Bedroom Light Exposure at Night and the Incidence of Depressive Symptoms: A Longitudinal Study of the HEIJO-KYO Cohort

Abstract

Previous studies have indicated that minimal exposure to light at night (LAN) increases depression risk, even at 5 lux, in nocturnal and diurnal mammals. Although such low-level LAN may affect human circadian physiology, the association between exposure to LAN and depressive symptoms remains uncertain. In the present study, bedroom light intensity was measured objectively, and depressive symptoms were assessed, during 2010–2014 in Nara, Japan. Of 863 participants (mean age = 71.5 years) who did not have depressive symptoms at baseline, 73 participants reported development of depressive symptoms during follow-up (median, 24 months). Compared with the “dark” group (average of <5 lux; n = 710), the LAN group (average of ≥5 lux; n = 153) exhibited a significantly higher depression risk (hazard ratio = 1.89; 95% CI: 1.13, 3.14), according to a Cox proportional hazards model adjusting for age, sex, body mass index, and economic status. Further, the significance remained in a multivariable model adjusting for hypertension, diabetes, and sleep parameters (hazard ratio = 1.72; 95% CI: 1.03, 2.89). Sensitivity analyses using bedroom light data with a cutoff value of ≥10 lux suggested consistent results. In conclusion, these results indicated that exposure to LAN in home settings was independently associated with subsequent depression risk in an elderly general population.

circadian rhythm, depression, depressive symptoms, epidemiology, light at night

Editor’s note:An invited commentary on this article appears on page 435, and the authors’ response appears on page 439.

The endogenous, circadian timing system in humans has adapted to the natural environment based on daytime light and nighttime dark. However, exposure to light at night (LAN) is the strongest disruptor of circadian physiology and behavior (1). Modern society is closely dependent on the convenient use of LAN, possibly resulting in circadian misalignment between internal biological and environmental rhythms (2). Thus, all modern humans may be at a risk of circadian misalignment and related diseases, given that previous evidence suggests that biological rhythmicity is affected by minimal LAN exposure (3).

Depression prevalence has increased in recent decades, and a significant public health burden associated with depression is also increasing because of the associated risk of dementia, cardiovascular diseases, and early mortality (4–6). However, depression etiology remains poorly understood, and an effective preventive method has not been established. Depression is frequently accompanied by sleep disturbances and circadian misalignment, such as altered patterns in sleep/wake cycles and melatonin secretion (7–9); these present a possible mechanism underlying depression incidence. A greater depression risk observed in night-shift workers is significantly correlated to LAN exposure and to circadian misalignment between internal biological and environmental rhythms (10, 11). A study using hamsters demonstrated that LAN exposure for 4 weeks significantly increased depression-like behaviors, even at 5 lux (12). Another study using diurnal rodents reported consistent results that 5-lux LAN exposure increases depressive behavior (13). In humans, 5-lux LAN exposure in home settings is significantly associated with depression symptoms; however, to our knowledge, evidence is limited to a cross-sectional analysis (14). Here we tested the longitudinal association of LAN exposure with the incidence of depressive symptoms in an elderly general population.

METHODS

Participants and study protocol

We analyzed data from the HEIJO-KYO cohort, focusing on the housing environments and health outcomes among 1,127 community-dwelling elderly individuals aged ≥60 years, most retired from their work, to minimize the effect of office environments and current work. A baseline survey was conducted during 2010–2014, and the study protocol was reported previously (14). In brief, the homes of participants were visited, predominantly on weekdays, and demographic and medical information were obtained using a standardized questionnaire. We then initiated measurements of bedroom light intensity and instructed the participants to maintain a standardized sleep diary for 2 nights. A follow-up survey regarding depressive symptoms was conducted between August 2014 and January 2015, using a validated tool. All participants provided written informed consent. The study protocol was approved by Nara Medical University’s ethics committee.

