Abstract
Background
Social participation may protect against dementia, but few studies have examined whether cessation of social participation during the period surrounding retirement is associated with subsequent dementia risk accounting for the competing risk of death. This study examined whether changes in social participation between ages 65 and 70 are associated with incident dementia in community-dwelling Japanese older adults.
Methods
This prospective cohort study analysed 1,681 participants of the NISSIN Project (Nisshin City, Aichi Prefecture, Japan) enrolled at age 65 (1996–2005), re-examined at age 70, and followed for incident dementia through the Japanese National Long-Term Care Insurance scheme until 2023. Social participation was assessed using 10 items at ages 65 and 70. Participants were classified into four groups: Continuation, Cessation, Initiation, and Persistent nonparticipation. Bayesian cause-specific hazard models were used to estimate hazard ratios for incident dementia, treating death without dementia as a competing event. Sensitivity analyses addressed reverse causation and model assumptions.
Results
During 22,116 person-years of follow-up (median 13.6 years), 497 participants developed dementia and 138 died without dementia. In the fully adjusted model, Cessation was associated with a higher hazard of dementia than Continuation (HR 1.27, 95% credible interval [CrI] 1.02 to 1.64; posterior probability of HR > 1, 0.98). This association remained after lag exclusion and adjustment for cognitive function at age 65. Neither Initiation nor Persistent nonparticipation was associated with dementia.
Conclusions
Cessation of social participation between ages 65 and 70 was associated with increased dementia incidence. Recent cessation of social activities may help identify older adults who could benefit from preventive support during the period surrounding retirement. Limitations include a single-suburb sample, which may limit generalisability, and identification of dementia through long-term care insurance certification rather than clinical diagnosis.
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1186/s12877-026-07932-7.
Keywords: Social participation, Dementia, Competing risks, Bayesian analysis, Cohort study
Background
Dementia affects over 55 million people worldwide, and the number of affected individuals is projected to reach 153 million by 2050 [1]. Poor social relationships and low social participation are consistently associated with increased dementia risk [2–4], potentially through reduced cognitive reserve, increased neuroinflammation, and fewer health-promoting behaviours [4, 5]. Greater social participation in midlife and late life has been associated with 30% to 50% lower incidence of dementia [6], and a meta-analysis of 38 longitudinal studies reported that cognitive, physical, and social leisure activities were all associated with reduced incidence of dementia, with cognitive activities showing the strongest association [7]. In a nationally representative English cohort of adults aged 50 and over, community cultural engagement was associated with lower incidence of dementia independently of social factors [8], and in a US cohort of older adults, the predicted age at dementia diagnosis differed by 4.5 years between the least and most socially active tertiles [9]. We previously reported that nonparticipation in social activities was associated with elevated incidence of dementia, particularly among participants with hypertension or diabetes [10]. In a nationwide longitudinal survey of Japanese adults aged 60 and over, retirement was associated with altered patterns of social activity participation [11].
Few studies have examined changes in social participation in relation to dementia. In the Whitehall II cohort, change in frequency of social contact was not significantly associated with subsequent dementia [12], whereas in a Japanese cohort of adults aged 65 and over (n = 47,698), a downward trajectory in social participation was associated with the highest incidence of dementia among all trajectory groups [13]. However, in a UK cohort of middle-aged women (n = 850,000), associations between nonparticipation and dementia attenuated substantially with longer follow-up, raising concerns about reverse causation [14]. Declining social participation before dementia diagnosis has been documented in an English cohort of adults aged 50 and over [15], and a similar temporal attenuation for physical inactivity has been reported in a European multi-cohort study [16]. However, none of these studies accounted for the competing risk of death, which can bias estimates of dementia incidence in ageing cohorts [17], nor did any focus on a specific life transition. In Japan, the statutory retirement age was set at 60 years during the study enrolment period (1996–2005), with legislation progressively extending employment opportunities toward age 65 [18]. Many workers in the study cohort are likely to have experienced a major occupational transition between ages 60 and 65, including mandatory retirement from their primary employer, with accompanying changes in daily routines and social networks. No study has examined whether cessation of social activities during this period independently predicts dementia incidence within such a framework.
The present study examined whether changes in social activity participation between ages 65 and 70 were associated with incident dementia in an age-specific cohort of community-dwelling Japanese older adults. Bayesian cause-specific hazard models were used with progressive covariate adjustment, lag exclusion analyses, and cognitive function adjustment to address reverse causation. Secondary analyses examined activity subtypes separately, and sex-interaction analyses explored potential sex differences.
Methods
Study design and population
The New Integrated Suburban Seniority Investigation (NISSIN) Project is an age-specific prospective cohort study reported in detail elsewhere [19]. Using the basic resident registry of Nisshin City in central Japan, all residents at age 65 years were invited to participate from 1996 to 2005; 3,073 individuals were enrolled (response rate, 43.9%). Participants underwent comprehensive baseline medical examinations and completed a self-report questionnaire at age 65, and were further invited to undergo follow-up examinations at age 70. After exclusions for missing data, prevalent dementia, and loss to follow-up (Fig. 1), 1,681 participants (880 men and 801 women) were analysed in the present study. This study was conducted in accordance with the principles of the Declaration of Helsinki. The study protocol was approved by the ethics committees of the participating institutions, including Hokkaido University Graduate School of Medicine (2014, no. 37), Aichi Medical University School of Medicine (2008, no. 558), National Center for Geriatrics and Gerontology (2006, no. 242), Nagoya University Graduate School of Medicine (2002 and 2004, no. 162), and Osaka City University School of Human Life Science (2021, no. 21 − 01). Written informed consent was obtained from all participants.
