Occupational (labour-type) physical activity (OPA). OPA was assessed using a standardised examination item: “Do you regularly engage in physical exercise or physical activity (including walking, farming, or domestic labour) for ≥30 min each time?” Participants reported the number of such activity sessions per week (sessions/week). This variable was treated as continuous in all primary analyses. For descriptive statistics, OPA was categorised as inactive (0 sessions/week), low (1–2 sessions/week), and moderate-to-high (≥3 sessions/week), following cut-offs consistent with national recommendations for older adults and rural activity profiles in China (33). For sensitivity and joint-exposure analyses, OPA was further operationalised as an ordinal variable and a binary indicator distinguishing lower-frequency (0–2 sessions/week) from high-frequency (≥3 sessions/week) activity. These categorical thresholds were specified a priori to enhance interpretability while maintaining adequate cell sizes for stable estimation in this rural cohort. In addition, the substantive inferences were verified using continuous (per-session) models and alternative operationalisations (ordinal and binary) in sensitivity and joint-exposure analyses, yielding materially consistent conclusions.

Alcohol consumption. Alcohol use was evaluated using self-reported drinking frequency (times/week) and beverage type. In this cohort, intake predominantly consisted of high-proof Chinese baijiu (a traditional distilled grain spirit; commonly ~35–60% alcohol by volume [ABV]) consumed within social contexts. Drinking frequency over the preceding 12 months was categorised as never/rare (<1 episode/month), occasional (1–3 episodes/month), and weekly (≥1 occasion/week), adhering to standard epidemiological definitions. In primary models, drinking frequency was treated as a continuous exposure (per +1 occasion/week). For joint-exposure and PAF analyses, drinking was additionally operationalised as none versus any weekly drinking (>0/week) to support interpretable contrasts and ensure robust estimation in fully adjusted models. When necessary, intake was approximated as grams of ethanol using typical conversions for baijiu, assuming ~50 mL per occasion at ~50% alcohol by volume (≈20 g ethanol per occasion) (34).

The primary outcome was hypertension, defined as higher-arm systolic blood pressure (SBP) ≥ 140 mmHg and/or diastolic blood pressure (DBP) ≥ 90 mmHg. This definition aligns with Chinese, European, and International Society of Hypertension guidelines and applies the higher-arm principle to enhance epidemiological accuracy (35). Antihypertensive medication status was recorded and reserved for sensitivity analyses.

Secondary cardiometabolic outcomes included dyslipidaemia (TG ≥ 2.3 mmol/L and/or HDL-C < 1.0 mmol/L) (36, 37), impaired fasting glucose (FBG ≥ 6.1 mmol/L) (37, 38), and obesity (BMI ≥ 28 kg/m²) (38). Continuous markers (SBP, DBP, TG, LDL-C, HDL-C, FBG, and BMI) were also analysed as secondary outcomes. An exploratory composite cardiometabolic risk score (0–4) was created for descriptive purposes but was not used in primary models (Tables 1–7).

All physiological assessments followed standardised national protocols for older adults health examinations (39). Bilateral blood pressure was measured after at least 5 min of seated rest using a calibrated Omron HBP-1300 professional electronic sphygmomanometer, validated for use in Chinese populations. Measurements were obtained at heart level with an appropriately sized cuff, and higher-arm values were utilised for primary analyses (35). Height and weight were recorded using a certified stadiometer and electronic scale with participants in light clothing and no shoes, from which body mass index (BMI) was calculated.

Fasting venous blood samples were collected after a minimum 8-h fast and transported under cold-chain conditions for processing at the Wangkui County Central Clinical Laboratory. The biochemical panel included triglycerides, HDL-C, LDL-C, fasting blood glucose, haemoglobin, and routine blood parameters. Automated analysers were operated with daily internal quality control and external proficiency testing, consistent with provincially accredited quality-control processes. Duplicate assays in 10% of participants demonstrated high reliability, with intraclass correlation coefficients exceeding 0.95. Resting heart rate, recorded immediately after blood pressure measurement, was winsorised at the 1st and 99th percentiles to mitigate the influence of extreme values (40).