Measurement of LAN exposure

Bedroom light intensity during the nighttime (self-reported bedtime to rising time) was measured for 2 consecutive nights at 1-minute intervals using a portable light meter (LX-28SD; Sato Shouji Inc., Kanagawa, Japan), which was placed facing the ceiling at the head of the participant’s bed, 60 cm above the floor (illuminance sensitivity, 0–100,000 lux; resolution, 1 lux (up to 2,000 lux); and accuracy, within 4% reading + 2 digits (up to 2,000 lux)). Three LAN parameters were defined: average intensity of nighttime light exposure (LAN average), total time of nighttime light intensity of ≥5 lux (LAN 5), and total time of nighttime light intensity of ≥10 lux (LAN 10). These predefined cutoff thresholds were based on previous knowledge underlying the association of LAN exposure with depressive symptoms (12, 13).

Ascertainment of depressive symptoms

Depressive symptoms were measured at baseline and follow-up using the short version of the Geriatric Depression Scale (GDS), a 15-item scale. Scores of ≥6 indicated the presence of depressive symptoms. Previous studies validating this cutoff value reported a sensitivity and specificity for clinically diagnosed depression of 0.85–0.91 and 0.68–0.78, respectively (15). The cutoff value of 6 might be optimal, although a cutoff value of 5, with higher sensitivity but lower specificity, was used in our previous study (14).

Measurement of covariates

Body mass index (BMI) was calculated as body weight (kg) divided by the square of body height (m²). Current lifestyle habits (drinking alcohol and smoking), socioeconomic status, and medication use were evaluated using a self-administered questionnaire. The presence of hypertension was determined on the basis of medical history and current use of antihypertensive drugs. Diabetes mellitus was determined on the basis of medical history, current use of antidiabetic therapy, and levels of fasting plasma glucose and glycated hemoglobin. A standardized sleep diary was used to estimate bedtimes and rising times. These sleep/wake patterns were objectively measured using an actigraph (Actiwatch 2; Philips Respironics Inc., Murrysville, Pennsylvania) worn on the nondominant wrist, where sleep onset and sleep offset were automatically detected by Actiware, version 5.5 (Philips Respironics Inc.), using the default algorithm. Sleep onset was the first minute followed by 10 minutes of immobility, and sleep offset was the last minute following the 10-minute period of immobility. Subjective sleep quality was assessed using the Pittsburgh Sleep Quality Index questionnaire, and sleep disturbances were defined as a score ≥6 (16). Physical activity was assessed using the International Physical Activity Questionnaire (Japanese version), containing questions on the time spent performing moderate and vigorous activities and walking per week (17). The day length (sunrise to sunset), based on the measurement days in Nara, Japan, was extracted from the website of the National Astronomical Observatory of Japan. Light exposures in the morning (4 hours after rising time) and evening (4 hours before bedtime) were measured for 2 consecutive days at 1-minute intervals using the wrist light meter noted above, worn on the nondominant wrist.

Statistical analysis

Means and medians were compared between the dichotomous groups using the unpaired t test and the Mann-Whitney U test, respectively. The χ² test was used to compare categorical data. Data were averaged on light exposure, time of sleep as registered in the diary, and day length on 2 measurement days. The risk of depressive symptoms associated with LAN exposure was estimated using a Cox proportional hazards model adjusting for age (per 5 years), sex, and variables associated with incident depressive symptoms (Web Table 1, available at https://academic.oup.com/aje), such as BMI (per unit), household income (≥4 million vs. <4 million Japanese yen/year (cutpoint: approximately $35,000)), hypertension (yes vs. no), diabetes (yes vs. no), sleep disturbances (yes vs. no), bedtime (per 1 hour), and duration in bed (per 1 hour). The mean, median, or proportion was substituted for missing data for these independent variables. All analyses were performed using SPSS, version 19.0, for Windows (SPSS Inc., Chicago, Illinois). A 2-sided P value of <0.05 was considered statistically significant. Time-dependent bedroom light intensity in groups with and without incident depressive symptoms was smoothed by a generalized additive model to minimize the generalized cross-validation score (18) using R (R Foundation for Statistical Computing, Vienna, Austria) (19).