Fig. 1.
Flow diagram of participant selection from the NISSIN Project cohort. Median follow-up from the age-70 examination was 13.6 years
Exposure
Social activity participation was assessed using 10 items from a validated instrument [20] (Supplementary Table 1) covering community-oriented (5 items) and interest/learning-oriented (5 items) activities. For each item, participants reported whether they were regularly, occasionally, or not doing the activity. This applied to all items, including the Silver Human Resources Centre, and therefore reflected actual engagement rather than registration alone. Those reporting regular or occasional participation in at least one of the 10 items were classified as participants. Supplementary Table 1 reports the proportion participating in each item at ages 65 and 70. Based on the transition in participation status between ages 65 and 70, four groups were defined: Continuation (participated at both ages), Cessation (participated at age 65 but not at age 70), Initiation (did not participate at age 65 but participated at age 70), and Persistent nonparticipation (did not participate at either age), with Continuation as the reference.
Outcome
Incident dementia was ascertained through the Japanese National Long-Term Care Insurance (LTCI) scheme [21]. Trained investigators assessed cognitive function, activities of daily living, and behavioural disorders during home visits using a standardised government assessment manual. Incident dementia was defined as cognitive disability of grade IIa or higher, corresponding to dementia-related symptoms that limit daily living outside the home. This threshold has demonstrated a sensitivity of 65% and a specificity of 93% for dementia according to DSM-IV criteria [22]. Follow-up began at the age-70 examination as the landmark time point and continued until the date of dementia certification, death, relocation, or December 31, 2023, whichever occurred first. Death without dementia was treated as a competing event.
Statistical analysis
Bayesian Weibull cause-specific hazard models were fitted to estimate hazard ratios for incident dementia, treating death without dementia as a competing event [23]. We chose a cause-specific hazard approach because our primary interest was the aetiological association between changes in social participation and subsequent dementia incidence, rather than prediction of the cumulative incidence in the presence of competing events. Baseline characteristics measured at the age-65 examination were used as covariates. Three covariate sets were specified with progressive adjustment: Model 1 adjusted for sex and enrolment year; Model 2 additionally adjusted for body mass index (< 18.5, 18.5 to 22.9, or ≥ 23 kg/m²), smoking status (never, former, or current), drinking status (never or past vs. current), educational attainment (lower than high school vs. high school or above), marital status (married vs. other), living arrangement (with family vs. alone), work status (not working vs. working), walking time (< 1 vs. ≥1 h per day), and exercise frequency (< 2 vs. ≥2 times per week); and Model 3 further adjusted for competence in daily living assessed by the Tokyo Metropolitan Institute of Gerontology Index of Competence (high vs. lower competence) [24], depressive symptoms assessed by the 15-item Geriatric Depression Scale (presence defined as a score ≥ 6) [25], and the presence of hypertension, diabetes mellitus, cardiovascular disease, and stroke history. Hypertension was defined as systolic blood pressure ≥ 140 mmHg, diastolic blood pressure ≥ 90 mmHg, or the use of antihypertensive medication. Diabetes mellitus was defined as fasting plasma glucose ≥ 126 mg/dL, haemoglobin A1c ≥ 6.5% (National Glycohemoglobin Standardization Program equivalent) [26], or self-reported use of hypoglycaemic medication. Weakly informative Normal(0, 2.5) priors were placed on all regression coefficients, and prior predictive checks confirmed clinically plausible survival time distributions. Each model was fitted with four Markov chain Monte Carlo chains of 6,000 iterations (2,000 warmup) using the No-U-Turn Sampler implemented in Stan via the brms R package [27]. Convergence was assessed using trace plots, R-hat statistics, effective sample size ratios, and posterior predictive checks (Supplementary Fig. 1). Missing covariate data were handled by multiple imputation using chained equations with 20 imputed datasets [28]. The imputation model included all analysis variables, and the exposure variable was not imputed. Posterior distributions were pooled across imputed datasets by combining posterior draws. Results are reported as posterior median hazard ratios with 95% credible intervals and the posterior probability that the hazard ratio exceeds 1. Secondary analyses examined community-oriented and interest/learning-oriented activities separately. Sensitivity analyses included an exploratory model additionally adjusted for cognitive function at age 65 (word recall, delayed recall, and Stroop interference scores), which was measured in a subset of participants and imputed for the remainder via multiple imputation; lag exclusion analyses reclassifying dementia events occurring within 3 or 5 years of the age-70 examination as censored; prior sensitivity analyses using Normal(0, 10) and Normal(0, 0.5) priors; sex-subgroup and sex-interaction analyses; and a Bayesian Cox proportional hazards model with leave-one-out cross-validation. Because dementia has a long preclinical phase, a diagnosis arising soon after the age-70 examination may reflect disease already present when social participation was measured, so that cessation would be an early consequence rather than a cause. Reclassifying these early events as censored, while retaining all participants in the analysis, examined whether the association persisted once the events most susceptible to reverse causation were removed. All analyses were conducted in R version 4.5.3 using the brms package version 2.23.0 [27] with CmdStan version 2.38.0.
