Ambient Temperature and 24-Hour Movement Behaviors in Community-Dwelling Older Adults in Japan: The Kyotango Longevity Cohort Study

Abstract

Background

Sedentary behavior (SB), physical activity, and sleep duration are 24-hour movement behaviors and key indicators of healthy aging. This study examined the association between daily mean ambient temperature and 24-hour movement behaviors among community-dwelling older adults.

Methods

Data were obtained from 769 participants aged ≥65 years (mean age 73.5±5.5 years; 59.0% female) in the Kyotango Longevity Cohort Study conducted in Kyotango City, Japan. SB and physical activity were objectively assessed in each participant using a triaxial accelerometer over 7 consecutive days (4,344 valid person-days in total). Associations between daily mean ambient temperature and 24-hour movement behaviors were evaluated using linear mixed-effects models with natural cubic splines. The association between ambient temperature and sleep duration, estimated from sleep diaries or standardized sleep windows when diary data were unavailable, was also examined.

Results

Significant nonlinear associations were observed between daily mean ambient temperature and all 24-hour movement behaviors (all p<0.05). SB exhibited a U-shaped pattern, with the predicted minimum at approximately 20.8°C. By contrast, light-intensity and moderate-to-vigorous-intensity physical activity showed inverted U-shaped associations, peaking at approximately 20.7°C and 21.1°C, respectively. Sleep duration remained relatively stable at low-to-moderate ambient temperatures but tended to decline above approximately 22°C.

Conclusion

Ambient temperature is an important environmental factor associated with 24-hour movement behaviors in older adults. These findings underscore the importance of considering thermal conditions when developing strategies to reduce SB, increase physical activity, and maintain appropriate sleep durations in older adults.

INTRODUCTION

Population aging is progressing rapidly worldwide.¹) With increasing longevity, prolonging a healthy life expectancy—the years lived in good health—has become a public health priority.²) Lifestyle behaviors are influential modifiable determinants of healthy aging. Sedentary behavior (SB), physical activity, and sleep are collectively regarded as 24-hour movement behaviors that are recognized as indicators of healthy aging.^3,4) Daily time allocation across these behaviors, including reduced SB, greater engagement in physical activity, and appropriate sleep duration, is associated with better cardiometabolic health and lower mortality risk.^5-8) However, daily movement patterns in older adults are often characterized by prolonged SB and limited physical activity.⁹)

Ambient temperature is a fundamental environmental factor that may influence day-to-day 24-hour movement behaviors. Older adults are particularly vulnerable to thermal stress because of age-related declines in the physiological aspects of thermoregulation.¹⁰) Although seasonal variation in physical activity has been documented,¹¹) day-to-day fluctuations in ambient temperature may more directly influence 24-hour movement behaviors. Given that both low and high temperatures may discourage outdoor activity and alter daily routines, the association between ambient temperature and the 24-hour movement behaviors may not be linear.

Wearable accelerometers have been used to objectively assess SB and physical activity in older adults.¹²) These small devices can be worn in daily life with minimal burden, enabling the classification of movement behaviors by intensity level. By linking daily accelerometer data with meteorological records, associations between ambient temperature and time spent in SB and physical activity at different intensities can be assessed.

This study examined potential nonlinear associations between daily mean ambient temperature and time spent in each of the 24-hour movement behaviors among community-dwelling older adults. In the primary analysis, behavior-specific optimal temperatures—defined as the ambient temperatures at which SB is minimized and physical activity is maximized—were estimated using accelerometer-derived waking-time activity data. A secondary analysis examined the association between ambient temperature and sleep duration.

