Association between irregular social rhythm and all-cause mortality, death by illness, suicide and accidental death: a 19-year community-based cohort study in Japan Authors: Megumi Hazumi, Zui C Narita, Naoaki Kuroda, Mayumi Kataoka, Hiroaki Hori, Yoshinori Cho, Mariko Hanafusa, Rieko Kanehara, Taiki Yamaji, Motoki Iwasaki, Manami Inoue, Shoichiro Tsugane, Norie Sawada Published: 2026 ABSTRACT AbstractIntroductionAlthough irregular social rhythms have been suggested to increase the likelihood of all-cause mortality and cause-specific mortality, their associations have not been investigated sufficiently. This study examined whether irregular social rhythms are associated with all-cause mortality, death by illness, suicide and accidental death.MethodsWe used data from the Japan Public Health Center-based prospective Study (JPHC Study), a community-based cohort of Japanese adults aged 45–74 years. Social rhythm regularity was assessed at baseline via a self-reported item categorising as regular or irregular. Participants were followed for 19 years, and all-cause mortality and cause-specific mortality, including death by illness, suicide and accidental death, were ascertained using national vital statistics. Risk differences (RDs) and risk ratios (RRs) at 5, 10, 15 and 19 years were estimated.ResultsAmong 98 349 participants included in the main analyses, those with irregular social rhythms showed a consistently increased risk of all-cause mortality at 5, 10, 15 and 19 years (19-year RD 2.12%, RR 1.11, 95% CI 1.07 to 1.15). Irregular social rhythms were also associated with risk of death by illness (19-year RD 1.85%, RR 1.11, 95% CI 1.07 to 1.15) and accidental death (19-year RD 0.25%, RR 1.29, 95% CI 1.06 to 1.53). Suicide risk was significantly higher only at 5 years (RD 0.06%, RR 1.50, 95% CI 1.02 to 2.02), with no significant association thereafter.ConclusionsIrregular social rhythms may serve as a long-term marker for all-cause mortality and accidental and illness-related mortality, and a possible short-term marker for suicide; however, the association remains uncertain due to the small number of suicide deaths and the null finding in the sensitivity analysis. Social rhythm regularity may be a useful behavioural marker for identifying groups at elevated mortality. WHAT IS ALREADY KNOWN ON THIS TOPIC Irregular social rhythms—disruptions in the regular timing of daily activities such as sleep and meals—have been linked to poor physical and mental health and higher mortality among shift workers. However, evidence directly examining their association with all-cause and cause-specific mortality in the general population remains limited. Irregular social rhythms—disruptions in the regular timing of daily activities such as sleep and meals—have been linked to poor physical and mental health and higher mortality among shift workers. However, evidence directly examining their association with all-cause and cause-specific mortality in the general population remains limited. WHAT THIS STUDY ADDS Using data from over 100 000 Japanese middle-aged and older adults followed for 19 years, this study found that irregular social rhythms were associated with persistently higher all-cause mortality, illness-related death and accidental death, and higher suicide mortality at 5 years. These associations remained even after adjusting for numerous lifestyle and health covariates. Using data from over 100 000 Japanese middle-aged and older adults followed for 19 years, this study found that irregular social rhythms were associated with persistently higher all-cause mortality, illness-related death and accidental death, and higher suicide mortality at 5 years. These associations remained even after adjusting for numerous lifestyle and health covariates. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY Social rhythm regularity may represent a modifiable behavioural marker for mortality. Further studies using more detailed and objective assessments are needed to clarify the public health relevance of social rhythm regularity. Social rhythm regularity may represent a modifiable behavioural marker for mortality. Further studies using more detailed and objective assessments are needed to clarify the public health relevance of social rhythm regularity. Introduction Social rhythm regularity has emerged as an important public health issue. Social rhythm refers to the regular timing of daily activities such as sleeping, eating and interacting with others. Irregular social rhythm is common among certain subpopulations, including shift workers, individuals experiencing changes in their living environments1 and those with unstable employment.2 A recent survey of the Japanese general population found that 26.0% of men and 21.7% of women report having an irregular social rhythm.3 These daily activities act as ‘social zeitgebers’ or time cues that help synchronise the body’s internal clock.45 According to the social zeitgeber theory, when these cues provided by social rhythm regularity are disrupted, sleep–wake rhythms become unstable.45 Consequently, social