Multivariable models were adjusted for a pre-specified set of clinically relevant covariates: age, sex, BMI, haemoglobin concentration, and winsorised resting heart rate (40). This core adjustment set was selected a priori to address major demographic and physiological confounders while ensuring model parsimony. Additional factors (e.g., dietary habits, smoking, and medication use) were incorporated into supplementary analyses (Section 3.6) to evaluate the robustness of the primary findings without materially expanding the main models.

Initial data inspection assessed completeness, physiological plausibility, and distributional characteristics using descriptive statistics, histograms, Q–Q plots, and Shapiro–Wilk tests. Implausible physiological values were screened against age-appropriate clinical reference ranges (42). Given that the clinical dataset had already undergone routine quality checks by local health authorities, no additional exclusions were applied beyond these pre-specified criteria. Resting heart rate occasionally exhibited extreme values and was therefore winsorised at the 1st and 99th percentiles to mitigate undue leverage on regression estimates. All other continuous variables (SBP/DBP, triglycerides, HDL-C, LDL-C, fasting glucose, and BMI) were analysed on their original scale in primary models, with alternative transformations or truncation rules examined only in sensitivity analyses.

Missingness in core analytic variables (blood pressure, OPA, alcohol use, BMI, haemoglobin, and resting heart rate) was less than 5%. Consequently, descriptive and prevalence analyses (Section 3.1) utilised the full sample (N = 2,270), whereas primary regression models (Sections 3.2–3.7) employed complete-case data (N = 2,194) to ensure consistent covariate adjustment. Multiple imputation was not performed because the low missingness rate allowed complete-case analysis to provide a transparent and consistent modelling sample; robustness was further supported by sensitivity analyses yielding estimates consistent with the primary results.

The primary outcome was hypertension, defined as higher-arm SBP ≥ 140 mmHg and/or DBP ≥ 90 mmHg, consistent with guideline recommendations (43, 44). Antihypertensive medication status was recorded but was incorporated only in supplementary analyses rather than the primary outcome definition. The main exposures were OPA frequency (sessions/week) and alcohol drinking frequency (occasions/week), both modelled as continuous variables with effects interpreted per +1 unit/week. In this rural context, reported physical activity predominantly reflected non-volitional occupational workload (e.g., farming and domestic labour) rather than leisure-time exercise, and was interpreted accordingly.

Associations between behavioural exposures and hypertension were estimated using multivariable logistic regression with HC3 heteroscedasticity-consistent standard errors (45). Two-sided p-values were reported and statistical significance was defined a priori as p < 0.05. All primary models adjusted for a pre-specified covariate set selected a priori based on biological relevance and prior evidence: age, sex, BMI, haemoglobin concentration, and winsorised resting heart rate. No automated variable selection procedures were used to avoid data-driven model specification. Adjusted odds ratios (aORs) with 95% confidence intervals were reported. The population-attributable fraction (PAF) for selected dichotomised exposures was estimated using Levin’s formula (Section 3.4).

Model stability and fit were evaluated using standard diagnostics. Multicollinearity was assessed with variance inflation factors (VIFs). Calibration was examined using the Hosmer–Lemeshow test based on deciles of predicted risk (g = 10). Influential observations were assessed using Cook’s distance, and linearity of the logit for continuous predictors was examined using Box–Tidwell-type log–product terms where applicable.

Dose–response relationships were evaluated by modelling OPA and alcohol frequency as continuous exposures. To assess potential nonlinearity, polynomial terms and restricted cubic splines were fitted and compared against linear specifications using likelihood ratio tests. To preserve interpretability and reduce overfitting risk, linear functional forms were retained unless model-comparison tests provided evidence of a materially superior fit for nonlinear specifications.

Combined exposure patterns were evaluated using a joint-exposure approach in which OPA was dichotomised as lower-frequency (0–2 sessions/week) versus high-frequency (≥3 sessions/week), and alcohol use as non-drinking (0 occasions/week) versus regular drinking (≥1 occasion/week). Four mutually exclusive groups were constructed, with the lower-frequency OPA/non-drinking group specified a priori as the referent. Effect heterogeneity was explored through pre-specified subgroup analyses stratified by sex, age (<70 vs. ≥70 years), and BMI category (<24, 24–28, and ≥28 kg/m²). Stratified models used the same adjustment set as the primary model, except that the stratifying variable was not additionally adjusted for in its corresponding stratum, and HC3 standard errors were applied throughout to maintain comparability.

