Association Between Physical Activity and Anxiety in A Rural Chinese Middle-Aged and Older Cohort

Introduction

The prevalence of anxiety disorders increases in China, driven by population growth and demographic aging.^1,2 Anxiety is associated with psychological distress, multi-system dysfunction, and more often than not this is accompanied by comorbidity with chronic physical conditions which can cause a reduction in the quality of life and social functioning.^3–7 There is growing support in the field of clinical practice to incorporate lifestyle intervention with conventional treatment.⁸ Regular physical activity (PA) is a cost-effective cornerstone of mental health lifestyle interventions through integrated psychobiological pathways.^9,10

The connection between PA and anxiety is complicated and controversial particularly in terms of domains and quantity of activities. To start with, the physical activity paradox has found that:^11–13 autonomous, enjoyable leisure-time physical activity (LTPA) is associated with lower anxiety, whereas occupational physical activity (OPA) is associated with increased fatigue and higher anxiety risk.^14–16 OPA is predominantly governed by external regulation such as financial reward, constraining autonomy over the nature and intensity of exertion. Conversely, LTPA is fueled by intrinsic motivation, marked by high self-determination and hedonic enjoyment.^17,18 Notably, the physical activity paradox framework has predominantly centered on the dichotomy between LTPA and OPA, largely overlooking household physical activity (HPA) as an independent domain. HPA is a part of daily movement and is explicitly recognized as a distinct dimension in widely used instruments such as the International Physical Activity Questionnaire (IPAQ). Existing evidence regarding the relationship between HPA and anxiety remains sparse and equivocal.^11,19 Second, the relationship between anxiety and PA volume is nonlinear.^20,21 With excessive intensity or prolonged duration, the association may weaken or reverse, and may be associated with negative effects due to an inability to adapt and recover.²² This highlights the need to identify domain-specific turning points to define the safe and effective PA range for anxiety management.

Rural middle-aged and older adults in China represent an ideal population for examining the physical activity paradox. Unlike their urban counterparts, who are predominantly engaged in sedentary or discretionary occupations, rural older adults often continue to perform heavy agricultural labor well into old age, rendering their OPA largely obligatory and associated with insufficient recovery time.^23,24 At the same time, this population faces unique psychosocial stressors, including poor access to mental health services, a high prevalence of the empty-nest syndrome resulting from adult children migrating to cities, and economic vulnerability.^25–27 Each of these factors may be associated with a higher baseline risk of anxiety. In addition, cultural background is associated with the expression of anxiety symptoms, pointing to the value of local evidence.²⁸ Against this background, the present study systematically examined domain-specific associations and nonlinear relationships between PA and anxiety in a rural middle-aged and older adult population. We propose the following hypotheses. First, based on the physical activity paradox, we hypothesize that LTPA is inversely associated with anxiety, whereas OPA is positively associated with anxiety. Second, given the inconsistent evidence on HPA and its limited investigation in this population, we do not prespecify a directional hypothesis and instead treat this association as exploratory. Third, we hypothesize that the dose-response relationship between OPA and anxiety is nonlinear, with an inflection point beyond which the risk pattern changes. By integrating activity domains and individual characteristics, this study aims to provide targeted evidence for public health guidance on anxiety prevention in China.

Materials and Methods

Study Design and Participants

This study was based on data from the Yangzhou Upper Gastrointestinal Disease Screening Cohort, which enrolled participants from 2017 to 2024 in Jiangsu Province. Of the 4904 enrolled participants, 510 were excluded due to missing covariates or prior cancer, yielding a final cohort of 4394 individuals (Figure 1). No imputation was performed for missing covariate data.

Figure 1 Flow chart for inclusion and exclusion criteria of the study population.

All participants underwent standardized questionnaires, physical examinations, fasting laboratory tests, and upper gastrointestinal endoscopy. Our research has been ethically approved by the Ethics Review Committee of Yangzhou University Affiliated Hospital. All procedures were performed in accordance with the Helsinki Declaration, and written informed consent was obtained from all participants before enrollment.

Assessment of PA

Self-reported PA was assessed using a questionnaire based on the IPAQ.²⁹ The questionnaire captured occupational, leisure-time, and household activities, recording key parameters for each: type, intensity, weekly frequency, session duration, and cumulative years of engagement. Commuting-related activities were excluded from the total physical activity (TPA) calculation because the predominant modes (motor vehicle/electric bicycle) in the study population involved energy expenditure comparable to sedentary levels. HPA was classified as light-intensity physical activity (LPA).