rhythm regularity is associated with not only mental and physical health,6–8 but also cognitive functioning.9 In this context, self-reported regularity of daily routines and activity schedules may be regarded as a broad proxy indicator of social rhythm regularity, although it does not capture the full complexity of social rhythms. Recent research increasingly suggests that an irregular social rhythm may be associated with mortality through multiple pathways, involving a broad range of causes of death. A large-scale cohort study using objective measures such as actigraphy has shown that individuals with more stable daily rhythms tend to have lower all-cause mortality rates.10 Consistent with these findings, shift workers, who are one of the populations experiencing irregular social rhythms, have been reported to have significantly higher all-cause mortality but also from specific illnesses such as cardiovascular disease and cancer.11 Data from the UK Biobank further indicate that irregular work patterns are associated with increased mortality from cancer, cardiovascular, respiratory and digestive diseases, underscoring the potential relevance of social rhythm disruption across multiple systems.12 Beyond death by illness, evidence also points to a link between shift work and higher suicide mortality,13 highlighting that social rhythm disruption may be linked to vulnerability to both physical and mental health outcomes. Despite accumulating evidence, comprehensive studies directly exploring the association between social rhythm regularity and cause-specific mortality, including all-cause mortality, death by illness, suicide and accidental death, remain scarce. For example, given the associations between irregular social rhythm and accelerated biological ageing, glycaemic and metabolic control, renal function and immune response,14–17 irregular social rhythm may be linked to death by various illnesses in addition to those mentioned above. In addition to physical health outcomes, irregular social rhythms beyond occupational shift work may also be linked to suicide, as suggested by the elevated suicidal ideation, planning and attempts among individuals with irregular social rhythms such as social jet lag.18–20 Moreover, in occupational contexts, shift work has been associated with accidents; previous studies have reported that shift workers are more likely to be involved in motor vehicle crashes or make medical errors due to disrupted social rhythms.2122 These findings suggest that irregular social rhythm may also be relevant to accidental deaths. Given the wide-ranging links between social rhythm irregularity and various physical and mental health, it is plausible that these pathways may be linked to multiple causes of death in the general population. Therefore, the present study aimed to examine the association between social rhythm regularity and cause-specific mortality, including all-cause mortality, death by illness, suicide and accidental death, in a large general population cohort. By evaluating these outcomes within the same population, this study seeks to clarify whether social rhythm irregularity is broadly linked to multiple causes of death or shows varying patterns. The findings may provide valuable evidence to inform future public health research on social rhythm regularity and premature mortality from both illness and external causes. Methods Participants and setting This study was conducted as part of the Japan Public Health Center-based prospective study (JPHC Study),23 a large-scale population-based cohort study designed to investigate associations between social rhythm factors and various health outcomes. The baseline survey was initiated between 1995 and 1998 across 11 public health centre areas in Japan, with the specific start year varying by region. The target population was resident Japanese nationals aged 45–64 years (Cohort I) and 45–74 years (Cohort II), who were identified from population registries in each area and received self-administered questionnaires. Cohort II was established 3 years after the initiation of Cohort I, with data collection beginning at that time. Self-administered questionnaires were distributed to eligible residents, and returning the completed questionnaire was regarded as provision of informed consent. Among respondents, we excluded participants who had duplicate responses, refused follow-up or additional contact, had a foreign nationality, had inconsistent dates for questionnaire responses or residential relocation or were outside the target age range. Participants with missing information on social rhythm regularity were excluded from the main analyses, although their baseline characteristics are shown separately in table 1. Table 1Participants’ characteristics at baselineSocial rhythm regularityOverallRegularIrregularMissingCovariates(n=103 461)(n=75 155, 72.6%)(n=23 194, 22.4%)(n=5112, 4.9%)Age (years), mean (SD)57.02 (7.97)57.27 (7.99)55.59 (7.61)59.81 (8.12)Sex, n (%) Female55 100 (53.3)41 206 (54.8)11 126 (48.0)2768 (54.1) Male48 361 (46.7)33 949 (45.2)12 068 (52.0)2344 (45.9)Body mass index≥25, n (%) Yes29 112 (28.6)20 093 (27.0)7594 (33.3)1425 (30.4) No72 699 (71.4)54 196 (73.0)15 245 (66.7)3258 (69.6) Missing1650 (1.6)866 (1.2)355 (1.5)429 (8.4)Smoking status, n (%) Less than once a week72 794 (70.4)55 111 (73.3)14 656 (63.2)3027 (59.2) Once a week or more24 082 (23.3)15 946 (21.2)7078 (30.5)1058 (20.7) Missing6585 (6.4)4098 (5.5)1460 (6.3)1027 (20.1) Alcohol consumption, g/day, mean (SD)15.47 (29.7)14.14 (27.2)20.24 (36.6)13.34 (27.7) Missing, n (%)3794 (3.7)2341 (3.1)785 (3.4)668 (13.1)Physical activity, n (%) Less than once a week76 301 (73.7)55 686 (74.1)18 380 (79.2)2235 (43.7) Once a week or more21 483 (20.8)16 954 (22.6)3996 (17.2)533 (10.4) Missing5677 (5.5)2515 (3.3)818 (3.5)2344 (45.9)Living arrangement, n (%) Living alone5497 (5.3)3521 (4.7)1743 (7.5)233 (4.6) Living with someone97 964 (94.7)71 634 (95.3)21 451 (92.5)4879 (95.4)Employment, n (%) Employed93 357 (90.2)67 648 (90.0)21 038 (90.7)4671 (91.4) Unemployed or houseworker10 104 (9.8)7507 (10.0)2156 (9.3)441 (8.6)Sleep duration, n (%) <7 hours28 498 (27.5)18 175 (24.2)9463 (40.8)860 (16.8) 7–9 hours64 389 (62.2)50 816 (67.6)12 122 (52.3)1451 (28.4) >9 hours7537 (7.3)5847 (7.8)1474 (6.4)216 (4.2) Missing3037 (2.9)317 (0.4)135 (0.6)2585 (50.6)Perceived stress, n (%) Low19 522 (18.9)16 400 (21.8)2733 (11.8)389 (7.6) Normal62 252 (60.2)47 550 (63.3)13 032 (56.2)1670 (32.7) High18 303 (17.7)10 725 (14.3)7240 (31.2)338 (6.6) Missing3384 (3.3)480 (0.6)189 (0.8)2715 (53.1)Number of physical illnesses, n (%)078 620 (76.0)57 635 (76.7)16 955 (73.1)4030 (78.8)119 717 (19.1)13 979 (18.6)4852 (20.9)886 (17.3)≥ 25124 (5.0)3541 (4.7)1387 (6.0)196 (3.8)History of physical illness includes stroke, myocardial infarction, diabetes, gastric or duodenal ulcer, gallstones, gastric cancer, lung cancer, colorectal cancer, liver cancer, breast cancer, uterine cancer or other cancers.SD, standard deviation. Participants’ characteristics at baseline History of physical illness includes stroke, myocardial infarction, diabetes, gastric or duodenal ulcer, gallstones, gastric cancer, lung cancer, colorectal cancer, liver cancer, breast cancer, uterine cancer or other cancers. SD, standard deviation. This study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement. Measurements The vital status was consistently recorded over a follow-up period of up to 19 years from the baseline survey, which was conducted between 1995 and 1998. The cause of death was ascertained using death certificates obtained from the Ministry of Health, Labour and Welfare, with permission,24 and classified according to ICD-10 codes.25 The outcomes were all-cause mortality (all ICD-10 codes), accidental death (ICD-10 codes V01–X59), suicide (ICD-10 codes X60–X84), death by illness (ICD-10 codes A00-R99, excluding deaths with accompanying external cause codes (V01–Y98)), death by cancer (ICD-10 codes C00–C97) and death by cardiovascular (CVD) disease (ICD-10 codes I00–I97). Social rhythm regularity was assessed as a practical proxy using the question, “Is your daily routine or activity schedule regular?” in the questionnaire used in the previous study, with the options ‘yes’ or ‘no’.3 Participants who responded with ‘no’ were considered to have an irregular rhythm; participants who responded with ‘yes’ were included as the reference group. Information on covariates collected at baseline included age (years),2627 sex (female, male),2829 body mass index (BMI≥25: yes, no),3031 current smoking status (yes, no),3233 alcohol consumption (mean grams/day),3435 which was calculated based on the question regarding frequency (rarely, 1–3 times per month, 1–2 times per week, 3–4 times per week, 5–6 times per week, everyday) and amount of consumption per day (question regarding type of alcohol and number of glasses per day),35 habitual exercise (less than once a week, once a week or more),3637 living arrangement (living alone, living with someone),3839 employment status (employed, unemployed or houseworker),240 sleep duration (<7 hours, 7–9 hours, >9 hours),4142 perceived stress (low, normal, high),4344 number of physical illness (0, 1, ≥2),45–47 defined as the presence of any of the following conditions: stroke, myocardial infarction, diabetes, gastric or duodenal ulcer, gallstones, gastric cancer, lung cancer, colorectal cancer, liver cancer, breast cancer, uterine cancer or other cancers and region (categorical, Ninohe, Yokote, Saku, Chubu, Katsushika, Mito, Nagaoka, Chuo-higashi, Kamigoto, Miyako and Suita).4849 Analyses Statistical analyses were conducted to examine the association between social rhythm regularity at baseline and the risk of all-cause mortality, death by illness, death by CVD, death by cancer, suicide and accidental death during the follow-up period. All analyses were performed using R (V.4.3.1). The analytical approach was based on methods outlined in a previous study using the same dataset.39 To account for potential bias due to non-random loss to follow-up and confounding by baseline characteristics, stabilised inverse probability weights were constructed