A comprehensive set of sensitivity analyses was conducted to evaluate the internal validity of the primary findings. We examined the robustness of the estimates to alternative hypertension definitions (mean-arm ≥140/90 mmHg and higher-arm ≥130/80 mmHg) and to the exclusion of participants reporting current antihypertensive medication use. We also assessed whether results were sensitive to outlier-handling strategies for behavioural exposures (winsorisation and upper-tail trimming) and to covariate specification using nested adjustment sets (minimal, physiological, and fully adjusted). Additional models included SBP-only and DBP-only outcome definitions, linear models for continuous SBP and DBP, and multiplicative interaction testing between OPA and alcohol frequency. Across these analyses, the direction and magnitude of associations were materially unchanged, supporting the robustness of the primary conclusions.

Quadratic and cubic terms were added to the PA model. The likelihood ratio test (LRT) comparing the linear and polynomial models indicated no improvement in model fit (χ² = 1.31, df = 1, p = 0.25), suggesting no evidence for a J-shaped, U-shaped, or threshold pattern.

The linear dose–response association remained significant. Each additional weekly PA session was associated with 23% higher odds of hypertension (adjusted OR = 1.23, 95% CI: 1.16–1.32; p < 0.001).

The predicted linear trend and 95% CI band are displayed in Figure 4A.

Similarly, polynomial models for alcohol drinking frequency showed no meaningful improvement over the linear specification (LRT χ² = 0.20, df = 1, p = 0.66).

Each additional weekly drinking occasion was associated with 20% higher odds of hypertension (adjusted OR = 1.20, 95% CI: 1.04–1.40; p = 0.014).

The estimated linear trend is illustrated in Figure 4B.

Together, these analyses support approximately linear, monotonic exposure–response relationships for both labour-type physical activity and alcohol frequency. Hypertension risk increased steadily with higher LPA frequency, without evidence of a protective range at lower exposure levels. Alcohol frequency showed a similarly monotonic gradient, with no detectable plateau.

Shaded areas represent 95% confidence intervals generated from logistic regression models with HC3 robust standard errors. All models were adjusted for age, sex, BMI, haemoglobin concentration, and winsorized resting heart rate. Nonlinearity tests (polynomial LRTs) indicated no significant deviation from linearity for either exposure (Figure 5).

High-frequency physical activity (≥3 sessions per week)—which in this context largely reflects manual agricultural or domestic labour—was highly prevalent, with 59.4% of Participants reporting such activity. Compared with lower activity levels, high-frequency physical activity was associated with a substantially higher risk of hypertension (adjusted OR = 1.90, 95% CI: 1.57–2.29).

The corresponding PAF was 34.8%, indicating that approximately one in three hypertension cases could theoretically be prevented if heavy, non-recoverable physical labour were replaced with safer and more sustainable activity patterns.

Regular drinking—defined operationally as any weekly drinking (>0 drinking sessions per week)—was reported by 7.9% of Participants. Despite its relatively low prevalence, habitual weekly alcohol consumption was associated with an elevated risk of hypertension (adjusted OR = 1.62, 95% CI: 1.11–2.36).

The corresponding PAF was 4.7%, suggesting that approximately one in 20 hypertension cases in this population may be attributable to weekly alcohol use.

Joint exposure to both high-frequency physical activity and regular drinking was present in 5.7% of Participants. This combined exposure was associated with a higher risk of hypertension (adjusted OR = 1.71, 95% CI: 1.11–2.67).

The corresponding PAF was 3.9%, reflecting an additional but modest population-level burden arising from the co-occurrence of heavy occupational workload and habitual drinking.

Collectively, the PAF estimates suggest that a substantial proportion of hypertension cases in this rural older adults population may be attributable to heavy occupational labour and habitual weekly drinking. Importantly, in this setting, “physical activity” represents physically demanding, non-volitional labour, rather than structured, moderate-intensity exercise typically promoted in clinical guidelines.

Public health strategies should therefore emphasise workload modulation (e.g., mechanisation, task redistribution, rest scheduling) and alcohol-use reduction, rather than simply encouraging higher overall volumes of physical activity.

