Exploring demographic-behavioural profiles in a large cohort of Japanese adults: a latent class analysis approach for 137 357 individuals

Abstract

Background

Understanding health behaviour patterns is critical to addressing public health challenges. In Japan, lifestyle factors such as smoking, physical inactivity, poor diet and insufficient sleep significantly contribute to non-communicable diseases (NCDs). Identifying how these behaviours cluster within populations offers valuable insights for effective prevention and intervention strategies. Latent class analysis (LCA) enables the identification of behavioural patterns to inform tailored public health initiatives.

Purpose

To identify and characterise demographic-behavioural profiles among Japanese adults using LCA.

Method

Data from 137 357 individuals participating in a health promotion programme in Japan (1 January 2022 to 31 December 2022) were analysed. Participants aged 20 years and older with complete records for all relevant health behaviour variables were included. Variables included age, gender, smoking status, alcohol consumption, sleep duration, physical activity, sedentary behaviour and salt intake. LCA was conducted to identify behaviour clusters, with the optimal number of classes determined using Bayesian and Akaike information criteria.

Results

From a cohort of 137 357 individuals, nine clusters were identified: (1) middle-aged women with short sleep duration (20.8%), (2) middle-aged men with elevated health behaviour risks (8.4%), (3) young adults with short sleep duration (18.4%), (4) healthy middle-aged men (2.6%), (5) sedentary middle-aged men with short sleep duration (25.3%), (6) typical middle-aged women (5.2%), (7) physically active young men with poor dietary habits (2.5%), (8) unhealthy middle-aged men with inconsistent behaviours (8.5%) and (9) older adults with moderate health concerns (8.4%).

Conclusion

LCA revealed diverse health behaviour patterns, emphasising the need for tailored health interventions targeting specific subgroups to mitigate NCD risks.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Previous studies have highlighted the clustering of health behaviours as a significant determinant of non-communicable disease (NCD) risk. However, these studies predominantly focus on high-risk individuals, and limited research exists on the broader patterns of health behaviours in the general Japanese population.

WHAT THIS STUDY ADDS

• This study provides a comprehensive characterisation of health behaviour patterns in Japan by leveraging a large cohort. Compared with previous work, our study includes factors such as sleep and sedentary behaviour and analyses a broader population beyond high-risk groups.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• This study highlights the importance of creating targeted public health programmes for specific groups to reduce NCD risks. Policymakers and practitioners can use these results to promote healthier behaviours like regular exercise and better diets. Researchers can build on this work by studying how these behaviour patterns affect health over time.

Introduction

Non-communicable diseases (NCDs), such as coronary heart disease, diabetes and hypertension, are leading causes of morbidity and mortality worldwide, and their prevalence is rising.¹ Modifiable lifestyle factors, such as smoking, physical inactivity, poor dietary habits and inadequate sleep, often influence these conditions. Addressing these risk factors is critical for preventing NCDs and reducing their global burden.² Furthermore, health behaviours do not occur in isolation; individuals often exhibit distinct patterns or clusters of behaviours that can synergistically increase health risks. Identifying these clusters can provide valuable insights into the combined impact of multiple behaviours on health outcomes and inform developing targeted public health interventions.³ For example, high levels of sedentary behaviour and poor dietary habits have been linked to increased risks of hypertension and metabolic syndrome, especially when they co-occur.⁴ Moreover, the simultaneous presence of multiple health risk behaviours amplifies the likelihood of developing NCDs and obesity.⁵

Japan has an ageing population and high life expectancy; thus, managing the burden of chronic diseases and promoting healthy ageing are critical public health priorities. Understanding how health risk behaviours cluster within the Japanese population is essential for identifying unique behavioural patterns that may differ from global trends due to cultural, demographic and societal factors.^6,7 More importantly, Japan serves as a critical case study for understanding the challenges of NCDs in the context of advanced demographic ageing.⁸ As one of the world’s first ‘super-aged’ societies, the health behaviour patterns observed in the Japanese population may act as a harbinger for trends that other developed nations in Europe, North America and Asia will face in the coming decades. Furthermore, while Japan boasts one of the highest life expectancies globally, a key public health challenge—shared by many other countries—is the extension of healthy life expectancy and the compression of morbidity.⁹ Understanding how modifiable lifestyle behaviours cluster is paramount to achieving this goal. Therefore, identifying distinct behavioural typologies within this context can provide valuable and transferable insights for designing targeted public health interventions in other high-income countries facing similar demographic and societal shifts. Using cluster analysis, we can identify latent patterns of health behaviours, enabling more nuanced interventions that address multiple risks simultaneously. This approach supports current public health strategies and highlights the need for tailored solutions that consider the combined effects of behaviour, environment and genetics in preventing NCDs.