RESULTS

Of 1,108 participants who completed the GDS questionnaire and LAN exposure measurement, 177 were excluded on the basis of reporting depressive symptoms (GDS score ≥6; n = 160) or on the basis of being diagnosed with depression and receiving antidepressants at baseline (n = 17) (Figure 1). Of the remaining participants, 863 (92.7%) participated in a follow-up assessment (median, 24 months) and completed the GDS questionnaire at that time.

Figure 1.

A flow chart of the inclusion of participants, HEIJO-KYO cohort, Nara, Japan, 2010–2014. Reasons for exclusion are stated on the right. GDS, geriatric depression scale; LAN, light at night.

Overall, the mean age of the participants was 71.5 (standard deviation, 7.0) years, and 413 (47.9%) participants were men. At baseline, compared with the “dark” group (LAN average of <5 lux, n = 710), the LAN group (LAN average of ≥5 lux, n = 153) showed significantly earlier bedtime and later rising time, resulting in longer duration in bed (Table 1). Older age, current smoking, and sleep disturbances were marginally, but not significantly, associated with higher LAN exposure.

Table 1.

Baseline Characteristics Stratified by Bedroom Light Levels (n = 863), HEIJO-KYO Cohort, Nara, Japan, 2010–2014

Characteristic	Bedroom Light Intensity						P Value
	LAN Average of ≥5 lux (n = 153)			LAN Average of <5 lux (n = 710)
	Mean (SD)	No. of Participants	%	Mean (SD)	No. of Participants	%
Basic parameters
Age, years	72.5 (7.2)			71.3 (6.9)			0.053
Male sex		75	49.0		338	47.6	0.75
Body mass indexa	23.4 (2.8)			23.1 (3.1)			0.23
Current smoker		11	7.2		26	3.7	0.051
Alcohol consumption (≥30 g/day)		19	12.4		105	14.8	0.45
Education (≥13 years)		43	28.1		197	27.7	0.93
Annual household income, ≥4 million JPYb		62	45.3		293	44.2	0.82
Clinical parameters
Hypertension		67	43.8		296	41.7	0.63
Diabetes		19	12.5		81	11.6	0.74
Sleep disturbances		55	36.2		202	28.5	0.06
Bedtime, clock time	22:13 (1:22)			22:36 (1:05)			0.002
Rising time, clock time	6:56 (1:03)			6:44 (0:52)			0.025
Duration in bed, minutes	523.0 (81.5)			488.3 (70.5)			<0.001
Day length, minutes	654 (616–681)c			655 (624–684)c			0.75
Physical activity, log MET-hours/week	3.1 (1.5)			2.9 (1.4)			0.23

Abbreviations: JPY, Japanese yen; LAN, light at night; MET, metabolic equivalent of task; SD, standard deviation.

^a Weight (kg)/height (m)².

^b Approximately $35,000.

^b Expressed as median values (interquartile ranges).

The median LAN average for all participants was 0.6 lux (interquartile range, 0.1–3.0), and 153 (17.7%) and 93 (10.8%) participants were exposed to LAN averages of ≥5 lux and 10 lux, respectively (Table 2). The median light intensities in the groups with LAN averages of ≥5 and <5 lux were 12.4 and 0.4 lux, respectively, and with LAN averages of ≥10 and <10 lux were 19.4 and 0.5 lux, respectively. The median LAN 5 and LAN 10 exposures were 7.0 (interquartile range, 0.5–29.5) minutes and 5.0 (interquartile range, 0–23.5) minutes, respectively, and 108 (12.5%) and 96 (11.1%) participants were exposed to LAN 5 and LAN 10 for ≥60 minutes, respectively. The median LAN 5 exposure in groups with LAN 5 ≥60 and <60 minutes were 97.5 minutes and 0.4 minutes, respectively, and the median LAN 10 exposure in groups with LAN 10 ≥60 and <60 minutes were 98.3 minutes and 0.5 minutes, respectively.