as the product of two components: (1) weights from a censoring model predicting the probability of remaining uncensored at each follow-up year, and (2) weights from an exposure model estimating the probability of having an irregular versus regular social rhythm. For the censoring weights, the numerator model included follow-up time, squared follow-up time and social rhythm regularity, while the denominator model additionally included all covariates listed above. The same censoring model specification was used across mortality outcomes, with the outcome definition changed for each analysis. The final weights were applied throughout the pooled logistic regression. As model diagnostics for the IPW models, we examined the distribution of the final stabilised weights by calculating the mean, SD, minimum and maximum values. Censoring weights and exposure weights were truncated at the 99th percentile. Inverse probability weighted pooled logistic regression was employed to estimate and visualise the cumulative incidence of mortality over the follow-up period, examining the association with social rhythm regularity.50 Follow-up time was divided into yearly intervals to create person-period data for the pooled logistic regression. Cumulative risks were then estimated and summarised at 5, 10, 15 and 19 years. Social rhythm regularity was defined as a dichotomous exposure variable, comparing individuals with an ‘Irregular’ rhythm to those with a ‘Regular’ rhythm, who served as the reference group. The model included a product term between the exposure and the time variables. Separate analyses were performed for all-cause mortality, death by illness, death by CVD, death by cancer, suicide and accidental death generating cumulative incidence curves plotted against the number of years to illustrate the risk progression for each exposure group over time. For analyses of all-cause mortality, only loss to follow-up was treated as censoring. In contrast, for cause-specific mortality analyses, deaths from other causes were additionally treated as censoring events, and the censoring model for the IPW was constructed accordingly. We analysed mortality outcomes across different follow-up durations of 5, 10, 15 and 19 years. For each outcome and time point, both the absolute risk difference (RD) and the rerative risk (RR) were calculated to compare the mortality risk in the irregular social rhythm group relative to the regular social rhythm group. To estimate the precision of these measures, bootstrap CIs were computed for both the RD and RR estimates. The number of deaths within each category (regular and irregular) for each cause and time point was also tabulated. Finally, to assess the robustness of the observed associations to potential unmeasured confounding, E-values were calculated for the RR estimates comparing irregular versus regular social rhythm for each mortality outcome (all-cause mortality, death by illness, death by CVD, death by cancer, suicide and accidental death) at the 5-year, 10-year, 15-year and 19-year follow-up points.51 As a complementary sensitivity analysis, these E-values were used to refine the interpretation of the observed RRs. These E-values were computed for both the point estimate of the RR and the lower limit of its CI to quantify the minimum strength of association an unmeasured confounder would need with both the exposure and the outcome to explain away the observed RR. We also conducted stratified analyses by sex, age group (<65 and ≥65 years) and employment status to examine whether the associations differed across these subgroups. For each stratum, RDs, RRs and E-values at 5, 10, 15 and 19 years were estimated using the same analytical approach. Missing data in covariates were imputed using random forest imputation.52 In the main analysis, social rhythm regularity and follow-up years were not imputed. As a sensitivity analysis, we additionally imputed missing values of social rhythm regularity together with baseline covariates using random forest imputation and repeated the main analyses. The imputation was performed using the missRanger package in R with 100 trees and predictive mean matching using three candidate donors. Results Participants characteristics Of the original 103 880 participants who answered the survey in 1995–1998 in the JPHC study, 419 individuals were excluded, resulting in 103 461 eligible participants (figure 1). Of these, 5112 participants had missing information on social rhythm regularity and were excluded from the main analyses comparing the regular and irregular rhythm groups. The main analyses therefore included 98 349 participants. Flow chart highlight. The mean age of the participants was 57.02 years (SD 7.97), and 46.7% (n=48 361) were male, while 53.3% (n=55 100) were female. 