Results
Figure 1 shows the flow of participant selection. A total of 1,681 participants (880 men and 801 women) were included in the analysis. Of these, 1,310 (77.9%) were classified as Continuation, 65 (3.9%) as Cessation, 205 (12.2%) as Initiation, and 101 (6.0%) as Persistent nonparticipation. Compared with the Continuation group, the other three groups had higher proportions of men, lower educational attainment, lower competence in daily living, and a higher prevalence of depressive symptoms (Table 1).
Table 1.
Baseline characteristics of the study participants according to social participation change groups
| Characteristic | Continuation (n = 1,310) |
Cessation (n = 65) |
Initiation (n = 205) |
Persistent nonparticipation (n = 101) |
|---|---|---|---|---|
| Sex | ||||
| Male | 661 (50%) | 36 (55%) | 123 (60%) | 60 (59%) |
| Female | 649 (50%) | 29 (45%) | 82 (40%) | 41 (41%) |
| Body mass index category, kg/m² | ||||
| < 18.5 | 48 (3.7%) | 4 (6.2%) | 8 (3.9%) | 8 (7.9%) |
| 18.5–22.9 | 613 (47%) | 28 (43%) | 90 (44%) | 41 (41%) |
| ≥ 23 | 649 (50%) | 33 (51%) | 107 (52%) | 52 (51%) |
| Smoking status | ||||
| Never | 740 (56%) | 32 (49%) | 96 (47%) | 49 (49%) |
| Former | 368 (28%) | 19 (29%) | 68 (33%) | 37 (37%) |
| Current | 202 (15%) | 14 (22%) | 40 (20%) | 15 (15%) |
| Missing | 0 | 0 | 1 | 0 |
| Drinking status | ||||
| Never/Past | 692 (53%) | 39 (60%) | 99 (48%) | 58 (57%) |
| Current | 617 (47%) | 26 (40%) | 106 (52%) | 43 (43%) |
| Missing | 1 | 0 | 0 | 0 |
| Educational attainment | ||||
| Lower than high school | 349 (27%) | 25 (38%) | 61 (30%) | 42 (42%) |
| High school or above | 960 (73%) | 40 (62%) | 144 (70%) | 59 (58%) |
| Missing | 1 | 0 | 0 | 0 |
| Marital status | ||||
| Married | 1,189 (91%) | 56 (86%) | 184 (90%) | 89 (88%) |
| Other | 121 (9.2%) | 9 (14%) | 21 (10%) | 12 (12%) |
| Living arrangement | ||||
| With family | 1,261 (96%) | 61 (94%) | 199 (97%) | 95 (94%) |
| Alone | 49 (3.7%) | 4 (6.2%) | 6 (2.9%) | 6 (5.9%) |
| Work status | ||||
| Not working | 763 (59%) | 36 (56%) | 108 (53%) | 60 (59%) |
| Working | 535 (41%) | 28 (44%) | 95 (47%) | 41 (41%) |
| Missing | 12 | 1 | 2 | 0 |
| Walking time | ||||
| < 1 h/day | 556 (43%) | 26 (40%) | 85 (42%) | 44 (44%) |
| ≥ 1 h/day | 747 (57%) | 39 (60%) | 119 (58%) | 57 (56%) |
| Missing | 7 | 0 | 1 | 0 |
| Exercise frequency | ||||
| < 2/week | 1,075 (83%) | 54 (83%) | 175 (86%) | 89 (88%) |
| ≥ 2/week | 226 (17%) | 11 (17%) | 28 (14%) | 12 (12%) |
| Missing | 9 | 0 | 2 | 0 |
| Competence in daily living | ||||
| High competence | 1,211 (92%) | 57 (88%) | 168 (82%) | 68 (67%) |
| Lower competence | 99 (7.6%) | 8 (12%) | 37 (18%) | 33 (33%) |
| Depressive symptoms | ||||
| Presence | 205 (16%) | 25 (38%) | 58 (29%) | 44 (44%) |
| Absence | 1,104 (84%) | 40 (62%) | 145 (71%) | 57 (56%) |
| Missing | 1 | 0 | 2 | 0 |
| Medical histories of | ||||
| Hypertension | ||||
| Presence | 540 (41%) | 29 (45%) | 94 (46%) | 49 (49%) |
| Diabetes mellitus | ||||
| Presence | 111 (8.5%) | 4 (6.2%) | 25 (12%) | 10 (9.9%) |
| Cardiovascular disease | ||||
| Presence | 50 (3.8%) | 1 (1.5%) | 7 (3.4%) | 4 (4.0%) |
| Stroke | ||||
| Presence | 150 (11%) | 6 (9.2%) | 10 (4.9%) | 12 (12%) |
| Cognitive function | ||||
| Word recall total (0–30) | 22.0 (19.0, 24.0) | 21.0 (18.0, 22.0) | 21.0 (18.0, 24.0) | 21.5 (19.0, 24.0) |
| Missing | 841 | 39 | 122 | 63 |
| Delayed recall (0–10) | 8.0 (7.0, 9.0) | 8.0 (7.0, 9.0) | 8.0 (7.0, 9.0) | 8.0 (6.0, 9.0) |
| Missing | 842 | 39 | 122 | 63 |
| Stroop interference, seconds | 13.6 (9.2, 17.9) | 15.0 (10.8, 20.0) | 14.0 (8.6, 18.0) | 14.0 (10.0, 19.6) |