MATERIALS AND METHODS

Participants

This study used data from the Kyotango Longevity Cohort Study, an ongoing community-based prospective cohort study led by the Department of Longevity and Regional Epidemiology, Kyoto Prefectural University of Medicine (Kyoto, Japan). The study area was Kyotango City in northern Kyoto Prefecture, Japan, a region known for longevity because it has a centenarian rate approximately three times the national average.¹³) A total of 1,244 community-dwelling older adults aged ≥65 years who participated in the Kyotango Longevity Cohort Study between August 2017 and September 2025 were included in the present analyses. Participants performed one or two assessment periods during the study period. Participants were recruited through the Municipal Longevity Health Check-up program, with announcements distributed via the official city website and community newsletters of Kyotango City Yasaka Hospital. Detailed data collection procedures have been described elsewhere.¹⁴)

Ethical Considerations

The study protocol was approved by the Ethics Review Board of Kyoto Prefectural University of Medicine (Approval No. ERB-C-885-9). All procedures were conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants prior to enrollment in the study. The study was reported in compliance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.¹⁵)

Data Collection

Data regarding the participants’ demographic characteristics (age and sex), health behaviors (current smoking and drinking status), socioeconomic factors (marital status and educational attainment [bachelor’s degree or higher]), and medical history (treated hypertension, diabetes, and dyslipidemia) were collected. Height and weight were measured, and body mass index was calculated as follows: weight (kg) divided by height (m) squared. To account for potential behavioral restrictions during the coronavirus disease 2019 (COVID-19) pandemic, assessment days corresponding with the Japanese government’s State of Emergency or Priority Preventive Measures were documented as a covariate.

Daily meteorological data were retrieved from the Japan Meteorological Agency’s official database¹⁶) and linked to each accelerometer assessment day. To ensure geographic correspondence between environmental exposure parameters and participants’ residences, data from the Taiza Observatory, located within Kyotango City, were used. The variables obtained included daily mean temperature (°C), precipitation (mm), sunshine duration (hour), daylight duration (hour), and mean wind speed (m/s). For the analysis, daily precipitation was categorized into three levels (<1 mm, 1–9.9 mm, and ≥10 mm) to reflect its varying impact on outdoor physical activity. The percentage of possible sunshine was calculated as follows: (daily sunshine duration / total daylight duration) × 100.

Assessment of 24-Hour Movement Behaviors

SB and physical activity were objectively assessed using a triaxial accelerometer (ActiGraph wGT3X-BT; ActiGraph LLC, Pensacola, FL, USA). Participants were instructed to wear the device on an elastic belt over the right hip for 7 consecutive days, removing it only during water-based activities (e.g., bathing or swimming). Accelerations were recorded at a sampling frequency of 30 Hz. Data were processed using ActiLife software (version 6.13.4) with the low-frequency extension filter enabled to improve sensitivity to the lower-intensity movements common among older adults.¹⁷) Data were aggregated into 60-second epochs. During waking time, non-wear time was defined as ≥90 consecutive minutes of zero counts. A valid day required ≥10 hours of wear time during waking time, and participants with ≥4 valid days were included in the analysis. Each day was partitioned into sleep and wake periods. Sleep periods were identified using participants’ sleep diaries. When diary data were unavailable, a standardized sleep window (23:00–06:00) was applied for sleep–wake classification. During waking hours, activity intensity was classified using vector magnitude counts per minute (cpm) based on established cut-points for SB (<200 cpm), light-intensity physical activity (LPA; 200–2,689 cpm), and moderate-to-vigorous-intensity physical activity (MVPA; ≥2,690 cpm),^18,19) corresponding to ≤1.5, 1.6–2.9, and ≥3.0 metabolic equivalents, respectively.²⁰)

Statistical Analyses

Participants’ characteristics are presented as means with standard deviations (SDs) for continuous variables and as counts with percentages for categorical variables. Daily mean ambient temperature was the primary exposure variable. Outcome variables included daily time spent in SB, LPA, MVPA, and sleep (minutes/day). Because the MVPA data exhibited a right-skewed distribution, a square root transformation was applied to improve normality in the models. Models were fitted on the transformed scale. MVPA was modeled on the square-root scale; therefore, back-transformation to the original scale (minutes/day) was performed without assuming symmetry of confidence intervals on the original scale. For predicted values and estimated marginal means, 95% confidence intervals were first calculated on the square-root scale and then back-transformed by applying the inverse transformation (squaring) to each confidence limit separately, resulting in asymmetric confidence intervals on the original scale. For pairwise contrasts and average marginal effects, standard errors were propagated to the original scale using a first-order delta-method approximation, and Wald 95% confidence intervals were subsequently calculated on the minutes/day scale. Multiplicity-adjusted confidence intervals were used for pairwise temperature contrasts where appropriate.