Bilateral blood pressure measurements showed excellent completeness, with missingness <0.2% for all four arm-specific variables. Left–right differences were small and symmetric, with a mean (±SD) of −0.6 ± 8.1 mmHg for systolic blood pressure (SBP) and −0.4 ± 4.9 mmHg for diastolic blood pressure (DBP). These findings indicate no systematic arm-specific measurement bias and support the use of the higher-arm reading as the primary criterion for hypertension classification, consistent with AHA and BHS recommendations.

To assess whether the associations between behavioural factors (weekly physical activity frequency and drinking frequency) and hypertension were sensitive to outcome definition, multivariable logistic regression models were re-estimated under three commonly used blood pressure thresholds. All models were adjusted for age, sex, BMI, haemoglobin concentration, and winsorized resting heart rate.

When hypertension was defined as higher-arm BP ≥ 140/90 mmHg (primary definition), physical activity frequency remained strongly associated with hypertension (OR = 1.23, 95% CI: 1.16–1.32), and drinking frequency also showed a positive association (OR = 1.20, 95% CI: 1.04–1.40). Model calibration was acceptable (Hosmer–Lemeshow p = 0.49).

Using the mean of left- and right-arm readings (mean-arm ≥140/90 mmHg) produced very similar effect estimates (physical activity: OR = 1.20, 95% CI: 1.13–1.27; alcohol: OR = 1.21, 95% CI: 1.05–1.39), with good calibration (Hosmer–Lemeshow p = 0.41).

When a lower diagnostic threshold of higher-arm BP ≥ 130/80 mmHg was applied, the association with physical activity persisted (OR = 1.17, 95% CI: 1.05–1.29), whereas the alcohol–hypertension association was weaker and not statistically significant (OR = 1.17, 95% CI: 0.91–1.50). Calibration remained acceptable (Hosmer–Lemeshow p = 0.34).

Taken together, these analyses show that physical activity frequency is a stable predictor of hypertension across multiple blood pressure definitions (OR range: 1.17–1.23). The association with drinking frequency is directionally consistent but more sensitive to the specific threshold used, becoming non-significant at the lower 130/80 mmHg cut-off. Acceptable goodness-of-fit across all models supports the internal validity of these findings.

First, three progressively adjusted logistic regression models were estimated to examine whether the associations of weekly labour-type physical activity (LPA) frequency and drinking frequency with hypertension were sensitive to the choice of covariates.

In the minimally adjusted model including age, sex, and BMI, both exposures showed positive associations with hypertension (OOPA: OR = 1.26; alcohol: OR = 1.23). After adding haemoglobin concentration and winsorized resting heart rate (physiological covariates), the effect sizes were slightly attenuated but remained very similar (OPA: OR = 1.23; alcohol: OR = 1.20). The final multivariable model, which corresponds to the primary specification used in Section 3.2, produced identical estimates (OOPA: OR = 1.23; alcohol: OR = 1.20).

The close agreement of odds ratios across all three models indicates that the observed associations are not artefacts of covariate selection and are unlikely to be explained by either omitted-variable or overadjustment bias.

To further address potential treatment-related bias, the final multivariable model was re-estimated after excluding all participants who reported current use of antihypertensive medications (N = 1,895). In this restricted sample, the associations between behavioural exposures and hypertension remained directionally consistent and of similar magnitude (OOPA: OR = 1.20; alcohol: OR = 1.24).

These findings suggest that the positive associations of OPA frequency and drinking frequency with measured hypertension are not driven by medication-related suppression of blood pressure.

Across all sensitivity analyses—minimal adjustment, addition of physiological covariates, full adjustment, and exclusion of treated individuals—the associations between behavioural exposures and hypertension were stable in both direction and magnitude. Additional checks using winsorized OPA and alcohol frequencies and excluding the top 5% of exposure values yielded nearly identical odds ratios (data not shown), further confirming the robustness of the main findings.

Clear sex differences were observed. Among men, both OPA and alcohol drinking frequency were positively associated with hypertension (OOPA: OR = 1.27, 95% CI: 1.16–1.40; alcohol: OR = 1.25, 95% CI: 1.07–1.47). Among women, OPA remained a significant predictor (OR = 1.20, 95% CI: 1.10–1.31), whereas alcohol consumption showed no significant association with hypertension (OR = 0.76, 95% CI: 0.48–1.21). These patterns suggest that the detrimental effect of alcohol on blood pressure is largely confined to men, while the impact of LPA is present in both sexes but more pronounced in men.

