could not be employed to diagnose clinically the subjects of our study. Many epidemiological studies have, in fact, relied on the CES-D scale, as its reliability and validity have been sufficiently established. Nevertheless, given that comparing disease prevalence among regions is problematic³⁷ and that clinical diagnoses and CES-D scale evaluations of depression do not always agree,³⁸ the use of other diagnostic methods should be considered in the future.

Third, because at the time of the baseline survey we did not ask participants whether they had ever had depression before, we may be examining the relationship between insomnia subtypes and recurrence of depression rather than between insomnia subtypes and initial onset of depression.

Fourth, as is commonly found in epidemiological work, we did not have objective data on disturbed sleep. While it is desirable to obtain objective data, it is very difficult to conduct such studies on a community scale. More objective and yet simple epidemiological research techniques should be developed. In addition, we did not make an insomnia diagnosis using established criteria. We simply used endorsement of specific insomnia subtypes which did not include a time frame or frequency. Moreover, because of the data limitation, we did not utilize daytime impairment/dysfunction as required by most current insomnia nosologies.

Fifth, the association between insomnia symptoms and the presence of depression was evaluated only at two points in the three-year period. No specific information on depression was obtained regarding when during the three years the condition developed, nor how long the condition lasted. In addition, three years may be too long to observe the association between insomnia symptoms and depression. Although, as in the present study, several longitudinal studies on sleep disorder have evaluated results using information from two study points, the number of observation points chosen and the duration of the study should receive careful consideration in future studies.

Lastly, we should point out the possibility of non-response bias in our study. Because a longitudinal study is destined to lose participants by death and lost-to-follow-up, only those who have managed to maintain health tend to be surveyed in the follow-up. In addition, proxy responses and non response to CES-D questions tend to increase as participants age. In order for the study of precursor symptoms and risk factors of insomnia to be meaningful, due attention in future studies should be given to the upper age limit in studies.

Despite these limitations, we hope that our study results will contribute to the future progress of mental health care and the solution of sleeping problems. If our findings are found to be true by clinical research, DIS subtype is possibly a better predictor of depression than overall insomnia severity. By understanding better the relationship between depression and insomnia subtypes, we may have a chance to lower the prevalence of depression among older adults in Japan.

DISCLOSURE STATEMENT

This was not an industry supported study. The authors have indicated no financial conflicts of interest.