| Missing | 760 | 38 | 111 | 55 |
Values are presented as n (%) or median (Q1, Q3) for cognitive function variables
Percentages were calculated using non-missing values for each variable
The analytic sample included 1,681 participants with non-missing 10-item social activity participation data at both ages 65 and 70
Baseline characteristics were measured at the age-65 health examination
Cognitive function was assessed in a subset of participants: word recall (n = 616), delayed recall (n = 615), and the Stroop test (n = 717)
Higher word recall and delayed recall scores indicate better performance; higher Stroop interference values indicate worse performance
During 22,116 person-years of follow-up (median 13.6 years), 497 participants developed dementia and 138 died without dementia (Table 2). The hazard ratio for cessation increased progressively across models: from 1.17 (95% CrI: 0.95 to 1.50) in Model 1 to 1.27 (95% CrI: 1.02 to 1.64; posterior probability 0.98) in the fully adjusted model. Neither Initiation nor Persistent nonparticipation was associated with incident dementia in any model. When community-oriented and interest/learning-oriented activities were examined separately, no group showed a clear association with incident dementia (Table 3). The association between cessation and incident dementia was robust across all sensitivity analyses: additional adjustment for cognitive function at age 65 yielded a strengthened rather than attenuated estimate (HR 1.31, 95% CrI: 1.05 to 1.71), and lag exclusion analyses reclassifying events within 3 years (HR 1.21, 95% CrI: 1.01 to 1.50) or 5 years (HR 1.20, 95% CrI: 1.02 to 1.47) as censored produced consistent results. Alternative prior specifications did not materially change the estimates. The sex-interaction model revealed that the association between initiation and dementia differed between men and women (interaction HR 0.68, 95% CrI: 0.53 to 0.87), with evidence of an elevated hazard in men and a point estimate below unity in women (Table 3; Fig. 2). All chains converged satisfactorily, with no divergent transitions observed (Supplementary Figs. 1 and 2).
Table 2.
Bayesian cause-specific hazard ratios for incident dementia according to changes in social participation between ages 65 and 70
| Continuation | Cessation | Initiation | Persistent nonparticipation | ||||
|---|---|---|---|---|---|---|---|
| n | 1,310 | 65 | 205 | 101 | |||
| Dementia events | 395 | 14 | 57 | 31 | |||
| Person-years | 17,268 | 882 | 2,669 | 1,297 | |||
| HR (95% CrI) | P(HR > 1) | HR (95% CrI) | P(HR > 1) | HR (95% CrI) | P(HR > 1) | ||
| Model 1 | Ref. | 1.17 (0.95, 1.50) | 0.93 | 1.04 (0.92, 1.17) | 0.72 | 0.98 (0.84, 1.15) | 0.40 |
| Model 2 | Ref. | 1.24 (1.00, 1.60) | 0.97 | 1.04 (0.92, 1.17) | 0.72 | 1.00 (0.86, 1.19) | 0.52 |
| Model 3 | Ref. | 1.27 (1.02, 1.64) | 0.98 | 1.07 (0.95, 1.21) | 0.85 | 1.04 (0.89, 1.24) | 0.70 |
Model 1: sex, enrolment year
Model 2: Model 1 + body mass index, smoking, drinking, education, marital status, living arrangement, work status, walking time, exercise frequency
Model 3: Model 2 + competence in daily living, depressive symptoms, hypertension, diabetes mellitus, cardiovascular disease, stroke. Estimates were obtained from Bayesian Weibull cause-specific hazard models with weakly informative priors (Normal(0, 2.5))
Posterior distributions were pooled across 20 multiply imputed datasets
Ref reference group, HR hazard ratio, CrI credible interval, P(HR > 1) posterior probability that the hazard ratio exceeds 1
Table 3.