A marked age gradient was also evident.

In Participants aged ≥70 years, OPA frequency showed a strong and statistically significant association with hypertension (OR = 1.32, 95% CI: 1.22–1.44), whereas alcohol use was not significantly related to hypertension (OR = 1.12, 95% CI: 0.92–1.37).

Conversely, among Participants aged <70 years, alcohol had a stronger effect (OR = 1.30, 95% CI: 1.04–1.63), while the association between OPA and hypertension was weaker and not statistically significant (OR = 1.08, 95% CI: 0.97–1.20).

Taken together, these findings indicate that alcohol contributes more to hypertension risk in the younger older adults, whereas occupational workload becomes the dominant behavioural risk factor among adults aged ≥70 years.

Substantial heterogeneity by BMI category was observed.

Among individuals with BMI < 24 kg/m², both exposures were strong predictors of hypertension (OOPA: OR = 1.28, 95% CI: 1.17–1.41; alcohol: OR = 1.37, 95% CI: 1.05–1.78). In the 24–28 kg/m² group, associations persisted but were weaker, with OPA remaining significant (OR = 1.20, 95% CI: 1.09–1.34) and alcohol showing a non-significant trend (OR = 1.16, 95% CI: 0.94–1.42).

Among Participants with BMI ≥ 28 kg/m², neither OPA nor alcohol was significantly associated with hypertension (OOPA: OR = 1.03, 95% CI: 0.84–1.26; alcohol: OR = 1.00, 95% CI: 0.66–1.49).

These results suggest that leaner individuals (BMI < 24 kg/m²) are Particularly susceptible to both occupational physical strain and alcohol-related increases in blood pressure, whereas these behavioural risks appear attenuated in those with higher BMI, potentially reflecting adiposity-related buffering or survivor bias.

Across stratified analyses, three consistent patterns emerged. First, alcohol-related hypertension risk was concentrated in men, adults aged <70 years, and leaner individuals (BMI < 24 kg/m²). Second, the association between labour-type physical activity and hypertension strengthened with age and was most pronounced among participants aged ≥70 years. Third, among individuals with BMI ≥ 28 kg/m², neither labour-type physical activity nor alcohol drinking frequency was significantly associated with hypertension. Collectively, these findings indicate meaningful effect heterogeneity and underscore the need for population-specific hypertension prevention strategies that account for sex, age, and adiposity profiles.

HC3-robust linear regression models showed no statistically significant associations between weekly labour-type physical activity (LPA) frequency, alcohol drinking frequency, and fasting lipid Parameters after covariate adjustment. Although small coefficients were observed for TG, LDL-C, and HDL-C, none reached conventional significance thresholds (all p ≥ 0.05).

These findings indicate that in this rural older adults cohort—where “physical activity” largely reflects heavy manual labour—variation in OPA or drinking frequency does not translate into systematic alterations in TG, LDL-C, or HDL-C at the population level. In other words, the behavioural exposures examined here do not appear to exert their hypertensive effects primarily through fasting lipid profiles (Table 8).

To assess whether lipid profiles contributed to the association between behavioural exposures and hypertension, triglycerides (TG), LDL-C, and HDL-C were added to the multivariable logistic regression model alongside labour-type physical activity frequency and alcohol drinking frequency (N ≈ 2,182). In this extended model, TG was positively associated with hypertension (OR = 1.10 per 1 mmol/L, 95% CI: 1.01–1.21; p = 0.032), whereas LDL-C was not (OR = 0.98, 95% CI: 0.88–1.09; p = 0.691). HDL-C also showed a positive association with hypertension (OR = 1.76 per 1 mmol/L, 95% CI: 1.26–2.47; p = 0.001), which is opposite to the typical cardioprotective expectation and is more plausibly explained by medication use, reverse causation, or survivor bias in an older population than by a causal adverse effect of HDL-C. Importantly, both behavioural exposures remained significant after lipid adjustment, with only modest attenuation (labour-type physical activity: OR = 1.24 per 1 session/week, 95% CI: 1.16–1.32; p < 0.001; alcohol drinking frequency: OR = 1.18 per 1 occasion/week, 95% CI: 1.02–1.36; p = 0.030). Together, these findings suggest that fasting lipids explain only a small proportion of the observed behavioural associations, and that non-lipid mechanisms—such as cumulative haemodynamic load, neurohumoral activation, and occupational strain—are likely to contribute more substantially (Table 9).