75 155 (72.6%) reported having a regular social rhythm, 23 194 (22.4%) reported having an irregular social rhythm and 5112 (4.9%) did not provide information on social rhythm regularity. Participants with irregular social rhythm regularity were more likely to be male, to smoke at least once per week, to consume more alcohol, to engage in physical activity less frequently (less than once per week), to live alone, to have shorter sleep duration, to experience higher stress and to have comorbid conditions compared with those with regular social rhythm. The proportions of missing covariate data are also shown in table 1. The distributions of the final stabilised weights used in the pooled logistic regression models are shown in online supplemental appendices 1 and 2. Across outcomes, the mean weights ranged from 0.99 to 1.00, the SD ranged from 0.28 to 0.30, and the maximum values ranged from 2.29 to 3.44. Cumulative incidence over time Figure 2 illustrates the cumulative incidence curves for different mortality outcomes based on social rhythm regularity. The cumulative incidence of all-cause mortality was consistently higher for individuals with irregular social rhythm compared with those with regular social rhythm during the entire observation period. For death by illness, the cumulative incidence was greater in the irregular social rhythm group compared with the regular social rhythm group across the years, as well as death by cancer and CVD (online supplemental appendix 3). A similar pattern was observed for suicide, where the irregular group showed a higher cumulative incidence over time than the regular group. Likewise, for accidental death, the cumulative incidence was consistently higher among individuals with an irregular social rhythm compared with those with a regular social rhythm throughout the follow-up period The cumulative incidence of deaths. Risk at the points of 5, 10, 15 and 19 years passed Table 2 presents the RDs and RRs comparing irregular versus regular social rhythm for various mortality outcomes at different follow-up times. All-cause mortality risk was significantly higher for the irregular social rhythm group throughout the follow-up: 5 years (RD 0.59%, 95% CI 0.39% to 0.77%; RR 1.26, 95% CI 1.17 to 1.34), 10 years (RD 1.34%, 95% CI 0.92% to 1.70%; RR 1.20, 95% CI 1.14 to 1.26), 15 years (RD 1.93%, 95% CI 1.34% to 2.43%; RR 1.15, 95% CI 1.10 to 1.19) and 19 years (RD 2.12%, 95% CI 1.38% to 2.72%; RR 1.11, 95% CI 1.07 to 1.15). Table 2Absolute and relative risks at 5-year, 10-year, 15-year and 19-year follow-upsBootstrap CIBootstrap CINumber of deathsRD (%)Lower (%)Upper (%)RRLowerUpperAll-cause mortality 5-yearRegular2972 Irregular10970.590.390.771.261.171.34 10-yearRegular6760 Irregular23241.340.921.701.201.141.26 15-yearRegular11 797 Irregular38231.931.342.431.151.101.19 19-yearRegular15 418 Irregular48832.121.382.721.111.071.15Death by illness 5-yearRegular2615 Irregular9310.500.340.691.251.171.35 10-yearRegular5972 Irregular20111.150.821.541.201.141.27 15-yearRegular10 593 Irregular33511.681.242.221.141.111.20 19-yearRegular17 996 Irregular55661.851.272.541.111.071.15Suicide 5-yearRegular137 Irregular650.060.0030.111.501.022.02 10-yearRegular253 Irregular1070.09−0.0170.161.290.951.59 15-yearRegular325 Irregular1260.07−0.070.151.150.851.37 19-yearRegular362 Irregular1340.04−0.110.141.070.801.27Accidental death 5-yearRegular136 Irregular620.060.010.111.491.052.02 10-yearRegular292 Irregular1250.140.030.241.411.101.77 15-yearRegular460 Irregular1830.210.060.351.341.101.61 19-yearRegular559 Irregular2200.250.050.421.291.061.53CI, Confidence Interval; RD, risk difference; RR, relative risk. Absolute and relative risks at 5-year, 10-year, 15-year and 19-year follow-ups CI, Confidence Interval; RD, risk difference; RR, relative risk. For death by illness, a consistently increased risk was found for the irregular group across all time points: 5 years (RD 0.50%, 95% CI 0.34% to 0.69%; RR 1.25, 95% CI 1.17 to 1.35), 10 years (RD 1.15%, 95% CI 0.82% to 1.54%; RR 1.20, 95% CI 1.14 to 1.27), 15 years (RD 1.68%, 95% CI 1.24% to 2.22%; RR 1.20, 95% CI 1.11 to 1.20) and 19 years (RD 1.85%, 95% CI 1.27% to 2.54%; RR 1.11, 95% CI 1.07 to 1.15). The relative risks of death by CVD remained consistently high, whereas the relative risks for death by cancer did not show such differences (Appendix 4). Regarding suicide, an elevated risk was observed at 5 years (RD 0.06%, 95% CI 0.003% to 0.11%; RR 1.50, 95% CI 1.02 to 2.02), but this association attenuated and became non-significant at 10 years (RD 0.09%, 95% CI −0.02% to 0.16%; RR 1.29, 95% CI 0.95 to 1.59), 15 years (RD 0.07%, 95% CI −0.07% to 0.15%; RR 1.15, 95% CI 0.85 to 1.37) and 19 years (RD 0.04%, 95% CI −0.11% to 0.14%; RR 1.07, 95% CI 0.80 to 1.27). For accidental death, individuals with an irregular social rhythm showed an increased risk compared with the regular group at 5 years (RD 0.06%, 95% CI 0.01% to 0.11%; RR 1.49, 95% CI 1.05 to 2.02), 10 years (RD 0.14%, 95% CI 0.03% to 0.24%; RR 1.41, 95% CI 1.10 to 1.77), 15 years (RD 0.21%, 95% CI 0.06% to 0.35%; RR 1.34, 95% CI 1.10 to 1.62) and 19 years (RD 0.25%, 95% CI 0.05% to 0.42%; RR 1.29, 95% CI 1.06 to 1.53). Stratified analyses by sex, age group, employment status and absence of physical comorbidity are shown in online supplemental appendixs 5–9. The associations with all-cause mortality and death