The clustering of health behaviours, rather than their occurrence in isolation, is internationally recognised as a critical determinant of NCD risk.³ Latent class analysis (LCA) has emerged as a useful statistical method used to identify underlying behavioural patterns.¹⁰ In recent years, interest in the clustering of health behaviours has grown in Japan, leading to several studies exploring these patterns.^6,7 However, previous studies in Japan were often limited by a focus on specific high-risk or demographic groups or by relatively small sample sizes. A comprehensive characterisation of health behaviour patterns that includes contemporary lifestyle factors such as sleep duration and sedentary behaviour across a broad and diverse adult population has remained a research gap. Therefore, this study aimed to identify and characterise demographic-behavioural profiles among Japanese adults participating in a health promotion programme using LCA. It aims to provide a broader understanding of patterns across diverse groups, informing the development of targeted public health interventions. This study is characterised by its large cohort size, inclusion of factors such as sleep and sedentary behaviour, and analysis of a broader population beyond high-risk groups.

Methods

Study design and participants

This cross-sectional study utilised data from individuals enrolled in Japan’s Sumitomo Life Vitality programme between 1 January 2022 and 31 December 2022. Health check-up and behavioural data used in the analysis were drawn from records collected between January 2021 and December 2023. Initially developed by Discovery Health, the Vitality programme is a comprehensive wellness initiative designed to promote healthy lifestyle behaviours through incentives and behavioural economics principles.¹¹ In Japan, Sumitomo Life offers the programme to its life insurance policyholders nationwide as a wellness service. Participants in this study were policyholders residing in Japan who voluntarily enrolled in the programme and completed standardised health behaviour assessments via the Vitality mobile application. Participants completed the assessment at a time of their own choosing after enrolment, independently of any clinical visit or health guidance session.

Cohort timeline and validation cohort

Two distinct time windows are relevant to this study. The enrolment window refers to the period during which participants joined the programme, and the measurement window refers to the period during which their health check-up and behavioural data were collected. The primary cohort consisted of participants who enrolled during the enrolment window of 1 January 2022 to 31 December 2022, with health check-up and behavioural data drawn from records collected between January 2021 and December 2023. To enable an independent test of model stability, a validation cohort was constructed from participants who enrolled during the non-overlapping subsequent enrollment window of 1 January 2023 to 31 December 2023, with no participant overlap with the primary cohort. The validation cohort was used solely for the sensitivity analysis.

Eligibility criteria and missing-data exclusions

Participants were eligible for inclusion if they met specific criteria related to demographic, clinical and health behaviour data. All participants were required to have valid health check-up records, including age as of 1 April 2023; body mass index (BMI) or records of both height and weight; systolic and diastolic blood pressure and either fasting blood glucose or HbA1c levels. Participants responded to standardised health behaviour assessments, including average daily sleep duration, frequency of salty food consumption as an indicator of nutrition, levels of physical activity, daily sedentary behaviour, alcohol use and smoking status. The smoking status was determined through self-reported non-smoking declarations obtained from personal health records. Participants were excluded if they met any of the following conditions: (1) missing data for any of the lifestyle or demographic variables; (2) no record of age as of 1 April 2023; or (3) missing information on gender. These exclusion criteria were applied to ensure data completeness and analytical reliability.

Data quality exclusions and final analytical sample

In addition to the missing data exclusions described above, prespecified data quality exclusions were applied before LCA. Statistical outliers in the raw alcohol use, physical activity and sedentary behaviour variables were removed using the 1.5× IQR rule, retaining values between Q1−1.5 × interquartile range and Q3+1.5 × interquartile range. Sleep duration was restricted to values of ≥3 and <12 hours/day. Health check-up values outside predefined Ministry of Health, Labour and Welfare-based validity thresholds were also excluded. These data-quality exclusions were applied before categorisation and model estimation.

Health behaviour measurements

We analysed two demographic variables (gender and age) and six health behaviour variables (smoking, alcohol consumption, sleep duration, physical activity, sedentary behaviour and salt intake), which have been selected as key dimensions of health in previous studies.¹² Table 1 provides a detailed description. For the smoking item specifically, only participants who provided an explicit non-smoking declaration were classified as non-smokers; all other participants, including those who did not respond to this item, were classified as current smokers. The smoking variable was, therefore, treated as fully observed by design.

Table 1. Characteristics of demographic and health behaviour variables.