Diagnostic checks for the final hypertension model—including OPA, alcohol frequency, and the core covariates—indicated adequate model performance.

Variance inflation factors showed no evidence of problematic multicollinearity (all VIF < 2.0). Influence diagnostics based on Cook’s distance indicated that no observation exerted undue influence on the model: the maximum Cook’s D was well below 0.5, and 0% of observations exceeded conventional thresholds of 0.5 or 1.0.

The model demonstrated acceptable calibration in the Hosmer–Lemeshow goodness-of-fit test (p ≈ 0.59), supporting the internal validity of the estimated odds ratios. Deviance residuals were approximately symmetric, with no extreme outliers, indicating no major violations of overall model fit.

Box–Tidwell tests suggested minor departures from the linearity-of-logit assumption for age and resting heart rate, whereas BMI and haemoglobin satisfied linearity. Given the small magnitude of these departures and the stability of the OPA and alcohol odds ratios across multiple sensitivity and subgroup analyses, these deviations were judged unlikely to materially affect the substantive conclusions. Overall, the behavioural–lipid–hypertension models appear statistically robust and consistent with the main findings reported in Sections 3.2–3.7.

While moderate, voluntary, and recovery-supported physical activity is well documented to enhance lipid oxidation, HDL-mediated cholesterol transport, and metabolic efficiency, the activity captured in this cohort largely reflects obligatory, high-load agricultural labour (55, 56). Such OPA, often performed for protracted hours under cold-climate conditions, may plausibly diminish recovery capacity and impose cumulative metabolic strain; however, intermediate biomarkers (e.g., catecholamines, oxidative markers) were not measured.

Alcohol consumption displayed a modest metabolic signature partially consistent with recognised hepatic dysregulation pathways via the ADH–ALDH2 system (57, 58). The extended logistic models detailed in Section 3.8.2 help characterise the contributions of lipid biomarkers to hypertension risk within this cohort. Specifically, while triglycerides exhibited a modest but statistically independent association with hypertension, no such relationship was observed for LDL-C. Interestingly, HDL-C demonstrated an unexpected positive association with elevated blood pressure. This pattern is more plausibly attributable to external factors—such as antihypertensive medication exposure, reverse causation, or survivor bias—rather than a direct detrimental physiological effect of HDL-C itself. Such an interpretation is particularly relevant given the cross-sectional nature of the data and the inherent complexities of lipid metabolism in an older population.

A more coherent explanatory pattern emerges from haemodynamic load and cumulative stress physiology. Repeated heavy labour—especially outdoors during Northeast China’s prolonged cold season—may plausibly elicit sustained sympathetic activation, cold-induced vasoconstrictive responses, elevated systemic vascular resistance, and baroreflex strain (59–61). These mechanisms align with a chronic stress-load model, in which non-volitional, environmentally imposed workload contributes to a steadily increasing haemodynamic burden.

The continuous blood pressure models (Supplementary Table S3) provide outcome-level support for this pathway by demonstrating that both SBP and DBP increase with higher OPA and drinking frequency, consistent with a cumulative pressor load rather than an isolated threshold phenomenon. Alcohol use may further potentiate these effects through autonomic imbalance, endothelial dysfunction, and acute fluctuations in vascular tone (62–64). The monotonic behavioural dose–response gradients and stable odds ratios observed across robustness checks (Sections 3.5–3.6) are congruent with a cumulative haemodynamic-stress pathway rather than a lipid-mediated mechanism. Although inflammatory and neurohumoral biomarkers were unavailable, the empirical patterns are broadly consistent with established conceptual models of cumulative physiological load in ageing populations.

The behavioural pathways described above were not uniform across subgroups (Section 3.7). Alcohol-related hypertension risk was concentrated in men, adults aged <70 years, and leaner individuals (BMI < 24 kg/m²). This pattern may reflect sex-based differences in drinking behaviours and sociocultural norms, and is potentially consistent with ALDH2-related metabolic sensitivity that is prevalent in East Asian populations (52, 65). In contrast, occupational physical activity (OPA) exhibited its strongest association with hypertension among adults aged ≥70 years, aligning with age-related declines in vascular compliance, autonomic buffering capacity, and overall physiological reserve that may heighten susceptibility to cumulative workload-related haemodynamic strain (66, 67).