by illness were generally in the same direction across subgroups, including among participants without physical comorbidity, whereas those with suicide and accidental death were less consistent and showed wide CIs in several subgroups. The sensitivity analysis using random forest imputation for missing social rhythm regularity yielded estimates similar to those of the main analysis, while the association with suicide at 5 years was slightly attenuated and not statistically significant (online supplemental appendix 10). Robustness to unmeasured confounders Table 3 and online supplemental appendix 4 present the E-values calculated to assess the robustness of the observed relative risks to potential unmeasured confounding. For all-cause mortality, the E-values for the point estimates were 1.83 at 5 years, 1.69 at 10 years, 1.57 at 15 years and 1.46 at 19 years, and the E-values for the CI limits were 1.62, 1.54, 1.43 and 1.34 across the respective time points. For death by illness, the E-values for the point estimates were 1.81 at 5 years, 1.69 at 10 years, 1.69 at 15 years and 1.46 at 19 years, with E-values for the CI limits of 1.62, 1.54, 1.46 and 1.34, respectively. E-values of cancer and CVD ranged from 1.21 to 1.34 and 1.67 to 2.08, respectively. Regarding suicide, the E-value for the point estimate was 2.37 at 5 years, decreasing to 1.90 at 10 years, 1.57 at 15 years and 1.34 at 19 years; the corresponding E-values for the CI limit were 1.16 at 5 years and 1.00 for the 10-year, 15-year and 19-year follow-ups. For accidental death, the E-value for the point estimate was 2.34 at 5 years, 2.17 at 10 years, 2.01 at 15 years and 1.90 at 19 years, while the E-value for the lower limit of the CI was 1.28, 1.43, 1.43 and 1.31, respectively. Table 3E-values sensitivity analysis for all mortality outcomes at 5-year, 10-year, 15-year and 19-year follow-upsPoint estimateLimit of CIAll-cause mortality 5-year1.831.62 10-year1.691.54 15-year1.571.43 19-year1.461.34Death by illness 5-year1.811.62 10-year1.691.54 15-year1.691.46 19-year1.461.34Suicide 5-year2.371.16 10-year1.901.00 15-year1.571.00 19-year1.341.00Accidental death 5-year2.341.28 10-year2.171.43 15-year2.011.43 19-year1.901.31CI, confidence interval. E-values sensitivity analysis for all mortality outcomes at 5-year, 10-year, 15-year and 19-year follow-ups CI, confidence interval. Discussion This study investigated the association between social rhythm regularity and subsequent all-cause mortality, death by illness, CVD and cancer, suicide and accidental death over a 19-year follow-up period. The findings revealed that individuals with an irregular social rhythm had a consistently higher all-cause mortality, death by illness, CVD and accidental death across all follow-up points compared with those with a regular rhythm, despite the fact that as many as 22.4% of participants had an irregular rhythm. In addition, a short-term association with suicide was suggested in the main analysis, although this finding should be interpreted cautiously because the sensitivity analysis yielded an attenuated estimate with a CI including the null. Irregular social rhythms were consistently associated with all-cause mortality across the 19-year follow-up. This finding aligns with prior large-scale cohort studies that have demonstrated links between circadian rhythm disruption and higher all-cause mortality. For instance, the Rotterdam Study showed that fragmentation and instability of daily activity rhythms predicted mortality,10 and meta-analyses of shift work have similarly reported higher all-cause mortality among workers with irregular schedules.1113 The magnitude of relative risk observed in our study is consistent with these previous findings, supporting the robustness of the association in a large Japanese general population cohort. A recent large-scale cohort study reported that night shift work was associated with faster biological ageing and reduced life expectancy, partly mediated by metabolic dysregulation.14 This suggests that irregular rhythms may be linked to a systemic ageing process that gradually increases vulnerability to multiple causes of death, thereby manifesting as elevated overall mortality beyond disease-specific mechanisms. For most mortality outcomes, the RDs tended to increase over time, whereas the RRs tended to decrease. This pattern may be explained by the difference between absolute and relative scales. As cumulative mortality risks increase with longer follow-up, including in the regular rhythm group, absolute differences can accumulate even when RRs attenuate.53 In addition, the attenuation of RRs may indicate non-proportional associations over time, because social rhythm regularity may change after baseline and a single baseline assessment may become less representative of long-term exposure as follow-up becomes longer. Another possible explanation is depletion of susceptible individuals, whereby individuals at higher risk die earlier and the remaining population becomes less vulnerable over time.54 Social rhythm regularity was also associated with