Linear mixed-effects models were fitted to evaluate nonlinear associations between daily mean ambient temperature and each outcome variable, with ambient temperature modeled using natural cubic splines. Two to four degree of freedom were compared using the Akaike Information Criterion, and three degree of freedom was selected because it improved model fit relative to two, and no further improvement was observed with four. The same specification was applied across outcome variables to ensure consistency. Models were adjusted for age, sex, body mass index, waking time (calculated as follows: SB + LPA + MVPA), day type (weekday or weekend), smoking status, alcohol consumption, marital status, educational attainment, treated hypertension, diabetes, dyslipidemia, precipitation category, percentage of possible sunshine, mean wind speed, COVID-19–related restrictions, and measurement occasion. Waking time was included to account for between-day differences in observed waking time. Measurement occasion was included as a fixed effect to account for systematic differences between the first and second assessments, whereas random intercepts were included for participant ID to account for within-person correlations across repeated measurements and for assessment date to account for clustering of observations measured on the same calendar day. The overall association of the spline terms was assessed using Type II Wald chi-square tests.

Temperature–outcome associations were visualized by plotting model-based estimated marginal means across the observed temperature range. Continuous covariates were fixed at their sample means, whereas categorical covariates were averaged over their observed distributions using proportional weights (emmeans package in R) to estimate population-average associations. Associations for covariates were summarized using average marginal effects. Model assumptions were evaluated by visual inspection of residual plots, multicollinearity was assessed using variance inflation factors, and model fit was summarized using marginal and conditional R² values.

Optimal ambient temperatures were defined as the ambient temperatures corresponding to the nadir of the model-predicted curve for SB and the peaks of the model-predicted curves for LPA and MVPA. For outcomes with sparse observations, in which predictions at extreme ambient temperatures were less stable, the search for optimal values was restricted to the 5th–95th percentile of the observed ambient temperature distribution to reduce the influence of extreme values. For sleep duration, a threshold ambient temperature was estimated based on the temperature-specific slope of the predicted estimated marginal mean. Marginal temperature trends were computed using the emtrends function (emmeans package) across the 5th–95th percentile ambient temperature range, averaging over categorical covariates with proportional weights. The threshold ambient temperature was defined as the first temperature at which the estimated trend became negative and statistically significant (two-sided p<0.05).

All analyses were performed using R (version 4.5.2). Linear mixed-effects models were fitted using lme4 (version 1.1-38), with p-values obtained via lmerTest (version 3.1-3). Natural cubic splines were implemented using the splines package (included in R version 4.5.2). Predicted means and average marginal effects were estimated using emmeans (version 2.0.1) and marginaleffects (version 0.31.0), respectively. Model diagnostics were conducted using performance (version 0.15.3).

RESULTS

The participant selection flowchart is shown in Fig. 1. After exclusions, 769 community-dwelling older adults were included in the present analyses. Of these, 667 participants contributed data from a single assessment period, and 102 contributed data from two assessment periods. Baseline characteristics are summarized in Table 1. During the study period, 123 participants performed assessments during COVID-19-related State of Emergency or Priority Preventive Measures periods.

Fig. 1.

Flowchart of participant selection and derivation of the final sample included in this study.

Table 1.

Characteristics of the study participants

Characteristic	Value
Demographic characteristics
Age (y)	73.5±5.5
Female sex	454 (59.0)
Height (cm)	157.4±8.4
Weight (kg)	57.1±10.0
Body mass index (kg/m2)	23.0±3.0
Health behaviors
Current smoker	31 (4.0)
Current drinker	302 (39.3)
Socioeconomic factors
Married	585 (76.1)
Educational attainment
Bachelor’s degree or higher	121 (15.7)
Comorbidities
Treated hypertension	279 (36.3)
Treated diabetes	88 (11.4)
Treated dyslipidemia	237 (30.8)

Values are presented as mean±standard deviation or n (%).