Secondary and sensitivity analyses for the association between changes in social participation and incident dementia
| Continuation (ref.) | Cessation | Initiation | Persistent nonparticipation | ||||
|---|---|---|---|---|---|---|---|
| n / events | HR (95% CrI) | P(HR > 1) | HR (95% CrI) | P(HR > 1) | HR (95% CrI) | P(HR > 1) | |
| Secondary analyses: Activity subtypes (Model 3) | |||||||
| Community-oriented (5 items) | 960 / 293 | 1.09 (0.92, 1.32) | 0.84 | 1.09 (0.99, 1.21) | 0.96 | 1.05 (0.94, 1.18) | 0.82 |
| Cessation: n = 101/25; Initiation: n = 353/102; Persistent nonparticipation: n = 267/77 | |||||||
| Interest/learning (5 items) | 1,002 / 302 | 1.08 (0.91, 1.32) | 0.81 | 1.04 (0.94, 1.16) | 0.79 | 1.09 (0.97, 1.23) | 0.92 |
| Cessation: n = 87/24; Initiation: n = 332/95; Persistent nonparticipation: n = 260/76 | |||||||
| Sensitivity analyses (Total social activity, Model 3) | |||||||
| + Cognitive function | 1,681 / 497 | 1.31 (1.05, 1.71) | 0.99 | 1.07 (0.95, 1.21) | 0.86 | 1.05 (0.89, 1.25) | 0.72 |
| 3-year lag exclusion | 1,681 / 478 | 1.21 (1.01, 1.50) | 0.98 | 1.05 (0.95, 1.17) | 0.83 | 1.05 (0.91, 1.22) | 0.74 |
| 5-year lag exclusion | 1,681 / 454 | 1.20 (1.02, 1.47) | 0.98 | 1.06 (0.97, 1.17) | 0.90 | 1.06 (0.93, 1.21) | 0.80 |
| Diffuse prior (Normal(0, 10)) | 1,681 / 497 | 1.27 (1.02, 1.64) | 0.98 | 1.07 (0.95, 1.21) | 0.85 | 1.04 (0.89, 1.24) | 0.70 |
| Sceptical prior (Normal(0, 0.5)) | 1,681 / 497 | 1.25 (1.01, 1.60) | 0.98 | 1.06 (0.95, 1.21) | 0.85 | 1.04 (0.89, 1.23) | 0.69 |
| Male only | 880 / 243 | 1.20 (0.86, 1.83) | 0.85 | 1.34 (1.10, 1.70) | 1.00 | 1.18 (0.93, 1.55) | 0.91 |
| Female only | 801 / 254 | 1.32 (0.98, 1.93) | 0.96 | 0.88 (0.75, 1.04) | 0.06 | 0.92 (0.74, 1.18) | 0.24 |
Reference group: Continuation
All analyses were based on Model 3 covariates (sex, enrolment year, body mass index, smoking, drinking, education, marital status, living arrangement, work status, walking time, exercise frequency, competence in daily living, depressive symptoms, hypertension, diabetes mellitus, cardiovascular disease, stroke)
Cognitive function adjustment additionally included standardised word recall, delayed recall, and Stroop interference scores imputed via multiple imputation. Lag exclusion analyses reclassified dementia events occurring within the specified period as censored
Sex-subgroup models excluded sex from covariates
Community-oriented activities: community events, neighbourhood associations, senior citizen clubs, volunteer activities, Silver Human Resources Centre. Interest/learning activities: hobby groups, sharing skills or experience, senior college, culture centre learning, public lectures or study groups
HR hazard ratio, CrI credible interval, P(HR > 1) posterior probability that the hazard ratio exceeds 1
Fig. 2.
Forest plots of Bayesian cause-specific hazard ratios for incident dementia associated with changes in social activity participation between ages 65 and 70. A Progressive covariate adjustment for cessation vs. continuation: Model 1 adjusted for sex and enrolment year; Model 2 additionally adjusted for body mass index, smoking, drinking, education, marital status, living arrangement, work status, walking time, and exercise frequency; Model 3 further adjusted for competence in daily living, depressive symptoms, hypertension, diabetes mellitus, cardiovascular disease, and stroke; Model 3 + Cognitive additionally adjusted for word recall, delayed recall, and Stroop interference scores. B Activity subtypes under the fully adjusted model (Model 3). Circles represent cessation, triangles represent initiation, and squares represent persistent nonparticipation, each compared with continuation as the reference. C Sensitivity analyses for cessation vs. continuation under Model 3. Posterior median hazard ratios and 95% credible intervals are annotated above each point estimate in all panels. All estimates were obtained from Bayesian Weibull cause-specific hazard models with weakly informative priors (Normal(0, 2.5)) and pooled across 20 multiply imputed datasets. The dashed vertical line indicates a hazard ratio of 1 (null)
Discussion
Cessation of social activity participation between ages 65 and 70 was associated with elevated dementia risk, with the association strengthening progressively across adjustment models and persisting after additional adjustment for cognitive function at age 65. Neither initiation nor persistent nonparticipation was associated with dementia risk, and neither activity subtype showed a clear individual association. A sex-specific pattern was observed for initiation, with an elevated hazard in men and estimates below unity in women.
The present finding that cessation of social activity participation was associated with elevated dementia risk is consistent with evidence from the JAGES (n = 47,698), where a downward trajectory in social participation was associated with the highest incidence of dementia among all trajectory groups [13]. In the Whitehall II Study, however, declining social participation was not significantly associated with dementia [12], possibly reflecting the longer interval between midlife exposure and late-life outcome. An earlier analysis of the NISSIN Project reported that nonparticipation in social activities was associated with elevated dementia risk, particularly among participants with hypertension or diabetes (HR 1.65, 95% CI: 1.21 to 2.26) [10]. While this earlier study examined the level of participation at a single time point, the present study demonstrates that cessation of social activities during the period surrounding retirement, when occupational roles, daily routines, and social networks are simultaneously disrupted, is associated with elevated dementia risk even after accounting for the competing risk of death. When community-oriented and interest/learning-oriented activities were examined separately, neither subtype showed a clear individual association with dementia, likely reflecting limited statistical power given the small number of participants classified as cessation within each activity category.
The null finding for persistent nonparticipation may reflect lifelong behavioural patterns or personality traits distinct from those who actively discontinued participation, or survivor bias among persistently nonparticipating individuals who reached age 70 without dementia. The null finding for initiation is consistent with evidence that social participation interventions have improved cognition but have not reduced dementia incidence [6], and the period of newly adopted activities beginning at age 70 may have been too brief to generate detectable effects. However, the sex-interaction analysis revealed that the association between initiation and dementia differed by sex: in men, initiation was associated with an elevated hazard, whereas in women, the point estimate was below unity. Sex-specific associations between social connection and dementia have been reported in the ASPREE cohort, where weak social connections were associated with elevated dementia risk among men but not women [29]. These patterns suggest that the protective potential of social engagement may differ between men and women, although the underlying mechanisms remain unclear.