death by illness, cardiovascular disease and cancer. These findings are consistent with previous studies reporting associations between irregular social rhythms and death by illness, such as cardiovascular disease and cancer.1011 One plausible explanation is that irregular rhythms may be associated with disrupted circadian regulation of physiological systems, thereby increasing vulnerability to metabolic and cardiovascular conditions.55 The association between irregular social rhythms and all-cause mortality may reflect associations with specific causes of death, including illness, accidents and suicide. Multiple mechanisms could underlie this relationship, such as disruptions in sleep, behavioural dysregulation and long-term psychological or physiological strain, all of which may be associated with mortality. Higher suicide mortality was indicated among individuals with irregular social rhythms at the 5-year follow-up, although this association attenuated and became statistically non-significant over longer durations. This finding is consistent with previous studies suggesting that irregular social rhythm is associated with suicidal ideation and attempts among specific subpopulations, such as adolescents,1819 individuals with mood disorders18 and workers,20 as well as death by suicide among workers.13 The present study examined death by suicide prospectively over nearly two decades in a large, community-based cohort. This design provides a more conservative and temporally nuanced evaluation of the associations and allows for broader generalisability beyond specific subpopulations by using a large community-based sample. One plausible mechanism is that irregular social rhythms may be associated with depressive states and susceptibility to suicidal thoughts and attempts,5657 which are known factors for suicide. The association between irregular social rhythms and suicide was not statistically significant after 10 years, while this finding should be interpreted with caution. The small number of suicide cases may have limited statistical power, increasing the possibility of type II error. Therefore, the non-significant estimates at later follow-up points should not be interpreted as evidence of no long-term association. In addition, earlier mortality in the irregular group may have led to a built-in selection bias, where those with greater vulnerability died earlier, leaving a survivor group with lower vulnerability. These factors could have contributed to an underestimation of the association in the later years. Individuals with irregular social rhythms may be associated with higher accidental death over the long term. Although previous studies have presented the associations between irregular social rhythm and accidents or death by causes other than accidents,101121 few have examined its relevance to accidental death. This study extended prior studies by a modest but persistent increase in accidental death over a 19-year follow-up period among individuals with irregular rhythms. Irregular social rhythm can result in circadian misalignment and sleep disturbances,45 which in turn may impair cognitive and mental regulation.956–58 These impairments are known to increase the vulnerability to accidental death,59–61 due to a lack of situational awareness and adaptive responses to environmental hazards.6263 Notably, the association with accidental death in individuals with irregular social rhythms persisted for up to 19 years. Supporting this long-term association, previous studies have shown that the health consequences of irregular social rhythms, such as shift workers, may continue even after retirement and in later life.864 Although causality cannot be established, these results suggest that irregular social rhythms may function as a long-term marker for accidental death, potentially reflecting long-term differences in mental health and cognitive capacity. In stratified analyses by sex, age group, employment status and absence of physical comorbidity, the associations of irregular social rhythms with all-cause mortality and death by illness were generally consistent across subgroups, which may suggest that these associations were not limited to a specific demographic group. In contrast, the associations with suicide and accidental death were less consistent, with several subgroup-specific estimates showing wide CIs and including the null value. Because the numbers of suicide and accidental death cases were relatively small, these findings should be interpreted with caution. Further studies with larger numbers of suicide and accidental death cases are needed to clarify these associations. In our cohort, over one-fifth of participants reported an irregular social rhythm, suggesting that irregular social rhythm is a common rather than a rare exception. Although we did not directly measure the prevalence of irregular social rhythms in independent cohorts using the same definition, some large-scale studies support the relevance of irregular social rhythms in general populations. Given that irregular social rhythms were associated