Table 2.

Baseline Light-at-Night Parameters (n = 863), HEIJO-KYO Cohort, Nara, Japan, 2010–2014

LAN Intensity and Duration	No. of Participants	%	Median (IQR)
LAN intensity, lux
Average at all intensities	863	100	0.6 (0.1–3.0)
Average ≥5	153	17.7	12.4 (7.4–22.8)
Average <5	710	82.3	0.4 (0.0–1.4)
Average ≥10	93	10.8	19.4 (13.8–31.2)
Average <10	770	89.2	0.5 (0.0–1.9)
LAN duration, minutes
5 lux	863	100	7.0 (0.5–29.5)
≥60 minutes	108	12.5	97.5 (74.8–147.9)
<60 minutes	755	87.5	0.4 (0.0–1.7)
10 lux	863	100	5.0 (0.0–23.5)
≥60 minutes	96	11.1	98.3 (72.6–148.4)
<60 minutes	767	88.9	0.5 (0.0–1.8)

Abbreviations: IQR, interquartile range; LAN, light at night.

During follow-up, 73 (8.5%) participants reported development of depressive symptoms (GDS score ≥6). Longitudinal analysis using the Cox proportional hazards models (Table 3) showed that the LAN group exhibited a significantly higher depression risk than the “dark” group (hazard ratio (HR) = 1.78, 95% confidence interval (CI): 1.07, 2.96 and P = 0.026; crude model). The significance remained after adjustment for age and sex (model 1); age, sex, and basic parameters (BMI and household income) (HR = 1.89, 95% CI: 1.13, 3.14 and P = 0.015; model 2); and age, sex, and clinical parameters (hypertension, diabetes, sleep disturbances, bedtime, and duration in bed) (HR = 1.72, 95% CI: 1.03, 2.89 and P = 0.039; model 3). In addition, the significance of the association of LAN exposure with incident depressive symptoms remained after adjustment for morning light exposure (median of mean intensity, 366.8 lux; interquartile range, 132.2–854.2) and evening light exposure (median of mean intensity, 25.0 lux; interquartile range, 16.5–37.9) (HR = 1.90, 95% CI: 1.14, 3.19 and P = 0.015; model 2). Furthermore, the adjustment for baseline mild depressive symptoms (GDS scores of 3–5, 33.4%), nocturia (void frequency ≥2 per night, 27.6%), cognitive impairment (Mini-Mental State Examination score of ≤23, 3.5%), or previous cataract surgery (16.0%) did not change the association of LAN exposure with incident depressive symptoms (data not shown; model 2).

Table 3.

Baseline Light-at-Night Exposure and the Risk of Depressive Symptoms (n = 863), HEIJO-KYO Cohort, Nara, Japan, 2010–2015

Comparison	Bedroom Light Intensity						P Value
	LAN Average of ≥5 lux (n = 153)			LAN Average of <5 lux (n = 710)
	No. of Participants	HR	95% CI	No. of Participants	HR	95% CI
No. with incident depressive symptoms	21			52
Follow-up duration, months		24 (17–35)a			23 (17–35)a
Crude model		1.78	1.07, 2.96		1.00	Referent	0.026
Model 1b		1.80	1.08, 2.98		1.00	Referent	0.024
Model 2c		1.89	1.13, 3.14		1.00	Referent	0.015
Model 3d		1.72	1.03, 2.89		1.00	Referent	0.039

Abbreviations: CI, confidence interval; HR, hazard ratio; LAN, light at night.

^a Expressed as median values (interquartile ranges).

^b Model 1 adjusted for age and sex.