The cognitive reserve framework posits that cognitively stimulating activities enhance the brain’s capacity to tolerate neuropathology [5, 30], and neuropathological evidence from the Rush Memory and Aging Project showed that larger social networks attenuated the association between Alzheimer’s disease pathology and cognitive decline [31]. The 2024 Lancet Commission proposed multiple protective pathways of social engagement, including building cognitive reserve, reducing neuroinflammation, and promoting healthy behaviours [4]. Cessation of social activities may disrupt these pathways simultaneously. In the NISSIN Project, continued regular social activity participation was associated with lower risk of developing depressive symptoms (RR 0.78, 95% CI: 0.62 to 0.97) [32], and depressive symptoms have been reported to mediate 41% to 79% of the association between social relationship factors and incident dementia [33]. Cessation of social activities during the period surrounding retirement may therefore increase dementia risk through multiple pathways, including reduced cognitive stimulation, loss of health-promoting behaviours, and the development of depressive symptoms.
Reverse causation is the foremost concern in this field: preclinical dementia may cause social withdrawal years before diagnosis. Declining social engagement trajectories before dementia diagnosis have been documented in the ELSA cohort [15], and in the UK Million Women Study (n = 850,000), no association between nonparticipation in cognitive or social activities and dementia was observed during the second decade of follow-up (RR 0.99, 99% CI: 0.95 to 1.03), leading to the conclusion that these activities had little or no causal effect on dementia incidence [14]. The present study addressed this threat through the landmark design ensuring temporal separation between exposure and outcome, lag exclusion analyses producing consistent results after reclassifying events within 3 and 5 years, and adjustment for cognitive function at age 65 yielding a strengthened rather than attenuated association. The hazard ratio point estimate for cessation increased progressively from 1.17 in the minimally adjusted model to 1.27 in the fully adjusted model, and further to 1.31 after additional adjustment for cognitive function at age 65, with the credible interval width remaining similar across models. If reverse causation were driving the observed association, adjustment for cognitive function would be expected to attenuate the estimate toward the null. The fact that the point estimate moved further from unity rather than toward it suggests that the association between cessation and dementia is not solely explained by preclinical cognitive decline. Nevertheless, reverse causation cannot be entirely excluded. In the JAGES (n = 47,698), targeted maximum likelihood estimation accounting for bidirectional feedback between social participation and cognitive decline showed that sustained social participation increased dementia-free survival by 3.2% points (95% CI: 1.9 to 4.5) [13], and the present findings provide complementary evidence from a Bayesian competing risk framework.
This study has several strengths, including the age-specific cohort design eliminating age as a confounder, the longitudinal exposure assessment capturing change in social participation during the period surrounding retirement, the long follow-up period, the Bayesian cause-specific hazard model explicitly accounting for death without dementia as a competing event, and multiple imputation addressing missing covariate data [34, 35].
Limitations
The study population was drawn from a single suburban municipality, and participants were somewhat healthier than the general Japanese population of comparable age [19], which may limit generalisability. Social participation was ascertained through self-report at two time points using a binary classification, which may obscure dose-response relationships. Participation was assessed only at ages 65 and 70, so the exposure reflects the net change between these anchors and does not capture the trajectory in between, including temporary cessation followed by resumption. Unrecorded transient cessation among participants classified as continuation would raise the risk in the reference group and thereby bias the cessation estimate toward the null, although for a four-category exposure the overall direction of such misclassification cannot be guaranteed. Incident dementia was identified through long-term care insurance certification rather than clinical diagnosis, which, although validated in Japan [22], may underascertain mild cases. Residual confounding by unmeasured factors such as personality traits and genetic risk cannot be excluded. In addition, the modest sample size, particularly in the cessation group, limited statistical power for secondary subgroup analyses. Accordingly, the activity-specific and sex-specific findings should be interpreted as exploratory. Caution is also warranted when interpreting the cognitive function-adjusted results given the high proportion of missing values in these variables. Finally, enrolment took place between 1996 and 2005. Because the exposure was defined as a change in participation rather than any specific activity, recent cessation may generalise across periods more readily than activity-specific findings, which supports its use as a behavioural marker in current practice. Caution is nonetheless required. Since enrolment, legislation has extended employment opportunities toward ages 65 and 70 [18], so the ages studied may now coincide with a period of continued employment rather than the years following retirement, and the relationship between this window and the period surrounding retirement differs from that in contemporary cohorts. In addition, the COVID-19 pandemic has shifted participation toward remote forms whose cognitive and social value may differ from in-person activity, so the magnitude of the association may not extrapolate directly to contemporary post-pandemic populations.
Conclusions
Cessation of social activity participation between ages 65 and 70 was associated with elevated dementia risk, with the association robust across adjustment models, lag exclusion analyses, and prior sensitivity analyses. A sex-specific pattern was observed for initiation. These findings may have practical implications for dementia prevention in later life. In clinical and community-based settings during the transition out of full-time work, recent cessation of social activity may serve as an early behavioural marker of increased dementia risk. Supporting the continuation of existing social activities may help identify and assist older adults at increased risk before functional decline becomes clinically apparent.