with several mortality outcomes, social rhythm regularity may be a useful behavioural marker in future epidemiological and intervention studies. Future research should assess social rhythm regularity using more detailed and validated approaches, such as the social rhythm metric, sleep diaries or ecological momentary assessment, which can capture day-to-day variability in sleep, meals, activity and social interaction. Future studies using more detailed and objective assessments are needed to clarify the association between social rhythm regularity and long-term mortality outcomes. Limitations There are some limitations. First, although we adjusted for a comprehensive set of covariates, residual confounding cannot be entirely ruled out due to the observational nature of the study and the assumptions required for IPW.65 In particular, although we adjusted for the number of physical illnesses and conducted stratified analyses among participants without physical comorbidity, underlying diseases may still have influenced both social rhythm regularity and mortality risk. The physical illness covariate did not fully capture illnesses or conditions that could influence social rhythm regularity, such as sleep-wake disorders, preclinical illness, depressive symptoms, social isolation and other chronic conditions. Reverse causation also cannot be completely ruled out, as irregular social rhythms may partly reflect unmeasured or preclinical health conditions at baseline. While the E-values calculated in this study do not eliminate the possibility of unmeasured confounding, they provide a supplementary indication of the robustness of the observed associations. Second, while social rhythm regularity was assessed using a brief self-report proxy measure without established validity, which may not fully capture the complexity of daily routines accurately, the measure has been used in previous epidemiological studies and captures functionally relevant aspects of behavioural regularity.3 Importantly, shift workers were likely included among those reporting irregular routines, but shift work was not explicitly assessed as a separate exposure. Because social rhythm regularity was assessed as a binary exposure at baseline without validation, we could not capture the complexity of daily rhythm regularity or evaluate the duration, changes, or accumulation of irregular social rhythms. This lack of validation may have led to exposure misclassification. Such measurement error would likely attenuate the estimated associations toward the null if it was non-differential with respect to mortality outcomes, although the direction of bias cannot be fully determined. Because social rhythm regularity and sleep duration were measured at the same baseline survey and detailed sleep quality data were not available, we could not formally examine sleep-related mediation pathways66; future studies with repeated assessments are needed. Together, these measurement limitations restrict the specificity of interpretation, although prior evidence shows that both shift work and irregular routines increase mortality.1114 Additionally, although the cohort represents a large, community-based population, the generalisability of these results to younger individuals or populations in different cultural or geographic settings may be limited. Third, in cause-specific mortality analyses, deaths from other causes were treated as censoring events; therefore, the cumulative risk estimates should be interpreted within this cause-specific framework. The number of suicide deaths was relatively small, which may have limited statistical power and contributed to wide CIs. Although missing covariate values were imputed using random forest imputation, several variables, including sleep duration and perceived stress, had relatively high proportions of missing data. As a result, the findings may be sensitive to assumptions about the missingness mechanism. In particular, if missingness was related to unobserved values or deviated from the missing-at-random assumption, the imputation procedure may have introduced bias. Finally, because the present study included middle-aged and older Japanese adults, caution is needed when applying the present findings to younger individuals or other populations. The degree and context of social rhythm regularity may differ across age groups and cultural settings.7 In addition, Japan has distinctive social rhythm contexts, such as long working hours.67 Conclusion This community-based prospective study found that irregular social rhythms were associated with all-cause mortality, death by illness, accidental death over the long term and suicide at the 5-year follow-up. These findings suggest that social rhythm regularity may be a useful behavioural marker for all-cause mortality, as well as death due to illness, suicide and accidents. To further validate these observations, future cohort studies are needed to elucidate the potential mechanisms linking irregular social rhythms to mortality, such as circadian disruption and sleep disturbances. Supplementary material