This study adopts a nationwide cross-sectional design to explore the relationship between cognitive function and physical performance, following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [20].

A multistage stratified sampling method was used to recruit 20,868 community-dwelling adults aged 60+ years from 6 provinces in China (Fujian, Guangdong, Nanjing, Sichuan, Xinjiang, and Guizhou) between January and August 2022. The required sample size was calculated using the formula , yielding a sample size of 13,830 for a 95% CI with a margin of error of 0.05 and SD of 3 [21]. To account for invalid questionnaires, a 20% adjustment was made, resulting in a final sample size of 17,288. Inclusion criteria: participants aged 60+ with adequate communication and complete assessments. Exclusion criteria: participants with severe physical disabilities preventing SPPB testing or acute medical conditions (eg, postoperative recovery).

Cognitive function was evaluated using the Mini-Mental State Examination (MMSE) [25], with the Chinese version validated by Wang et al [26] showing high test-retest reliability (intraclass correlation coefficient=0.91). The MMSE assesses five cognitive domains: orientation (time and place), memory (immediate and delayed recall), attention and calculation, language, and visuospatial ability. Scores ranged from 0 to 30, with higher scores indicating better cognitive function. The criteria for normal cognitive function vary by educational level: (1) middle school or above: score >24 (normal), ≤24 (impairment); (2) primary school: score >20 (normal), ≤20 (impairment); (3) illiteracy: score >17 (normal), ≤17 (impairment).

Sociodemographic characteristics and health behaviors were considered covariates. Sociodemographic characteristics included age, sex (male and female); BMI (<18.5, 18.5‐23.9, ≥24 kg/m²), educational level (primary school or below, middle school, high school or vocational school, and college or above), marital status (married vs other, including divorced, widowed, or single), number of children (none, 1, 2, 3, or more), residential area (rural, urban), living arrangement (living alone, living with relatives, living with a spouse), geographic region (eastern or western), data source (home-based or community health service center), and monthly household income (<5000 or ≥5000).

Health behaviors included the presence of chronic diseases (defined as having at least one: hypertension, coronary heart disease, cerebrovascular disease, or diabetes), long-term medications use (none, 1‐2, or ≥3), alcohol consumption (never, formerly, or currently), smoking status (never, formerly, or currently), weekly social participation frequency (none, 1‐3, or ≥4 d), weekly exercise frequency (none, 1‐3, or ≥4 d), self-reported social support (insufficient, inadequate, or sufficient), and self-reported visual and hearing impairment (no or yes).

A designated assessor was assigned to each research site to oversee data collection using the “Jingyice Elderly Function Assessment Platform,” a WeChat (Tencent) mini-program developed by the research team. All assessors, registered nurses from community health centers, nursing homes, or hospitals, received uniform online training on the platform and study procedures. Before assessment, participants were informed of the study’s aims and procedures, and written informed consent was obtained. Assessors conducted face-to-face interviews and functional tests, entering responses and test results (eg, cognitive assessments and physical performance measures) into the mini-program, which synchronized data to a secure web-based platform for real-time monitoring. To ensure data quality, the backend recorded the time and duration of each evaluation, and the research team reviewed data weekly, focusing on submissions completed between 10 PM and 8 AM or within 15 minutes. Questionnaires and test records with incomplete scale data, >15% missing demographic information, patterned responses, or logical inconsistencies were excluded.

Continuous variables are presented as mean SD, while categorical variables are summarized as frequencies (%) via Pearson chi-square tests. To assess bidirectional associations, cognitive function (MMSE score) was modeled as both continuously and categorically (impaired or normal), whereas physical performance (SPPB total score) was analyzed as a continuous variable in linear regression and dichotomized (poor or normal) in logistic regression. All models were adjusted for covariates listed in the “Measures” section.

Nonlinear relationships were assessed using restricted cubic splines (RCS), with the optimal model fit chosen based on the Akaike Information Criterion from models with 3 to 6 knots. A 2-stage segmented linear regression approach identified threshold effects using the likelihood ratio test, and the optimal threshold was determined via an iterative grid search.

Subgroup analyses by sociodemographic factors (eg, age, education, and income) were conducted, and interaction effects were assessed using product terms (eg, MMSE×income). All analyses were performed with R (version 4.2; R Core Team) and SPSS (version 25.0; IBM Corp). Statistical significance was set at a 2-tailed P value <.05.

This study was approved by the Ethics Committee of Beijing Hospital (approval number 2023BJYYEC-446‐02). Written informed consent was obtained from all participants prior to participation. All data were anonymized, participants received no compensation, and no identifiable images are included in the article or supplementary materials. The study was conducted in accordance with the ethical standards of the Declaration of Helsinki.

A total of 20,671 participants were included in the final analysis, with a mean age of 72.19 (SD 12.238) years; 48% (9931/20,671) were male and 52% (10,740/20,671) were female (Table 1). The majority of participants (12,703/20,671, 61.5%) had low education levels. Cognitive impairment was identified in 14.4% (2975/20,671), and 62.2% (12,857/20,671) exhibited poor physical performance with a mean SPPB score of 8.4 (SD 2.94). Domain-specific impairments were prevalent: 26.7% (5521/20,671) had balance deficits, 67.3% (13,906/20,671) showed reduced gait speed, and 58.4% (12,071/20,671) demonstrated muscle weakness, with gait speed showing the most pronounced deficit (mean score 2.34, SD 1.36; Figure 1). Furthermore, there were also differences in SPPB physical performance across various covariate subgroups (Table 1).

RCS analyses showed significant nonlinear associations between cognitive function (MMSE) and physical performance measures (SPPB, gait speed, balance, and muscle strength; Figure 2 ; Figures S1-S3 in Multimedia Appendix 1). These relationships remained consistent in gender and education-stratified analyses (Figures 3 and 4; Figures S1A-S3B in Multimedia Appendix 1), confirming robustness across demographic subgroups.