Supplementary Information
Acknowledgements
We thank the staff of the Nisshin Medical and Dental Associations and the Health Center and Hygiene Department of Nisshin City for their cooperation and great efforts in conducting our study. An AI-assisted language tool (Claude, Anthropic) was used to assist with English-language editing of the manuscript. The authors take full responsibility for the content of the publication.
Clinical trial number
Not applicable.
Abbreviations
- CrI
Credible interval
- DSM-IV
Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition
- ELSA
English Longitudinal Study of Ageing
- GDS
Geriatric Depression Scale
- HR
Hazard ratio
- JAGES
Japan Gerontological Evaluation Study
- LTCI
Long-Term Care Insurance
- MCMC
Markov chain Monte Carlo
- MICE
Multivariate imputation by chained equations
- NISSIN
New Integrated Suburban Seniority Investigation
- TMIG
Tokyo Metropolitan Institute of Gerontology
Authors' contributions
S.U. contributed to the methodology, formal analysis, investigation, data curation, writing of the original draft, visualisation, and funding acquisition. W.Z., S.O., T.K., Y.S., H.W., M.A., K.W., and K.T. contributed to study design, data acquisition, interpretation of the data, and critical revision of the manuscript. A.T. contributed to study design, data acquisition and interpretation, supervision, project administration, funding acquisition, and critical revision of the manuscript. All authors read and approved the final manuscript.
Funding
This work was supported by a Grant-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science, and Technology of Japan (JP15390197, JP25893003, JP26460760, JP26520105, JP20K02392 [CoBiA], and JP24K05554 [CoBiA]), Uehara Memorial Foundation, Mitsui Sumitomo Insurance Welfare Foundation, Health Promotion Foundation, and Pfizer Health Research Foundation. The funders had no role in the design of the study; the collection, analysis, and interpretation of the data; or the writing of the manuscript.
Data availability
The data that support the findings of this study are not publicly available due to privacy and ethical restrictions. Data are available from the corresponding author upon reasonable request and with permission of the relevant institutions.
Declarations
Ethics approval and consent to participate
This study was conducted in accordance with the principles of the Declaration of Helsinki. The study protocol was approved by the ethics committees of the participating institutions, including Hokkaido University Graduate School of Medicine (2014, no. 37), Aichi Medical University School of Medicine (2008, no. 558), National Center for Geriatrics and Gerontology (2006, no. 242), Nagoya University Graduate School of Medicine (2002 and 2004, no. 162), and Osaka City University School of Human Life Science (2021, no. 21-01). Written informed consent was obtained from all participants.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.GBD 2019 Dementia Forecasting Collaborators. Estimation of the global prevalence of dementia in 2019 and forecasted prevalence in 2050: an analysis for the Global Burden of Disease Study 2019. Lancet Public Health. 2022;7:e105–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Kuiper JS, Zuidersma M, Oude Voshaar RC, Zuidema SU, van den Heuvel ER, Stolk RP, Smidt N. Social relationships and risk of dementia: a systematic review and meta-analysis of longitudinal cohort studies. Ageing Res Rev. 2015;22:39–57. [DOI] [PubMed] [Google Scholar]
- 3.Penninkilampi R, Casey A-N, Singh MF, Brodaty H. The association between social engagement, loneliness, and risk of dementia: a systematic review and meta-analysis. J Alzheimers Dis. 2018;66:1619–33. [DOI] [PubMed] [Google Scholar]
- 4.Livingston G, Huntley J, Liu KY, et al. Dementia prevention, intervention, and care: 2024 report of the Lancet standing Commission. Lancet. 2024;404:572–628. [DOI] [PubMed] [Google Scholar]
- 5.Stern Y. What is cognitive reserve? Theory and research application of the reserve concept. J Int Neuropsychol Soc. 2002;8:448–60. [PubMed] [Google Scholar]
- 6.Sommerlad A, Kivimäki M, Larson EB, et al. Social participation and risk of developing dementia. Nat Aging. 2023;3:532–45. [DOI] [PubMed] [Google Scholar]
- 7.Su S, Shi L, Zheng Y, et al. Leisure activities and the risk of dementia: a systematic review and meta-analysis. Neurology. 2022;99:e1651–63. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Fancourt D, Steptoe A, Cadar D. Community engagement and dementia risk: time-to-event analyses from a national cohort study. J Epidemiol Community Health. 2020;74:71–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Chen Y, Grodstein F, Capuano AW, Wang T, Bennett DA, James BD. Late-life social activity and subsequent risk of dementia and mild cognitive impairment. Alzheimers Dement. 2025;21:e14316. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Taniguchi R, Ukawa S, Zhao W, et al. Associations of hypertension and/or diabetes and the risk of dementia according to participation in social activities from an age-specific community-based cohort study (the NISSIN Project). Arch Gerontol Geriatr Plus. 2024;1:100023. [Google Scholar]
- 11.Kobayashi E, Sugawara I, Fukaya T, Okamoto S, Liang J. Retirement and social activities in Japan: does age moderate the association? Res Aging. 2022;44:144–55. [DOI] [PubMed] [Google Scholar]
- 12.Sommerlad A, Sabia S, Singh-Manoux A, Lewis G, Livingston G. Association of social contact with dementia and cognition: 28-year follow-up of the Whitehall II cohort study. PLoS Med. 2019;16:e1002862. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Matsuyama Y, Shirai K, Aida J. Sustained social participation and dementia: evidence from a Japanese longitudinal cohort study with a time-varying exposure analysis. Soc Sci Med. 2025;385:118608. [DOI] [PubMed] [Google Scholar]
- 14.Floud S, Balkwill A, Sweetland S, et al. Cognitive and social activities and long-term dementia risk: the prospective UK Million Women Study. Lancet Public Health. 2021;6:e116–23. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Hackett RA, Steptoe A, Cadar D, Fancourt D. Social engagement before and after dementia diagnosis in the English Longitudinal Study of Ageing. PLoS ONE. 2019;14:e0220195. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Kivimäki M, Singh-Manoux A, Pentti J, et al. Physical inactivity, cardiometabolic disease, and risk of dementia: an individual-participant meta-analysis. BMJ. 2019;365:l1495. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Fine JP, Gray RJ. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc. 1999;94:496–509. [Google Scholar]
- 18.Ministry of Health, Labour and Welfare. Act on Stabilization of Employment of Elderly Persons (Act No. 68 of 1971, as amended). Accessed April 2026. https://www.japaneselawtranslation.go.jp/en/laws/view/2621.