The dataset comprised 4,344 valid person-days (mean 5.6 days per participant), with a mean waking time of 965.6±78.6 minutes per day. During waking time, 464.0 ± 112.4 minutes per day was spent in SB, 456.3±109.6 minutes per day was spent performing LPA, and 45.3±34.4 minutes per day was spent performing MVPA. Mean sleep duration was 453.4±65.5 minutes per day. Sleep periods were identified using sleep diaries for 3,971 person-days (91.4%), whereas a standardized sleep window (23:00–06:00) was applied to 373 person-days (8.6%). Daily mean temperatures during the assessment periods ranged from −2.1°C to 31.8°C (mean 17.3±7.6 °C; Supplementary Table S1).

As shown in Table 2, significant nonlinear associations were observed between daily mean ambient temperature and all outcome variables. Across all models, conditional R² values were substantially higher than marginal R² values, indicating considerable between-individual variability in 24-hour movement behaviors. All variance inflation factors were <2, suggesting no meaningful multicollinearity. As shown in Fig. 2, optimal ambient temperature patterns differed between behavior categories. SB exhibited a U-shaped pattern, with the predicted mean reaching its minimum at an optimal ambient temperature of approximately 20.8°C. By contrast, LPA and MVPA showed inverted U-shaped associations, peaking at the optimal ambient temperatures of approximately 20.7°C and 21.1°C, respectively. Sleep duration was relatively stable at low-to-moderate ambient temperatures and declined as temperature increased; the approximate ambient temperature threshold above which sleep duration began to decline was 22.0°C. Predicted 24-hour movement behavior values at these optimal ambient temperatures and the threshold temperature are presented in Table 3.

Table 2.

Nonlinear associations between daily mean temperature and 24-hour movement behaviors

Outcomes	Wald χ²	p-value	Marginal R²	Conditional R²
SB (min/day)	18.88	< 0.001	0.233	0.718
LPA (min/day)	15.89	0.001	0.282	0.733
MVPA (min/day)	14.74	0.002	0.126	0.690
Sleep (min/day)	12.02	0.007	0.502	0.769

Values are derived from linear mixed-effects models in which daily mean temperature was modeled using a natural cubic spline (with three degree of freedom). The overall association of the spline terms was assessed using a Type II Wald chi-square test. Marginal R² values indicate the variance explained by the fixed effects, whereas conditional R² values indicate the variance explained by the full model (fixed and random effects).

SB, sedentary behavior; LPA, light-intensity physical activity; MVPA, moderate-to-vigorous-intensity physical activity.

Fig. 2.

Nonlinear associations between daily mean ambient temperature and 24-hour movement behaviors: (A) SB, (B) LPA, (C) MVPA, and (D) sleep. Predicted daily time spent in SB, LPA, MVPA, and sleep was estimated using linear mixed-effects models with natural cubic splines (three degree of freedom), adjusted for all covariates. Solid lines and shaded areas represent estimated marginal means and 95% confidence intervals, respectively. Filled circles indicate the behavior-specific optimal ambient temperatures (the ambient temperature at which SB reached its minimum and those at which LPA and MVPA reached their maximum predicted values) and the threshold temperature for sleep (the temperature above which sleep duration began to decline). Vertical dashed lines indicate the 5th and the 95th percentile of daily mean temperature in the dataset. SB, sedentary behavior; LPA, light-intensity physical activity; MVPA, moderate-to-vigorous-intensity physical activity.

Table 3.