Variable	Description	Categories
Demographics
Gender	Participants were categorised based on self-report.	Female/male
Age	Participants’ age as of 1 April 2023.	<30/30–39/40–49/50–59/60+
Health behaviours
Smoking	Participants were classified based on their self-reported current smoking status. Non-smokers included both ex-smokers and never-smokers.	Current smoker/non-smoker
Alcohol consumption	Categorised based on self-reported daily frequency.	High (>1 drink/day)/low (≤1 drink/day)
Sleep duration	Self-reported average hours of sleep per day. Categorisation is based on Japanese public health guidelines.41	Short (<7 hours)/medium (≥7 & <9 hours)/long (≥9 hours)
Physical activity	Self-reported minutes of moderate-to-vigorous physical activity (MVPA) per week. Categorisation is based on WHO recommendations.42	Active (≥150 min)/inactive (<150 min)
Sedentary behaviour	Self-reported hours of sedentary activity per day. The cut-off is informed by recent meta-analyses.21	High (≥10 hours)/low (<10 hours)
Salt intake	Self-reported frequency of consuming salty foods, used as an indicator of dietary habits.	Usually or frequently/occasionally/rarely or never

All items asked about habitual behaviours without a specified recall window.

Statistical analysis

LCA was employed to identify clusters based on the participants’ health behaviours. LCA is a statistical technique to uncover latent classes within multivariate categorical data. This method is particularly suitable for analysing health behaviour variables because it accommodates categorical and dichotomous data types more effectively than other clustering methods, such as traditional cluster or factor analysis.¹³ Additionally, LCA offers flexibility by reducing arbitrariness in selecting clustering criteria.¹⁴ LCA is particularly informative for describing health behaviours,¹⁰ as it better accommodates these variables’ categorical nature.¹⁵ To determine the optimal number of latent classes, we sequentially estimated models ranging from 2 to 14 classes. Models were estimated using three random starts (nrep=3; tol=1e-10; maxiter=50 000), retaining the solution with the highest log-likelihood. To verify the stability of the selected solution, we additionally repeated the estimation with 50 random starts; 14 of 50 starts converged to the same maximum log-likelihood, supporting the reproducibility of the cluster structure. The selection of the final model was guided by multiple criteria: the Bayesian Information Criterion (BIC), Akaike Information Criterion (AIC) and entropy.¹⁶ For both AIC and BIC, lower values suggest a better model fit. BIC is the most reliable fit statistic and should be routinely reported.^17,18 Additionally, it is recommended that a model that minimises the BIC and/or AIC values be selected.¹⁹ Beyond statistical indices, considerations of interpretability and theoretical consistency were also incorporated into the model selection process.²⁰ This approach ensured that the chosen model would not only achieve statistical adequacy but also offer meaningful insights for subsequent analysis. All statistical analyses were conducted using R (V.4.4.1). For the LCA, all demographic and health behaviour variables were treated as categorical (factor-type) variables. Continuous variables such as age were converted into ordered categories (eg, ‘<30’, ‘30–39’, etc). Dichotomous variables, including smoking, alcohol consumption, physical activity and sedentary behaviour, were coded into two levels (eg, ‘current smoker’/‘non-smoker’, ‘active’/‘inactive’). Multilevel categorical variables included sleep duration (‘short’, ‘medium’, ‘long’) and salt intake frequency. The poLCA package in R, which was used for this analysis, treats all these factor variables as nominal, meaning no specific order is assumed in the model estimation. Missing data were handled through complete-case analysis; participants with missing values for any of the included variables were excluded prior to the analysis. The LCA was performed using the ‘poLCA’ package (V.1.1). Finally, to assess the robustness of our model, we conducted a sensitivity analysis by replicating the entire LCA procedure on the validation cohort described above. The same eligibility criteria, variable definitions, and LCA estimation procedure were applied. The primary aim of this sensitivity analysis was to assess the stability of the optimal number of latent classes; we did not perform an individual-level cross-validation, as the two cohorts comprised different participants. Validation was considered successful if (1) BIC reached its minimum at the same number of classes as in the primary analysis and (2) the resulting class profiles were qualitatively similar in terms of the dominant response patterns.

Patient and public involvement

As this study involved the secondary analysis of anonymised programme data, there was no direct patient or public involvement in the study design, conduct or dissemination plans.