- 19.Kitamura T, Kawamura T, Tamakoshi A, Wakai K, Ando M, Ohno Y. Rationale, design, and profiles of the New Integrated Suburban Seniority Investigation (NISSIN) Project: a study of an age-specific, community-based cohort of Japanese elderly. J Epidemiol. 2009;19:237–43. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Aoki R, Ohno Y, Tamakoshi A, et al. Lifestyle determinants for social activity levels among the Japanese elderly. Arch Gerontol Geriatr. 1996;22:271–86. [DOI] [PubMed] [Google Scholar]
- 21.Tamiya N, Noguchi H, Nishi A, et al. Population ageing and wellbeing: lessons from Japan’s long-term care insurance policy. Lancet. 2011;378:1183–92. [DOI] [PubMed] [Google Scholar]
- 22.Noda H, Yamagishi K, Ikeda A, Asada T, Iso H. Identification of dementia using standard clinical assessments by primary care physicians in Japan. Geriatr Gerontol Int. 2018;18:738–44. [DOI] [PubMed] [Google Scholar]
- 23.Alvares D, Lázaro E, Gómez-Rubio V, Armero C. Bayesian survival analysis with BUGS. Stat Med. 2021;40:2975–3020. [DOI] [PubMed] [Google Scholar]
- 24.Koyano W, Shibata H, Nakazato K, Haga H, Suyama Y. Measurement of competence: reliability and validity of the TMIG Index of Competence. Arch Gerontol Geriatr. 1991;13:103–16. [DOI] [PubMed] [Google Scholar]
- 25.Yesavage JA, Sheikh JI. Geriatric Depression Scale (GDS): recent evidence and development of a shorter version. Clin Gerontol. 1986;5:165–73. [Google Scholar]
- 26.Kashiwagi A, Kasuga M, Araki E, et al. International clinical harmonization of glycated hemoglobin in Japan: from Japan Diabetes Society to National Glycohemoglobin Standardization Program values. J Diabetes Investig. 2012;3:39–40. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Bürkner P-C. brms: an R package for Bayesian multilevel models using Stan. J Stat Softw. 2017;80:1–28. [Google Scholar]
- 28.van Buuren S, Groothuis-Oudshoorn K. mice: multivariate imputation by chained equations in R. J Stat Softw. 2011;45:1–67. [Google Scholar]
- 29.Htun HL, Teshale AB, Ryan J, et al. Gender-specific analysis of social connection patterns and risk of dementia in community-dwelling older people. Alzheimers Dement. 2024;20:4879–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Stern Y, Arenaza-Urquijo EM, Bartrés-Faz D, et al. Whitepaper: defining and investigating cognitive reserve, brain reserve, and brain maintenance. Alzheimers Dement. 2020;16:1305–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Bennett DA, Schneider JA, Tang Y, Arnold SE, Wilson RS. The effect of social networks on the relation between Alzheimer’s disease pathology and level of cognitive function in old people: a longitudinal cohort study. Lancet Neurol. 2006;5:406–12. [DOI] [PubMed] [Google Scholar]
- 32.Shan Y, Zhao W, Hao W, et al. Changes in behavioral activities and transition of depressive symptoms among younger-old community-dwelling adults during 6 years: an age-specific prospective cohort study. Int J Geriatr Psychiatry. 2022;37:e5778. [DOI] [PubMed] [Google Scholar]
- 33.Duffner LA, Deckers K, Cadar D, de Vugt ME, Köhler S. Social relationship factors, depressive symptoms, and incident dementia: a prospective cohort study into their interrelatedness. Psychol Med. 2024;54:3115–25. [DOI] [PubMed] [Google Scholar]
- 34.Sterne JAC, White IR, Carlin JB, et al. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ. 2009;338:b2393. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Woods AD, Gerasimova D, Van Dusen B, et al. Best practices for addressing missing data through multiple imputation. Infant Child Dev. 2024;33:e2407. [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data that support the findings of this study are not publicly available due to privacy and ethical restrictions. Data are available from the corresponding author upon reasonable request and with permission of the relevant institutions.