Optimal and threshold ambient temperatures and predicted durations for the 24-hour movement behaviors

Movement behaviors	Optimal/threshold temperature (°C)	Predicted value	95% CI
			Lower	Upper
SB (min/day)	20.8	453.5	444.4	462.6
LPA (min/day)	20.7	468.4	460.2	476.6
MVPA (min/day)	21.1	38.8	36.1	41.5
Sleep (min/day)	22.0	455.0	451.5	458.4

Predicted values are model-based estimates (min/day) with 95% confidence intervals (CIs) from fully adjusted linear mixed-effects models. The optimal/threshold temperature indicates the behavior-specific ambient temperature derived from the fully adjusted models. For SB, LPA, and MVPA, the value represents the optimal ambient temperature at which the predicted mean reached its minimum for sedentary behavior (SB) or maximum for light-intensity physical activity (LPA) and moderate-to-vigorous-intensity physical activity (MVPA). For sleep duration, the value represents the threshold ambient temperature above which the predicted mean began to decline.

DISCUSSION

This study examined the association between daily mean ambient temperature and 24-hour movement behaviors among community-dwelling older adults. Significant nonlinear associations were observed in all behaviors, such that SB showed a U-shaped association with ambient temperature, whereas LPA and MVPA demonstrated inverted U-shaped associations. The predicted optimal ambient temperatures were remarkably similar across activity intensities, occurring at approximately 21°C. These findings suggest that day-to-day fluctuations in ambient temperature are closely associated with the distribution of time spent in the 24-hour movement behaviors in later life.

The inverted U-shaped associations of ambient temperature with LPA and MVPA may reflect behavioral responses to thermal comfort, whereby moderate environmental temperatures create conditions that facilitate voluntary movement in daily life. Similar nonlinear associations between ambient temperature and physical activity have been reported in previous studies of older adults, with activity levels increasing under moderate temperatures and declining under both colder and hotter conditions.²¹) Interestingly, the optimal ambient temperature observed in the present study (approximately 21°C) was slightly higher than the approximately 17°C peak reported previously.²¹) This difference may partly reflect variations in study design, including differences in activity metrics (step counts vs. accelerometer-derived activity intensities) and analytical approaches used to model temperature–activity associations. Nevertheless, these findings consistently suggest that moderate ambient temperatures are most conducive to physical activity among older adults. This pattern may be partly explained by age-related declines in thermoregulatory capacity, evidenced by impairments in temperature sensitivity as well as physiological and behavioral responses,^10,22,23) which may influence how older adults perceive and respond to ambient temperature. When ambient temperatures decrease, unfavorable weather conditions and reduced accessibility to outdoor environments may discourage outdoor activity, leading to increased sedentary behavior. Conversely, as temperatures rise, thermal discomfort and perceived environmental strain may similarly limit engagement in physical activity. These behavioral responses may partly contribute to the U-shaped pattern observed in SB and the inverted U-shaped patterns observed in LPA and MVPA. Nevertheless, the relatively low marginal R² of the MVPA model suggests that ambient temperature accounted for a smaller proportion of the population-level variability in MVPA than in the other movement behaviors. Therefore, factors other than ambient temperature likely contribute substantially to MVPA among older adults.

Another feature of the temperature–MVPA association was a slight increase in MVPA toward the lower end of the observed ambient temperature range (Fig. 2). This pattern may reflect region-specific behavioral adaptations to relatively cooler temperatures. Kyotango City, where this study was conducted, is located along the Sea of Japan and experiences snowfall during winter. Such conditions may involve physically demanding activities (e.g., snow removal), which could contribute to episodic increases in MVPA despite overall reductions in LPA. However, estimates near the boundaries of the distribution should be interpreted cautiously because of wider confidence intervals.