Results

Participant selection and characteristics

The initial sample consisted of 812 658 individuals enrolled in the Vitality programme between 1 January 2022 and 31 December 2022. Of these, 778 845 were insured Vitality members, and 526 958 had valid health check-up records, including BMI or height and weight, systolic and diastolic blood pressure and either HbA1c or fasting blood glucose within the prespecified measurement window. Among individuals with valid health check-up records, 483 962 participants (91.8%) initiated the health behaviour questionnaire, and 217 651 participants (41.3% of those with valid health check-up records) provided valid responses to the required health behaviour items. Among questionnaire initiators, item-level missingness was 16.3% for sleep duration, 27.8% for physical activity, 30.7% for salt intake, 31.5% for sedentary behaviour and 39.4% for alcohol consumption; smoking status was treated as fully observed by design. After additionally requiring complete analytic records within the prespecified measurement window, 169 691 participants remained. Prespecified data quality exclusions, as described in the Methods section, were then applied sequentially. Statistical outlier exclusions for alcohol use, physical activity and sedentary behaviours left 141 365 participants. The sleep duration validity restriction left 141 125 participants. Finally, exclusion of participants with health check-up values outside predefined validity thresholds yielded the final analytical sample of 137 357 participants. Overall, data-quality exclusions removed 32 334 participants, corresponding to 19.1% of the complete analytic sample. The participants’ mean age was 47.07 years (SD: 11.83) and the largest group belonging to the 50–59 age range (30.5%), followed by the 40–49 age range (27.2%). Among the participants, 42.9% were female (n=58 855) and 57.1% were male (n=78 502). Table 2 provides detailed participant demographics and health behaviour characteristics.

Table 2. Baseline characteristics of participants.

	Level	Overall	Female	Male
N (%)		137 357	58 855 (42.9)	78 502 (57.1)
Age (%)	<30	12 813 (9.3)	6212 (10.6)	6601 (8.4)
	30–39	24 817 (18.1)	10 845 (18.4)	13 972 (17.8)
	40–49	37 395 (27.2)	15 828 (26.9)	21 567 (27.5)
	50–59	41 849 (30.5)	17 070 (29.0)	24 779 (31.6)
	60 over	20 483 (14.9)	8900 (15.1)	11 583 (14.8)
BMI (mean (SD))		23.53 (4.17)	22.54 (4.14)	24.27 (4.03)
Smoking (%)	Yes	11 477 (8.4)	2224 (3.8)	9253 (11.8)
	No	125 880 (91.6)	56 631 (96.2)	69 249 (88.2)
Alcohol consumption (%)	Low frequency	132 938 (96.8)	57 928 (98.4)	75 010 (95.6)
	High frequency	4419 (3.2)	927 (1.6)	3492 (4.4)
Average hours of sleep per day among participants (%)	Long	977 (0.7)	496 (0.8)	481 (0.6)
	Medium	48 678 (35.4)	21 389 (36.3)	27 289 (34.8)
	Short	87 702 (63.8)	36 970 (62.8)	50 732 (64.6)
Physical activity (%)	Active	23 949 (17.4)	8472 (14.4)	15 477 (19.7)
	Inactive	113 408 (82.6)	50 383 (85.6)	63 025 (80.3)
Frequency of consuming salty foods (%)	Frequently or usually	35 849 (26.1)	14 590 (24.8)	21 259 (27.1)
	Occasionally	57 270 (41.7)	25 151 (42.7)	32 119 (40.9)
	Rarely or never	44 238 (32.2)	19 114 (32.5)	25 124 (32.0)
Sedentary behaviour (%)	Low sedentary time	92 489 (67.3)	43 338 (73.6)	49 151 (62.6)
	High sedentary time	44 868 (32.7)	15 517 (26.4)	29 351 (37.4)

BMI, body mass index.

Model selection for LCA

Models with varying class numbers were evaluated using BIC and AIC to identify the optimal number of latent classes. The BIC value reached its minimum at the nine-class model, suggesting it as the most appropriate solution (figure 1). While AIC continued to decrease with additional classes, it did not indicate a clear minimum. In addition to these statistical indices, the interpretability, distinctiveness and theoretical plausibility of the class profiles were also evaluated. The nine-class model allowed us to distinguish health behaviour patterns across different demographic groups in a way that was easy to understand and apply. Therefore, the nine-class model was selected as the final solution to achieve the best balance between statistical fit, conceptual clarity and practical utility in describing participant health behaviours. The entropy of the nine-class model was 0.435, indicating moderate class separation, which is typical for LCA applied to multivariate behavioural data in large samples. Entropy values for the full range of 2–14 class solutions are reported in online supplemental table 1. Observed response distributions and demographic characteristics based on modal class assignment are presented in online supplemental table 2.

Figure 1. Comparison of AIC and BIC for model selection in latent class analysis. The plot displays the Akaike information criterion (AIC) and the Bayesian information criterion (BIC) for models with 2 to 14 classes. The BIC reaches its minimum at the nine-class model.

Model validation through sensitivity analysis

To confirm the robustness of the preliminarily selected nine-class model, we conducted a sensitivity analysis on an independent validation cohort of 110 086 individuals who enrolled in the programme during the non-overlapping subsequent enrollment window of 1 January 2023 to 31 December 2023. The entire LCA procedure was replicated on the independent validation cohort. This validation analysis provided confirmation for our model selection: the BIC again reached its minimum at nine classes, supporting the stability of the model structure (online supplemental table 1). In addition, the dominant response patterns of the nine-class solution were broadly similar between the primary and validation cohorts, supporting the reproducibility of the class profiles.