^c Model 2 adjusted for age, sex, and basic parameters associated with incident depressive symptoms (body mass index and household income).

^d Model 3 adjusted for age, sex, and clinical parameters associated with depressed mood (hypertension, diabetes, sleep disturbances, bedtime, and duration in bed).

These results were consistent in the sensitivity analysis using a cutoff value of ≥10 lux (HR = 1.93, 95% CI: 1.03, 3.59 and P = 0.039; model 3, Table 4). Further, longer LAN 5 and LAN 10 exposures were consistently associated with incident depressive symptoms (for LAN 5 ≥60 minutes, HR = 2.83, 95% CI: 1.61, 4.97 and P < 0.001; for LAN 10 ≥60 minutes, HR = 2.45, 95% CI: 1.35, 4.46 and P = 0.003; model 3). However, no significant association trends were observed between quartiles in LAN average or LAN 5 exposures and incident depressive symptoms (P for trend = 0.50 and 0.29, respectively; model 1). The other sensitivity analysis using light data between actigraphic sleep onset and sleep offset showed a strengthened association of higher LAN average exposure with incident depressive symptoms (for LAN average of ≥5 lux, HR = 2.47, 95% CI: 1.37, 4.46 and P = 0.003; for LAN average of ≥10 lux, HR = 2.88, 95% CI: 1.42, 5.87 and P = 0.004; model 2).

Table 4.

Baseline Light-at-Night Exposure With Different Thresholds and the Risk of Depressive Symptoms (n = 863), HEIJO-KYO Cohort, Nara, Japan, 2010–2015

LAN Exposure	HR	95% CI	P Value
Crude model
LAN average of ≥10 lux vs. <10 lux	1.87	1.03, 3.41	0.041
LAN 5 lux for ≥60 vs. <60 minutes	2.68	1.57, 4.59	<0.001
LAN 10 lux for ≥60 vs. <60 minutes	2.41	1.36, 4.28	0.003
Model 1a
LAN average of ≥10 lux vs. <10 lux	2.04	1.12, 3.74	0.021
LAN 5 lux for ≥60 vs. <60 minutes	2.76	1.61, 4.74	<0.001
LAN 10 lux for ≥60 vs. <60 minutes	2.48	1.39, 4.40	0.002
Model 2b
LAN average of ≥10 lux vs. <10 lux	2.13	1.16, 3.90	0.015
LAN 5 lux for ≥60 vs. <60 minutes	2.91	1.69, 5.00	<0.001
LAN 10 lux for ≥60 vs. <60 minutes	2.59	1.45, 4.62	0.001
Model 3c
LAN average of ≥10 lux vs. <10 lux	1.93	1.03, 3.59	0.039
LAN 5 lux for ≥60 vs. <60 minutes	2.83	1.61, 4.97	<0.001
LAN 10 lux for ≥60 vs. <60 minutes	2.45	1.35, 4.46	0.003

Abbreviations: HR, hazard ratio; CI, confidence interval; LAN, light at night.

^a Model 1 adjusted for age and sex.

^b Model 2 adjusted for age, sex, and basic parameters associated with incident depressive symptoms (body mass index and household income).

^c Model 3 adjusted for age, sex, and clinical parameters associated with depressed mood (hypertension, diabetes, sleep disturbances, bedtime, and duration in bed).

Time-dependent changes in mean bedroom light intensities at night between the groups with and without incident depressive symptoms are shown in Figure 2A. The incident group was exposed to a higher average of bedroom light intensity throughout the night at baseline than the nonincident group. The mean difference in average bedroom light intensity between the 2 groups was significant and greater in the first half of the night, particularly in the first hour, than that in the second half (Figure 2B).

Figure 2.