For quantitative analysis, PA volume was computed through MET values, which were assigned based on the intensity of each activity:^30,31 3.3 METs for LPA (such as walking, light household chores), 4.0 METs for moderate-intensity PA (MPA, such as brisk walking, Tai Chi), and 8.0 METs for vigorous-intensity PA (VPA, such as running, heavy manual labor). The weekly PA volume for each domain d (occupational, leisure-time, or household) was calculated as:

where MET_i, Frequency_i, and Duration_i represent the assigned MET value, weekly session days, and session duration in minutes for activity i, respectively.

No observations were excluded from the main analysis. IQR‑based outlier removal was used only in sensitivity analyses to test robustness.

Assessment of Anxiety

Anxiety was evaluated through the 7-item Generalized Anxiety Disorder Scale (GAD-7).³² On this scale, participants rated each symptom’s two-week frequency from 0 to 3, with a score ≥5 indicating the presence of anxiety symptoms.³³

Covariates

Sociodemographic and health-related data were collected through standardized questionnaires and physical examinations. Marital status was categorized as married or unmarried/living alone. Education was categorized as ≤9 years, 9–12 years, and >12 years of schooling. BMI was classified into <25 kg/m², 25–30 kg/m², and ≥30 kg/m². Length of sleep was classified as short (<6 h/day), suitable (6–8 h/day), and long (>8 h/day). Cumulative smoking exposure was quantified in pack-years and categorized as never, light-to-moderate (<20 pack-years), or heavy (≥20 pack-years). Alcohol drinking was defined as current (≥1 standard drink per week on average over the past year) versus non-use (never or former drinking). Based on their self-reported psychological stress over the past month, participants were categorized into low, moderate, and high groups. Hypertension was defined as self-reported diagnosis or SBP≥140 mmHg and/or DBP≥90 mmHg. Diabetes was defined as self-reported diagnosis or current use of glucose-lowering medication. Heart disease was defined as physician-confirmed coronary heart disease, myocardial infarction, or heart failure. Gastric disease was defined as present if endoscopy and histopathology showed atrophy, intestinal metaplasia, dysplasia, or cancer, and absent if normal mucosa or non-atrophic gastritis was found. Three inflammation-based composite indices were computed from the complete blood count: the neutrophil-to-lymphocyte ratio (NLR) and the monocyte-to-lymphocyte ratio (MLR) were calculated as the ratios of absolute neutrophil or monocyte counts to the absolute lymphocyte count, respectively; the systemic immune-inflammation index (SII) was calculated as platelet count × neutrophil count/lymphocyte count.^34,35 Pepsinogen ratio (PGR) was computed as serum pepsinogen I divided by pepsinogen II (PG I/PG II).³⁶

Statistical Analysis

Continuous variables were presented as means ± standard deviations, while categorical variables were reported as frequencies and corresponding percentages. Multivariable logistic regression was applied to quantify the association between PA and anxiety, yielding adjusted odds ratios (ORs) with 95% confidence intervals (CIs). Three hierarchical logistic regression models were fitted: Crude (unadjusted); Model 1 (adjusted age and sex); Model 2 (additionally adjusted marital status, education, BMI, length of sleep, pack-years of smoking, alcohol drinking, psychological stress, hypertension, diabetes, heart disease and MLR). PA was analyzed as a continuous variable (per 1-SD increase in MET-min/week) and in tertiles. We used restricted cubic splines (RCS) fitted with four knots (at the 5th, 35th, 65th, and 95th percentiles) to model potential nonlinear relationships. Subgroup analyses were conducted according to age, sex, BMI, length of sleep, pack-years of smoking, and alcohol drinking. To assess the robustness of our main findings, we performed a sensitivity analysis in which TPA outliers identified by the IQR method (observations below Q1−1.5×IQR or above Q3+1.5×IQR) were excluded. Finally, the Boruta algorithm was employed to identify factors significantly associated with anxiety.

Results

Baseline Characteristics

The study included 4394 participants, with 622 (14.2%) screening positive for anxiety symptoms. The mean age of the cohort was 56.67±8.73 years, and 41.1% were male. TPA averaged 5853.03±2283.28 MET-min/week, driven mainly by OPA (3224.48±1964.73 MET-min/week), followed by HPA (1797.13±1326.12 MET-min/week) and LTPA (831.43±265.52 MET-min/week). Anxiety was significantly more prevalent among females, younger individuals, short sleepers, non-drinkers, lighter smokers, and those with higher psychological stress, heart disease, or lower MLR (Table 1).