Identified groups’ descriptions from the LCA

Nine clusters were identified, each with distinct demographic and health behaviour profiles. A summary of these clusters is provided in table 3 and figure 2. Statistically significant differences were observed across the nine clusters for all analysed variables (all p<0.001). The detailed characteristics for each cluster are available in online supplemental table 2.

Table 3. Summary of demographic-behavioural profiles distributions across clusters.

Class	Gender	Age	Smoking	Alcohol	Sleep	Physical activity	Sedentary Behaviour	Salt intake	Summary
1	Female	40s, 50s	–	–	80% Short sleepers	–		–	Middle-aged women with short sleep duration
2	Mostly male	40s, 50s, 60s	30% smokers	12% high-frequency alcohol consumption	–	–	–	–	Middle-aged men with elevated health behaviour risk
3	60% female	20s, 30s	99.9% non-smokers	–	50% short sleepers	–	–	–	Young adults with short sleep duration
4	90% male	Various age groups, mainly 50s	98.7% non-smokers	–	–	Regular exercise	–	90% low salt intake	Healthy middle-aged men
5	Male	Mostly in 50s	–	–	80% short sleepers		60% sedentary	–	Sedentary middle-aged men with short sleep duration
6	Female	40s	–	–	–	–	–	–	Typical middle-aged women
7	Male	20s, 30s, 40s	98.5% non-smokers	–	–	Regular exercise	–	50% frequent salt intake	Physically active young men with poor dietary habits
8	Mostly male	40s	30% smokers	–	80% short sleepers	No exercise	Mostly short sedentary times	Rarely/never salty food consumption was uncommon	Unhealthy middle-aged men with inconsistent behaviours
9	Both genders	60+	–	–	60% short sleepers	–	Mostly short sedentary times	–	Older adults with moderate health concerns

Figure 2. Health behaviour distributions across the nine identified clusters. This figure illustrates the profile of each cluster by showing the distribution of demographic characteristics and health behaviours.

Cluster 1: middle-aged women with a short sleep duration (20.8%)

This cluster consists exclusively of women, primarily in their 40s and 50s, with a high prevalence (85.8%) of short sleep duration (<7 hours). This group includes 28 611 individuals and represents ‘middle-aged women with short sleep duration’. Despite otherwise moderate health behaviours, the significant sleep deprivation in this group may elevate their risk for cardiometabolic conditions, including hypertension, insulin resistance and mood disorders. This cluster may particularly benefit from sleep-focused behavioural interventions, such as improved sleep hygiene and work–life balance strategies tailored for midlife women.

Cluster 2: middle-aged men with elevated health behaviour risks (8.4%)

This cluster consists predominantly of men (98.1%), mainly aged 40–60. The group comprises 11 585 individuals, of whom approximately 30% are current smokers and 12% report daily alcohol consumption. These behaviours suggest sustained exposure to modifiable lifestyle risk factors. Although other health behaviour indicators in this cluster were average, the co-occurrence of smoking and frequent alcohol consumption may increase the risk of NCDs, particularly cardiovascular and hepatic conditions. Interventions that focus on smoking cessation and responsible alcohol consumption, especially in workplace-based or insurer-supported health promotion programmes, may be beneficial for this cluster of participants.

Cluster 3: young adults with short sleep duration (18.4%)

This cluster consists primarily of individuals in their 20s and 30s, with 62% being female. It includes 25 118 participants, nearly all of whom are non-smokers (99.9%), suggesting a health-conscious orientation regarding tobacco use. However, approximately 50% are classified as short sleepers, indicating a high prevalence of insufficient sleep in this demographic. This pattern may reflect lifestyle pressures typical of early adulthood, such as work demands or social routines that compromise sleep duration. Although other health behaviours are generally unremarkable, the widespread sleep deficit may pose long-term risks for metabolic, cognitive and psychological health. This cluster could benefit from preventive interventions aimed at promoting sleep hygiene and stress management, particularly through digital health tools or workplace wellness programmes.

Cluster 4: healthy middle-aged men (2.6%)

The cluster designated ‘healthy middle-aged men’ comprises 3538 individuals, predominantly male (86.9%), with a notable concentration in their 50s. All members reported regular physical activity, and 90% reported infrequent consumption of salty foods. These findings imply the adoption of health-conscious behaviours, which may contribute to favourable cardiometabolic profiles. The combination of regular physical activity and a reduced salt intake is of particular relevance in the prevention of hypertension and cardiovascular disease (CVD), both of which become increasingly prevalent with advancing age. This cluster may reflect individuals who are proactively managing their health, thus rendering them a potential model population for workplace wellness strategies or community-based preventive health programmes.