Time-dependent changes in bedroom light between groups with (n = 73) and without (n = 790) incident depressive symptoms, HEIJO-KYO cohort, Nara, Japan, 2010–2014. A) Time-dependent changes in mean bedroom light intensities at night. Solid line, incident group; dotted line, nonincident group. B) Time-dependent changes in mean differences in bedroom light intensities at night between the groups with and without incident depressive symptoms. Solid line, mean difference in light intensity at night; dotted line, 95% confidence interval.

DISCUSSION

The present study identified LAN exposure in home settings as an independent risk factor for depressive symptoms in a general elderly population. To our knowledge, this is the first longitudinal evidence for the association between LAN exposure and depressive symptoms in humans. The observational method used here is considered the best study design to evaluate the long-term effects of LAN exposure on mood because an interventional method, such as a randomized controlled trial using a LAN exposure, is not possible because of ethical concerns. Other strengths of this study included the large sample size, use of objective measurement of LAN exposure, and an important sensitivity analysis using different LAN parameters or bedroom light data gathered during the actual sleep period defined by actigraphy.

Our observations add new information to the current understanding of depression risk—that bedroom LAN exposure in home settings, including low average light intensity, is significantly associated with depression risk. In our study, the median LAN intensities in the groups with LAN averages of ≥5 and ≥10 lux were 12.4 and 19.4 lux, respectively, although light intensity reaching the retina through closed eyelids would be substantially lower than bedroom light levels. However, LAN groups can be exposed to higher LAN intensity than the intensity mentioned above, because the LAN intensity was the average bedroom light intensity during the in-bed period. In addition, some participants might open their eyes after sleep onset but while still in bed or in the course of nocturnal voiding. In a previous study in humans, it was shown that LAN exposure with high intensity but short duration could affect human circadian physiology (20). Furthermore, a previous study suggested that biological rhythmicity is affected by minimal LAN exposure (3 lux) (3), and recent human studies have shown that LAN of 5–10 lux during sleep might affect sleep physiology and daytime brain function (21, 22). These studies support our findings of an effect of bedroom LAN exposure on mood. Although the possibility remains that LAN exposure was an initial sign of depressed mood, the adjustment for baseline mild depressive symptoms (GDS scores of 3–5) did not change the association of LAN exposure with incident depressive symptoms. Further studies related to the effects of low LAN intensity on mood are needed.

Although mechanisms underlying the association between LAN exposure and depressive symptoms remain uncertain, previous studies have suggested the possibility that LAN induces sleep disturbances, impaired melatonin section, and circadian misalignment between sleep/wake behavior and internal biological rhythm (7–9), and depression is frequently accompanied by these conditions. In a human study, LAN exposure was found to exert a dose-dependent alerting effect as assessed by subjective and objective ratings (23). Also, it is well established that LAN suppresses melatonin secretion, which may occur at low light intensity (3). We previously reported that LAN exposure in home settings (bedroom horizontal levels) is significantly associated with poor sleep quality but not with the amount of nocturnal melatonin secretion (24, 25). In the present study, the association between LAN exposure and depression risk was independent of sleep disturbances. Further, the mean difference in LAN exposure intensity between the groups with and without incident depressive symptoms was greater 2 hours after bedtime than at other periods, which is consistent with the phase-response curve to light (26), that LAN exposure around bedtime has the greatest effect on the circadian phase delay. On the other hand, LAN exposure might directly induce depressive symptoms, rather than indirectly via sleep disturbances, melatonin suppression, and circadian misalignment. A recent study suggested that aberrant light cycles increase depression-like behaviors in mice despite normal circadian and sleep structures (27). Furthermore, our previous report included associations of LAN exposure with other depression-associated parameters, such as obesity and blood pressure (28, 29); however, the association between LAN exposure and depression risk was also independent of BMI and hypertension.