Notably, associations for both exposures appeared attenuated among participants with a BMI ≥ 28 kg/m². This observation may reflect treatment differences, survivorship dynamics, or other unmeasured factors, and should therefore be interpreted cautiously as hypothesis-generating rather than mechanistically definitive. Collectively, these findings reinforce the view that behavioural strain interacts with age-related vascular physiology and body composition to produce differential susceptibility across demographic groups. Consistent with this interpretation, Supplementary Figure S1 illustrates a steeper OPA–hypertension gradient at older ages, and Supplementary Table S5 provides statistical evidence that age modifies the physical activity association (OPA × age interaction).

Across these pathways, our findings support a coherent behavioural–physiological sequence in which environmental constraints shape daily behaviours, which in turn generate a cumulative haemodynamic load with at most modest metabolic strain; this cumulative burden was associated with elevated hypertension risk. This integrated interpretation is consistent with the approximately linear behavioural gradients observed for occupational physical activity (OPA) and alcohol frequency (Sections 3.2–3.4), the limited evidence that fasting lipid markers provide meaningful mediation of these associations (Section 3.8), and the stability of estimates across alternative hypertension definitions and sensitivity analyses (Section 3.6). It also aligns with the subgroup-specific vulnerability patterns by sex, age, and BMI (Section 3.7).

Taken together, the evidence supports a context-sensitive mechanistic narrative in which hypertension risk among rural older adults reflects the intersection of environmental constraints, non-volitional labour demands, habitual alcohol use, and age-related physiological vulnerability. This framework is intended as an interpretive, biologically plausible synthesis rather than a causal claim, and it provides the conceptual basis for the prevention implications discussed in Section 4.4 (5). Finally, the absence of a statistically significant OPA × alcohol interaction (Supplementary Table S4) suggests that prevention strategies should address both exposures as parallel and potentially cumulative targets, rather than presuming synergistic amplification on the multiplicative scale.

The findings of this study challenge conventional assumptions regarding physical activity in ageing populations and highlight the need to distinguish between beneficial, recovery-supported exercise and compulsory, high-load labour. In this rural context, cardiovascular health may depend less on accumulating total activity volume than on achieving a “Healthy Labour Balance”—a concept emphasising appropriate intensity, sufficient recovery, environmental safety, and an age-appropriate workload (68). Consistent with the age-amplified OPA gradient (Supplementary Table S5; Supplementary Figure S1), workload-modulation interventions may be particularly relevant for the oldest adults, for whom incremental increases in OPA frequency correspond to steeper increases in predicted hypertension probability.

From a public health perspective, integrating routine alcohol-use screening, labour-intensity profiling (e.g., brief workload screening items), and bilateral blood pressure measurement (using the higher-arm value) into rural health surveillance could improve the identification of individuals at elevated vascular risk. Community-level prevention should prioritise structured, low-intensity activity programmes, evidence-based alcohol-reduction interventions, and cold-weather protection strategies, alongside strengthened access to primary and preventive care within township-level health systems. Because alcohol-related risk appears concentrated in specific subgroups—notably men, leaner individuals, and the younger older adults—targeted approaches (e.g., brief interventions and culturally adapted counselling) may be more efficient than uniform messaging.

These recommendations align with the WHO Healthy Ageing framework and the Healthy China 2030 agenda, while contributing to emerging international discussions on place-based cardiovascular prevention (68). Importantly, they emphasise that behavioural recommendations must be adapted to local labour patterns, climatic conditions, and social norms, rather than transferred wholesale from urban or high-income settings.

This study has several methodological strengths. It leveraged standardised, government-administered examination data collected under uniform protocols. Blood pressure was measured in both arms, and the higher-arm value was used for primary classification, consistent with AHA and BHS recommendations. Physiological and laboratory assessments were conducted under strict quality-control procedures, and behavioural exposures were assessed for both occupational physical activity (OPA) and alcohol use. The analytic strategy was rigorous, applying HC3 heteroscedasticity-consistent standard errors and confirming robustness across alternative hypertension definitions and multiple sensitivity analyses. In addition, subgroup analyses showed consistent heterogeneity patterns across sex, age, and BMI strata. Supplementary outcome operationalisations—including SBP-only/DBP-only definitions (Supplementary Table S2) and continuous blood pressure models (Supplementary Table S3)—further provide convergent evidence that the behavioural associations are not artefacts of a single diagnostic threshold.