Cluster 5: sedentary middle-aged men with short sleep duration (25.3%)

The cluster designated ‘sedentary middle-aged men with short sleep duration’ comprises 34 761 individuals, primarily men in their 50s. The results of the study indicate that approximately 88.5% of the participants reported a short sleep duration, and 65.9% engaged in prolonged sedentary behaviour. This combination of inadequate sleep and physical inactivity constitutes a worrisome health risk profile, as both behaviours have been independently associated with elevated risks of CVD, obesity and all-cause mortality. The co-occurrence of these two detrimental behaviours in a single cluster emphasises the necessity for targeted interventions, particularly in workplace settings where sedentary time is frequently unavoidable. It is recommended that this cluster benefits from multicomponent strategies that address both sleep hygiene and physical activity.

Cluster 6: typical middle-aged women (5.2%)

The cluster designated ‘typical middle-aged women’ comprises 7186 individuals, all women, predominantly in their 40s. This group did not exhibit strongly distinctive health behaviours, representing a broadly average profile for women in this age group. Most participants reported moderate levels of physical activity and sedentary time, along with a generally healthy lifestyle. However, dietary behaviours showed some variability, particularly in salt intake frequency. The absence of marked risk behaviours suggests that this cluster may not be a high-priority target for intensive intervention but could benefit from continued support for health maintenance, including dietary counselling and physical activity promotion to prevent age-related health decline.

Cluster 7: physically active young men with poor dietary habits (2.5%)

The cluster designated ‘physically active young men with poor dietary habits’ comprises 3369 individuals, predominantly men in their 20s to 40s. Nearly all members reported engaging in regular physical activity, suggesting a favourable behaviour in terms of cardiometabolic health. However, approximately 50% of participants reported frequent salt intake, indicating a tendency towards suboptimal dietary practices. This combination indicates a discrepancy between physical activity and nutritional habits. While the regular practice of exercise has been demonstrated to offer a degree of protection, the consumption of a diet high in salt has been shown to potentially nullify these benefits by increasing the risk of hypertension and other cardiovascular conditions. This cluster underscores the importance of integrated health education that emphasises the complementary role of both exercise and nutrition.

Cluster 8: unhealthy middle-aged men with inconsistent behaviours (8.5%)

The cluster designated ‘unhealthy middle-aged men with inconsistent behaviours’ comprises 11 666 individuals, predominantly men in their 40s. Approximately 80% of participants reported short sleep duration, 29% were smokers and none engaged in regular physical activity. The majority reported low sedentary time, creating a discordant health behaviour profile. The presence of multiple risk factors, particularly physical inactivity and insufficient sleep, suggests increased vulnerability to cardiometabolic disease. Tailored interventions for this cluster should consider both the negative and potentially modifiable aspects of their behaviour profile, particularly emphasising sleep improvement and promoting feasible physical activity routines.

Cluster 9: older adults with moderate health concerns (8.4%)

The cluster designated ‘older adults with moderate health concerns’ comprises 11 523 individuals almost all aged 60 years and older, with an even gender distribution (53.6% female, 46.4% male). Approximately 60% of the participants reported short sleep duration, while the majority had relatively short sedentary time. Although this group did not exhibit extreme health behaviour risks, the prevalence of insufficient sleep suggests a need for supportive interventions aimed at promoting sleep hygiene and sustaining physical function. This cluster may reflect a population segment transitioning into older adulthood, where moderate health concerns begin to emerge.

Discussion

Our study used LCA on a large dataset of 137 357 Japanese adults and identified nine distinct demographic-behavioural profiles. Because age and gender were included alongside behavioural variables in the LCA model, several of the resulting profiles reflect demographic structure as well as behaviour (eg, older adults, middle-aged women), rather than behaviour alone. We retained this approach because age-specific and gender-specific behavioural patterns are central to designing targeted public health interventions in Japan’s demographically diverse adult population. The large sample size is a key strength of our study, allowing us to see a more detailed picture of lifestyle patterns than smaller studies could provide. A key finding is that people’s health risks often come from specific combinations of behaviours, not just single factors. First, short sleep was a common found in many different groups, such as middle-aged women (cluster 1), young adults (cluster 3) and sedentary middle-aged men (cluster 5). Second, we found groups with mixed health behaviours, like young men who exercise but have poor dietary habits (cluster 7). Identifying these specific groups, which was facilitated by our large-scale analysis, suggests the value of looking beyond simple labels like ‘healthy’ or ‘unhealthy’. Our findings suggest that public health efforts should focus on the whole pattern of behaviours for each subgroup.