Table 1 Baseline Characteristics of Participants

Associations Between PA and Anxiety

Logistic regression analysis showed domain-specific associations between PA and anxiety (Table 2). In the fully adjusted model, TPA and OPA were positively associated with anxiety risk. Specifically, each 1-SD increment in TPA or OPA was associated with an OR of 1.22 for anxiety (95% CI: 1.11–1.34, P<0.001). When analyzed by tertiles, participants in the highest TPA tertile (T3) had higher odds compared with the lowest tertile (T1) (OR=1.35, 95% CI: 1.07–1.70, P=0.010); for OPA, the T3 versus T1 comparison was not statistically significant (OR=1.22, 95% CI: 0.98–1.53, P=0.078). HPA showed a marginally significant positive association with anxiety in the fully adjusted model (OR=1.10, 95% CI: 1.00–1.21, P=0.050). LTPA exhibited an inverse association with anxiety: each 1-SD increment was associated with an OR of 0.88 (95% CI: 0.80–0.96, P=0.004), and the T3 versus T1 comparison yielded an OR of 0.74 (95% CI: 0.59–0.94, P=0.012).

Table 2 Associations Between PA and Anxiety

RCS analysis showed reverse L-shaped nonlinear associations with anxiety symptoms for both TPA (inflection point: 7098.72 MET-min/week) and OPA (inflection point: 4200.11 MET-min/week) (Figure 2). Below these thresholds, the associations were not statistically significant (P>0.05). Above the thresholds, each 100 MET-min/week increase in TPA and OPA was associated with higher anxiety odds (OR=1.02, 95% CI: 1.01–1.03, P=0.001 for TPA; OR=1.02, 95% CI: 1.01–1.04, P=0.003 for OPA). In contrast, HPA and LTPA showed no significant nonlinear associations with anxiety (Table 3 and Table 4).

Table 3 Association Between TPA and Anxiety Using Two-Piecewise Regression Models

Table 4 Association Between OPA and Anxiety Using Two-Piecewise Regression Models

Figure 2 Restricted cubic spline analysis of the association between PA and anxiety. (A) total physical activity (TPA), (B) occupational physical activity (OPA), (C) household physical activity (HPA), (D) leisure-time physical activity (LTPA).

Sensitivity Analysis

Subgroup analyses by age, sex, BMI, sleep duration, pack-years of smoking, and alcohol drinking showed that the associations of TPA, OPA, and HPA with anxiety remained stable across all strata (Figures 3 and S1–S3). Although the P value for the LTPA–smoking interaction was below 0.05, this finding should be interpreted with caution given multiple testing. Furthermore, the results remained robust after excluding IQR-defined outliers (Table S1). As shown in Figure 4, the Boruta algorithm identified 11 variables as the strongest predictors of anxiety after 500 iterations. Ranked by their z-scores, these variables were: psychological stress, sex, sleep duration, OPA, HPA, pack-years of smoking, age, heart disease, education, MLR, and LTPA.

Figure 3 Association between TPA and anxiety. Each stratification was adjusted for age, sex, marital status, education, BMI, length of sleep, pack-years of smoking, alcohol drinking, psychological stress, hypertension, diabetes, heart disease, MLR except the stratification factor itself. Bold font indicates the overall effect and each subgroup itself.

Figure 4 Feature selection results of the Boruta algorithm. Predictor importance for anxiety according to the Boruta algorithm. The blue box plots indicate the Z-scores of the shadow attributes; the red, yellow, and green box plots represent the rejected, tentative, and confirmed scoring attributes, respectively. A predictor is deemed important (green) if its mean importance Z-score is significantly higher than the maximum value of the shadow variables, and excluded (red) if significantly lower than the maximum value of the shadow variables. Predictors marked in yellow are classified as tentative, indicating that their importance remains uncertain and requires further evaluation.

Discussion

The study examined domain-specific associations of PA with anxiety based on the Yangzhou Upper Gastrointestinal Disease Screening Cohort, providing local empirical support for the physical activity paradox. The main findings were as follows: TPA was positively associated with anxiety; OPA was positively associated, LTPA negatively associated, and HPA weakly positively associated. Furthermore, RCS analysis indicated reverse L-shaped associations of both TPA and OPA with anxiety.

We found a significant positive association between PA and anxiety among rural middle-aged and older adults engaged in heavy labor. While seemingly at odds with the consensus that “exercise promotes mental health”,^37,38 this finding highlighted the critical domain-specific associations of PA. In this cohort, PA was predominantly comprised of OPA, which was characterized by high psychological pressure, low autonomy, and inadequate recovery.^12,39 Functioning as a chronic passive stressor rather than an intrinsically motivated health behavior, OPA may be associated with attenuated or even negated protective associations.^40–42 At a biological level, prolonged occupational exertion is linked to anxiety risk through sustained sympathetic overactivation and dysregulation of the HPA axis feedback,^43,44 reduced BDNF synthesis, and elevated systemic inflammation.^45,46 To explore whether inflammation is involved in the above association, we calculated NLR, MLR, and SII. The results did not support this hypothesis: there were no differences in NLR or SII between the anxiety and non-anxiety groups, while MLR was unexpectedly lower in the anxiety group. This counterintuitive finding is unlikely to reflect immunosuppression; rather, it is more likely attributable to baseline characteristics of the anxiety group, including a higher proportion of females and lower rates of smoking and alcohol consumption. Thus, the inflammatory pathway was not supported in our cohort, and psychosocial factors may be more strongly associated with anxiety.