Our results suggested a potential threshold effect of LAN on mood. We explored several LAN exposure parameters with different thresholds based on previous knowledge. The results of this analysis revealed that LAN duration above the thresholds of 5 and 10 lux are more strongly associated with an increased depression risk than the average parameters. For instance, the number of participants exposed to LAN 10 lux for ≥60 minutes and LAN average of ≥10 lux was comparable (11.1 vs. 10.8%, respectively), but depression risk was higher in relation to LAN 10 lux for ≥60 minutes than LAN average of ≥10 lux (HR = 2.45 vs. 1.93, respectively; Table 4). Our results indicated that greater LAN exposure (top 10.8%–17.7%) was associated with a greater depression risk, although there were no significant dose-response association trends between LAN exposure quartiles and depression risk. Hence, further studies are warranted exploring the optimal threshold for LAN exposure in relation to depression risk.

The present study highlights the need for further research to investigate the association between LAN exposure and depressive symptoms in younger populations—LAN may have greater effects in younger individuals than the elderly. The present cohort included only elderly participants; therefore, the generalizability of our findings to a younger population is uncertain. Age-related cloudiness of the crystalline lens causes decreased light reception to the retina, even before cataract diagnosis, and the capacity for light reception of a 70-year-old is one-fifth of that of a teenager (30). Further, an age-related decline in suprachiasmatic nucleus function has been reported (31). Together, these findings imply that younger individuals might be more sensitive to LAN than are the elderly.

Our study has several potential limitations. First, participants were not randomly selected, possibly leading to selection bias. However, the parameters of BMI and estimated glomerular filtration rate were similar to corresponding national data in Japan, and the follow-up rate was sufficiently high. Second, bedroom light intensity was measured without eye-level light meters, and sleep/awake status was measured only over 2 days. In addition, some measures included data on weekends, and these possibly led to misclassification of LAN status, although our previous studies reported moderate day-to-day reproducibility of LAN exposure and bedtime (25, 32). Future studies measuring eye-level LAN exposure and measuring for multiple nights would reveal more appropriate associations. In addition, wavelength measurements would be important because shorter wavelengths have the greatest impact on circadian physiology (30).

In conclusion, the present study identified LAN exposure in home settings as an independent risk factor for depressive symptoms in an elderly general population. Maintaining darkness in the bedroom at night might be a novel and viable option to prevent depression. Interventional studies reducing LAN exposure are warranted.

ACKNOWLEDGMENTS

Author affiliations: Department of Community Health and Epidemiology, Nara Medical University School of Medicine, Nara, Japan (Kenji Obayashi, Keigo Saeki, Norio Kurumatani).

This work was supported by research funding from the Department of Indoor Environmental Medicine, Nara Medical University; JSPS KAKENHI (grants 24790774, 22790567, 25860447, 25461393, 15H04776, and 15H04777); Mitsui Sumitomo Insurance Welfare Foundation; Meiji Yasuda Life Foundation of Health and Welfare; Osaka Gas Group Welfare Foundation; Japan Diabetes Foundation; Daiwa Securities Health Foundation; Japan Science and Technology Agency; YKK AP Inc.; Ushio Inc.; Nara Prefecture Health Promotion Foundation; Nara Medical University Grant-in-Aid for Collaborative Research Projects; Tokyo Electric Power Company; EnviroLife Research Institute Co., Ltd.; and Sekisui Chemical Co., Ltd.

We thank Sachiko Uemura, Naomi Takenaka, and Keiko Nakajima for their valuable help with data collection.

Parts of this work were presented at 26th Annual Scientific Meeting of the Japan Epidemiological Association, Tottori, Japan, January 21–23, 2016.

Conflict of interest: K.O. and K.S. received research grant support from YKK AP Inc., Tokyo Electric Power Company, EnviroLife Research Institute Co. Ltd., and Sekisui Chemical Co. Ltd. N.K. reported no conflicts.

Abbreviations

 
• BMI

  body mass index

 
• CI

  confidence interval

 
• GDS

  Geriatric Depression Scale

 
• LAN

  light at night

 
• HR

  hazard ratio