This study identified several distinct clusters, including groups characterised by prolonged sedentary behaviour and inadequate sleep duration. Both sedentary behaviour and short sleep duration have been associated with various adverse health outcomes, including increased risks of hypertension, diabetes and other NCDs.^21,22 Cluster 5 had particularly high levels of sedentary time with short sleep duration, while cluster 1 had short sleep duration and cluster 8 had short sleep duration but predominantly short periods of sedentary time. Although this study did not find a direct link to cardiometabolic outcomes, the behaviours observed in cluster 5, such as insufficient sleep and prolonged sedentary behaviour, are consistent with known risk factors for cardiometabolic disease. For example, a study found that short sleep duration increased the risk of type 2 diabetes. Although a healthy diet reduced the overall risk, the association between short sleep and diabetes remained, emphasising the importance of sleep in prevention efforts.²³ A meta-analysis demonstrated a U-shaped relationship between sleep duration and diabetes risk, with both short and long sleep durations associated with higher risks compared with 7–8 hour of sleep per day.²⁴ Evidence also shows that the interplay between sedentary behaviour, physical activity and cardiorespiratory fitness significantly influences cardiometabolic risk. Recent study suggests that not only the total amount but also the pattern of sedentary time impacts cardiometabolic health. A randomised trial demonstrated that interrupting prolonged sitting every 30 to 60 min with short walks was effective in acutely lowering both blood pressure and postmeal glucose levels, highlighting a clear, actionable strategy to counteract the risks of a sedentary lifestyle.²⁵ These findings suggest the importance of reducing sedentary behaviour (in cluster 5) and addressing inadequate sleep duration (in clusters 1, 5 and 8) to reduce disease risk factors.

Cluster 4, characterised by high physical activity and low salt intake, exemplifies well-documented behaviours supporting cardiovascular health. Regular physical activity has been consistently associated with reduced risks of hypertension, obesity and type 2 diabetes, all of which are key risk factors for CVDs. A meta-analysis demonstrated that individuals who engage in at least 150 min of moderate-intensity physical activity per week experience significant reductions in blood pressure and improvements in lipid profiles, which are critical for maintaining cardiovascular health.²⁶ Similarly, adherence to a low-sodium diet has been linked to reductions in systolic and diastolic blood pressure, particularly among individuals at high risk of hypertension. A review demonstrated a clear relationship between sodium reduction and lowering blood pressure, with more significant reductions observed in populations with higher baseline blood pressure, older individuals and non-white populations.²⁷ While the direct relationship between sodium intake and blood pressure was not measured in this study, identifying this cluster suggests the potential for tailored interventions targeting similar behaviours. For instance, promoting low-sodium diets in combination with regular physical activity may yield substantial benefits, particularly for high-risk populations such as older adults or those with preexisting hypertension.

Smoking is a well-established risk factor for numerous chronic conditions, including CVD, chronic obstructive pulmonary disease and certain types of cancer. Among the clusters identified in this study, smoking behaviours were particularly evident in clusters 2 and 8, where a significant proportion of individuals were current smokers. These findings are consistent with the existing research, showing a clear link between smoking and adverse cardiometabolic outcomes, including elevated blood pressure, insulin resistance and systemic inflammation.^28–30 In contrast, clusters 3, 4 and 7 exhibited lower smoking prevalence, suggesting healthier behavioural patterns that align with reduced health risks. These clusters highlight the potential benefits of smoking cessation as part of broader lifestyle interventions targeting cardiovascular and metabolic health. A previous study indicated that smoking cessation, even among individuals with a mild history of smoking, leads to significant improvements in vascular function and a reduced risk of heart disease and stroke. A study found that the CVD risk of individuals who quit smoking rapidly declined, with the risk of coronary heart disease decreasing sharply within 1 to 2 years after cessation.²⁸ Additionally, quitting smoking reduces markers of inflammation and hypercoagulability, improves high-density lipoprotein cholesterol levels and slows the progression of atherosclerosis, all contributing to a lower risk of CVD.³¹ In addition to individual health impacts, smoking also imposes substantial economic burdens through increased healthcare costs and reduced productivity.³² Targeted smoking cessation programmes for high-risk clusters, such as clusters 2 and 8, could not only improve individual health outcomes but also alleviate the societal costs associated with smoking-related diseases.³³ Early intervention among younger individuals, particularly within high-risk clusters, may prevent the initiation of smoking and its subsequent health consequences. Evidence suggests that antismoking campaigns and policies targeting youth are particularly effective in reducing smoking prevalence.³⁴ Effective smoking cessation interventions often combine behavioural counselling, pharmacological aids and social support.³⁵