In contrast, LTPA is associated with better mental health, possibly through fulfilling the basic psychological needs for autonomy, relatedness and competence.^11,18,47 This positive engagement parallels the neurobiological findings of moderate exercise—which upregulates neurotrophic factors (IGF-1, BDNF, VEGF) to support neuroplasticity and cerebrovascular health,^48,49 modulates monoamine and endogenous opioid systems for emotion regulation,^50–52 and improves cerebral blood flow and metabolism.^53,54

Notably, HPA showed a marginally significant positive association with anxiety. While an early meta-analysis (over 640,000 participants) found no significant association with mental health,¹¹ a 2025 meta-analysis (over 3.3 million participants) reported a weak positive association with mental health.¹⁹ This shift may reflect lifestyle changes: more people now engage in enjoyable, autonomous household activities like gardening or parent-child play, rather than only traditional chores.^17,19 In our cohort, the persistently weak positive association suggests that rural middle-aged and older adults perform traditional, livelihood-oriented, duty-driven household activities. Although such activities may provide daily structure and a sense of family role identity, their autonomy is far lower than that of LTPA, and they often entail minor burdens such as time pressure and fatigue. These characteristics likely explain the weak positive association.

Our RCS analysis indicated a reverse L-shaped relationship between PA and anxiety risk, with an inflection point at 7098.72 MET-min/week. This finding aligned with existing research in confirming a complicated relationship between PA and anxiety, while highlighting the significant moderating roles of population characteristics and activity domains. For instance, a study of 30054 Chinese adults revealed an inverse J-shaped relationship, with benefits peaking at 903 MET-min/week;⁵⁵ this lower threshold was thought to reflect the optimal mental health benefit zone in general populations dominated by LTPA. In contrast, the substantially higher inflection point identified in our study may represent a critical point of psychophysiological tolerance in a population primarily engaged in OPA. We posit that long-term adaptation to high-intensity labor elevates this statistical threshold. However, this very adaptation implies a sustained high-load physiological state; consequently, exposure beyond this critical point may be associated with acute homeostatic overload and higher anxiety risk.⁵⁶ This pattern is further contextualized by a study of a primarily urban, younger Japanese population, which reported a U-shaped relationship with the lowest anxiety at an intermediate threshold of 5277 MET-min/week,²¹ bridging the gap between the leisure-focused Chinese cohort and our occupationally active sample. Taken together, these comparisons suggest that anxiety-related PA thresholds are not fixed values but exist along a dynamic gradient, mediated by the dominant domain of activity and socio-demographic settings.

This study has several limitations. First, its cross-sectional design precludes causal inference and cannot rule out reverse causality—preexisting anxiety or economic hardship may drive excessive OPA, while psychological barriers may limit LTPA. Longitudinal data are needed to clarify temporal relationships. Second, PA was self-reported, introducing potential recall bias; although the questionnaire followed the IPAQ guidelines and drew on validated Chinese cohort instruments, it lacked formal psychometric validation. Third, commuting activities were excluded (due to predominant motor vehicle/electric bicycle), which may have underestimated TPA for walkers or cyclists—future studies should capture detailed commuting modes. Fourth, we used psychological stress as a proxy for broader psychological distress, and residual confounding by depression or unmeasured economic pressure remains possible. Notably, direct socioeconomic indicators were not included in the analysis due to low response rates and reporting bias—this may overestimate the observed association between OPA and anxiety. Finally, the findings are based on rural, middle-aged and older manual workers in Yangzhou, which limits their generalizability to other populations or regions.

Conclusion

In this rural, occupation-dominated middle-aged and older population, we identified significant domain-specific associations between physical activity and anxiety. OPA exhibited a reverse L-shaped nonlinear relationship: no significant association below a specific threshold, but higher odds of anxiety above that threshold. LTPA showed a linear inverse association. HPA presented a marginally significant positive association. For rural adults engaged in heavy occupational labor, these results suggest that avoiding excessive OPA and engagement in LTPA are associated with more favorable anxiety profiles than following the general recommendation to exercise more. Future public health guidelines may need to distinguish between activity domains and account for population characteristics in order to inform more precise intervention strategies.