Several limitations warrant cautious interpretation (69). The cross-sectional design precludes causal inference and limits temporal interpretation; therefore, residual confounding and reverse causation cannot be excluded. Behavioural exposures were self-reported, which may introduce recall and social-desirability bias. The anonymised secondary database did not include month- or season-specific examination timestamps; therefore, we could not formally evaluate seasonal variation (e.g., harvest periods or extreme winter cold) in occupational workload or OPA reporting, nor stratify analyses by agricultural or climatic period. Future surveillance with season-resolved timestamps or longitudinal follow-up would help disentangle climatic and agricultural-cycle influences on workload exposure. Moreover, key covariates—such as detailed dietary patterns, psychosocial stressors, medication adherence, and objective environmental exposures (e.g., cold exposure and workload intensity)—were not systematically measured. Accordingly, the proposed mechanistic pathways were inferred from epidemiological patterns rather than directly tested using biomarker assays (e.g., autonomic indicators or inflammatory cytokines). Finally, interaction models (Supplementary Tables S4, S5) assessed effect modification on the multiplicative scale; thus, the non-significant OPA × alcohol term should be interpreted as an absence of multiplicative synergy in these data rather than evidence against a potentially meaningful combined public-health burden, including additive joint effects. Collectively, these limitations highlight key empirical gaps and underscore the need for future multi-site longitudinal studies.

Future research should prioritise longitudinal cohort designs to characterise behavioural–metabolic–haemodynamic trajectories over time and to validate the dual-pathway framework proposed in this study. Integrating multi-omics approaches—including metabolomics, lipidomics, proteomics, and inflammatory profiling—may help clarify whether the observed behavioural associations operate predominantly through cumulative haemodynamic stress, metabolic dysregulation, or coupled pathways linking the two.

Objective exposure assessment should be strengthened using wearable sensors capable of capturing labour intensity, daily physical load, recovery duration, heart-rate variability, and cold-weather exposure in a low-burden manner suitable for rural settings (14). In parallel, AI-enabled modelling tools could support personalised risk stratification, provided they are implemented ethically and evaluated for calibration and clinical utility within local primary-care capacity constraints.

Given the pronounced sex- and age-specific vulnerability patterns, future intervention trials should test tailored strategies—for example, recovery-oriented activity programmes for older women and targeted alcohol-reduction interventions for high-risk subgroups such as men and the younger older adults. Incorporating community-based participatory approaches may further enhance feasibility, cultural acceptability, and adherence in resource-limited settings.

Taken together, this study provides a behavioural–metabolic–environmental framework for understanding hypertension risk among rural older adults in Northeast China. Occupational physical activity (OPA) and alcohol consumption—behaviours shaped by persistent structural, environmental, and cultural constraints—appear to exert convergent haemodynamic and modest metabolic pressures that collectively elevate vascular risk (70).

These findings reinforce that hypertension prevention in rural settings cannot rely on generic recommendations (e.g., “increase physical activity” or “reduce drinking”) without accounting for the lived realities and structural exposures faced by older populations. Instead, effective strategies should be context-adapted, environmentally sensitive, and aligned with community labour patterns and social practices (70). Multiple supplementary analyses further supported the robustness of the observed associations. Specifically, findings were consistent under SBP-only/DBP-only definitions (Supplementary Table S2), and both OPA and alcohol frequency were linearly related to continuous SBP and DBP (Supplementary Table S3). The OPA–hypertension association strengthened with age (Supplementary Table S5; Supplementary Figure S1), whereas the absence of a multiplicative interaction (Supplementary Table S4) suggests that these exposures represent parallel, potentially cumulative targets for intervention.

By situating these behavioural gradients within a place-based framework, this study advances understanding of how cardiovascular risk emerges at the intersection of necessity-driven behaviour, environmental stressors, and ageing physiology, and it provides a foundation for future mechanistic research and policy innovation in rural geriatric health.