Practical implications for tailored interventions

The significance of identifying these clusters lies in the ability to tailor public health interventions. By understanding the unique characteristics of each cluster, targeted interventions can be designed to address the specific needs of different subgroups, moving beyond one-size-fits-all approaches. Our findings suggest distinct strategies for different high-risk groups. For cluster 5, characterised by sedentary middle-aged men with short sleep, a workplace wellness programme could promote physical activity and integrate sleep hygiene education.³⁶ For clusters 2 and 8, where smoking was prevalent among middle-aged men, targeted smoking cessation programmes combining counselling and pharmacological aids could be considered.³⁵ Similarly, community-based interventions focusing on nutritional education, such as reducing salt intake, could be relevant for cluster 7 (‘physically active young men with poor dietary habits’).³⁷ Such interventions may contribute to reducing individual health risks and alleviating the broader societal burden of NCDs. Furthermore, our results highlight opportunities for early, preventive interventions. For example, public health campaigns for cluster 3 (‘young adults with short sleep duration’) could emphasise the long-term importance of adequate sleep, may help reduce the future risk of metabolic syndrome.^34,38 The ability to focus on these groups with tailored strategies presents a valuable opportunity to prevent health risks from escalating and improve overall population health.

Limitations and strengths

This study has several limitations to acknowledge. First, the study’s data were derived from participants enrolled in the Sumitomo Life Vitality programme, which may only partially represent the broader Japanese adult population. We compared the demographic and behavioural profile of our cohort with publicly available national statistics. Our cohort had a higher proportion of men (57.1%) than the Japanese adult population (approximately 49% male, based on the 2020 National Census)³⁹ and was concentrated in working ages (mean age 47.1 years; 76% aged 30–59), with limited representation of adults aged 65 years and over. Smoking prevalence in our cohort (8.4% overall; 11.8% in men and 3.8% in women) was lower than that reported in the 2019 National Health and Nutrition Survey (16.7% overall; 27.1% in men and 7.6% in women),⁴⁰ consistent with self-selection of more health-conscious individuals into a wellness programme. Prefecture-level geographic distribution data were not available to us. These comparisons suggest that our findings primarily characterise working-age individuals who voluntarily participate in a nationwide insurer-linked wellness programme. Therefore, these results may not necessarily apply to the broader Japanese adult population. Beyond this self-selection at the programme level, item-level missingness within the questionnaire was the principal source of further participant loss, ranging from 16.3% (sleep duration) to 39.4% (alcohol consumption) among questionnaire initiators. If non-response is correlated with the behaviours being measured, the prevalence of high-risk profiles in our clusters may be underestimated. The clusters should, therefore, be interpreted as patterns observed among engaged respondents. Second, some health behaviour data relied on self-reported measures, such as physical activity and dietary habits, subject to recall and social desirability biases. These biases may have affected the reported behaviours’ accuracy and, consequently, the clustering results. For example, people may tend to over-report healthy habits like physical activity and under-report unhealthy ones like high salt intake. If this happened in our study, we might be underestimating the true level of risk in our clusters. This means the ‘healthier’ groups could be less healthy than they appear, and the ‘unhealthier’ groups could be at an even higher risk. Future research using objective measures, like activity trackers, would be helpful to confirm our findings. Third, although the health behaviour variables were selected based on prior literature and public health guidelines, the measurement instruments themselves were not subjected to formal psychometric validation. This may limit the interpretability of behavioural constructs at a psychometric level. Fourth, the LCA results depended on the selected model parameters and assumptions, including the number of clusters. While model selection criteria, such as the BIC, were used to determine the optimal number of clusters, the identified groupings may vary with different analytical approaches or alternative datasets. However, we argue that the core thematic patterns identified, such as the pervasiveness of short sleep across diverse groups and the existence of discordant behaviours, likely represent robust underlying phenomena that would emerge regardless of the specific clustering method used.

Despite these limitations, this study has several strengths. First, it applied LCA to an exceptionally large and diverse cohort of 137 357 participants. This large-scale approach offers several key advantages. First, the statistical power of our large sample enhances the reliability and stability of the identified clusters. Second, this scale enables us to detect smaller yet distinct subgroups, such as ‘healthy middle-aged men’ (2.6%) and ‘physically active young men with poor dietary habits’ (2.5%), which might be overlooked in smaller studies. Although the cohort was not nationally representative, its large nationwide scale enhances the practical relevance of the findings for insurer-linked and working-age health promotion settings.

Conclusion

This study identified nine distinct demographic-behavioural profiles in a large Japanese cohort, highlighting the need to move beyond single risk-factor models in public health. Our primary recommendation is for a strategic shift from one-size-fits-all policies to tailored interventions targeting the characteristics of these subgroups. In particular, greater attention to combinations of behaviours—such as the co-occurrence of poor sleep and sedentary behaviour—and to groups with discordant habits may be warranted. Future research should examine long-term health outcomes and assess whether profile-based approaches can inform NCD prevention strategies.

Data availability statement

No data are available.