The present findings also demonstrate that sleep duration tended to decline once ambient temperatures exceeded approximately 22°C. This pattern is consistent with previous evidence showing that higher nighttime temperatures are associated with shorter sleep durations.²⁴) Elevated temperatures may impair sleep initiation and maintenance through thermoregulatory mechanisms.²⁵) Older adults may be particularly vulnerable because thermoregulatory capacity declines with age,¹⁰) which may partly explain the decline in sleep duration with rising ambient temperature observed in the present study. Sleep duration also appeared to be slightly longer toward the lower end of the observed ambient temperature range. Nevertheless, as with MVPA, estimates near the boundaries of the distribution had wider confidence intervals and should therefore be interpreted cautiously. Notably, the ambient temperature associated with peak physical activity (approximately 21°C) was similar to the temperature at which sleep duration began to decline (approximately 22°C). This narrow gap of approximately 1°C suggests that environmental conditions favorable for daytime activity may already approach the upper limit for maintaining comfortable sleep conditions at night. Therefore, ambient temperature influences multiple 24-hour movement behaviors in different directions.

From a public health perspective, the present findings highlight the potential importance of reducing SB and promoting LPA and MVPA in older adults, while also suggesting that warmer conditions may warrant greater attention to nighttime thermal environments. Furthermore, climate-responsive behavioral and environmental strategies may help support healthier daily movement patterns under varying environmental conditions. Such insights may have increased public health relevance in Japan and other aging societies, where population aging is progressing rapidly. As the climate continues to warm, understanding how ambient temperature influences 24-hour movement behaviors may help inform strategies that promote active lifestyles and healthy aging among the growing population of older adults.

This study has several limitations. First, the observational study design precludes drawing causal inferences from the findings. Second, ambient temperature was derived from a single meteorological observatory, which may not capture individual-level exposure, microclimatic variation, or indoor environmental conditions. In addition, older adults spend a substantial proportion of their time indoors, and information on household cooling and heating use was unavailable. Therefore, indoor thermal environments could not be considered in the present study. Third, sleep duration was estimated using sleep diaries or standardized sleep windows. However, a sensitivity analysis restricted to person-days with diary-confirmed sleep yielded results that were highly consistent with the primary analysis, suggesting that the use of the standardized sleep window had minimal influence on the observed associations (Supplementary Table S2). Sleep quality indicators such as sleep efficiency, sleep onset latency, and nocturnal awakenings could not be evaluated. The potential influence of these unmeasured sleep characteristics on our findings warrants further investigation. Fourth, because each measurement period spanned only 7 consecutive days, the ambient temperatures experienced by an individual participant were strongly influenced by the season of enrollment. Therefore, the observed associations may partly reflect cross-seasonal differences rather than solely within-person temperature responses. Although no significant temperature-by-season interactions were observed (Supplementary Table S2), future studies with repeated assessments across multiple seasons are needed to better distinguish temperature effects from broader seasonal behavioral patterns. Fifth, COVID-19 restriction periods were included as a binary covariate, which may not have fully captured differences in the scope and stringency of restrictions across phases of the pandemic. Therefore, some residual confounding may remain, particularly during colder periods that overlapped with COVID-19 restrictions. Sixth, estimates at the extremes of the ambient temperature range were based on relatively few observations. Finally, participants were relatively healthy and active older adults living in a region known for longevity, which may limit the generalizability of the present findings to more vulnerable or less active populations.

In conclusion, daily mean ambient temperature showed clear nonlinear associations with multiple 24-hour movement behaviors among community-dwelling older adults. An optimal ambient temperature of approximately 21°C was associated with reduced SB and increased LPA and MVPA, whereas sleep duration began to decline as ambient temperature exceeded approximately 22°C. These findings suggest that considering thermal conditions may help inform strategies to reduce SB and increase LPA and MVPA while maintaining appropriate sleep durations among older adults. Maintaining comfortable thermal environments and encouraging physical activity during thermally favorable conditions may help support healthier 24-hour movement behavior patterns in later life. As the climate continues to warm, consideration of ambient temperature may be useful when designing community-based interventions and public health strategies for older adults.

SUPPLEMENTARY MATERIALS

Supplementary materials can be found via https://doi.org/10.4235/agmr.26.0065.

Table S1.

Summary of daily meteorological data (n=4,344)

Table S2.

Sensitivity analyses examining the robustness of ambient temperature and 24-hour movement behavior associations