This study revealed threshold effects and dynamic changes between cognitive function and physical performance using RCS analyses. Significant inflection points were found at MMSE=24 for overall performance (SPPB), balance, and gait speed, and at MMSE=29 for FTSST (Table 4). In the impaired cognitive period (MMSE <24), cognition improvements protected physical performance, reducing the risk of SPPB by 14% (OR 0.86, 95 CI 0.84‐0.87), gait speed by 16% (OR 0.84, 95% CI 0.82‐0.85), and balance by 4.8% (OR 1.05, 95% CI 1.03‐1.06). In the optimal cognitive period (MMSE ≥24), the protective association remained significant for balance (risk reduced by 31%, OR 0.69, 95% CI 0.68‐0.70) but weakened for overall performance (SPPB,P=.054), and a marginal positive association was observed for gait speed (OR 1.05, 95% CI 1.04-1.07), likely because cognitively intact individuals were already near the upper limit of normal performance, rather than reflecting a harmful relationship. The protective association on muscle strength persisted, with the most significant impact at MMSE >29 (risk reduced by 32%; OR 0.68, 95% CI 0.63‐0.74).

In addition, we identified a second inflection point in over physical performance (SPPB) at MMSE=19 using second derivative analysis (Figure S4 in Multimedia Appendix 1). After adjusting for confounding factors, the MMSE 19‐24 group showed a 34% lower risk of physical performance impairment (SPPB) compared to the MMSE <19 group (OR 0.66, 95% CI 0.57‐0.77) (Table S3 in Multimedia Appendix 1).

Subgroup analyses revealed significant heterogeneity in the association between cognitive impairment and physical performance across demographic and socioeconomic strata (Figure 5). The strongest associations were observed in older adults aged ≥90 years (OR 3.33, 95% CI 2.04‐5.42), those with college or above level education (OR 3.98, 95% CI 1.48‐10.70), high-income households (≥5000 yuan per month [US $1 = ¥7.113]: OR 5.24, 95% CI 3.91‐7.02), and urban residents (OR 3.85, 95% CI 3.28‐4.52). Interaction models further demonstrated graded effects: the detrimental impact of cognitive impairment increased with advancing age (β=0.45‐0.71 per age group), higher education (β=0.55‐0.77), and socioeconomic advantage (β=0.64‐0.85 for urban residency and high income), suggesting these factors may amplify vulnerability to physical decline (Table S4 in Multimedia Appendix 1).

This study advances the understanding of the cognitive-physical performance relationship through 3 key innovations. First, we demonstrated a bidirectional yet asymmetric relationship: cognitive impairment exerted a stronger detrimental association on physical function than vice versa. Second, our nonlinear threshold analysis revealed critical inflection points (MMSE=19, 24, 29), defining distinct intervention phases. The MMSE 19‐24 range marked a critical intervention window where cognitive improvements maximally protected physical function, whereas MMSE >24 signaled a ceiling effect for global physical performance and a paradoxical reversal in gait speed, aligning with the cautious gait hypothesis. Finally, subgroup analyses uncovered a socioeconomic paradox: higher socioeconomic status (SES), indicated by advanced education, income, and living in an urban area, exhibited heightened vulnerability to physical functional decline. These innovations combine mechanistic research with clinical applications, offering strategies to address the disability burden in an aging society.

Our findings corroborate and extend the evidence on cognitive-physical interdependencies, while challenging the prevailing assumption of symmetrical reciprocity. Longitudinal studies have shown cognitive function as a predictor of physical decline, particularly fall risk [27] and gait deterioration [28]. A cross-sectional study [29] found that increased cognitive impairment severity leads to worsened balance control, consistent with our observation of bidirectional associations. Notably, while Li et al [28] reported a bidirectional relationship between cognitive function and gait speed, our analysis revealed a critical asymmetry: cognitive impairment exerted a disproportionately stronger impact on physical disability than the reverse. This divergence challenges the notion of equivalent bidirectional association and positions cognitive preservation as a pivotal strategy to disrupt the disability cascade.

This study found a nonlinear relationship between cognitive function and physical performance across individuals of various genders and educational levels. It identified 3 protective thresholds at scores of 19, 24, and 29, offering key evidence for developing precise intervention strategies to maintain functionality in older individuals. At the stage where cognitive function may be impaired (MMSE <24), it significantly protects all physical performance dimensions, with the greatest association observed in the 19‐24 score range, suggesting it should be a target for clinical intervention. These results align with previous research [30], which found a positive correlation between cognitive and physical function, suggesting cognitive intervention can delay physical decline. Our study further quantifies cognitive function threshold points. Thus, it is recommended to assess functional levels at this stage and conduct adaptive dual-task training targeting both cognitive and physical functions to enhance both [31].

In the optimal cognitive function stage (MMSE ≥24), physical performance exhibits dimension-specific changes: overall physical performance (SPPB) shows a clear “ceiling effect,” suggesting that further improvements in physical performance may have reached physiological limits, leaving limited room for cognitive function to contribute. Gait speed, however, exhibits an unusual reversal phenomenon, consistent with George et al [32] inverted U-shaped gait-cognition curve, attributed to the “cautious gait hypothesis.” This hypothesis suggests that older adults with higher cognitive function consciously regulate their movements, resulting in a more cautious and controlled walking pattern [33]. Cognitive function exerts a continuous protective association on balance ability, most pronounced when MMSE >29. This finding supports the view of Kuan et al [34] that early-stage mild cognitive impairment, regardless of musculoskeletal function, can impact balance, potentially due to the weakening of the cortical-vestibular network in individuals with cognitive impairment, particularly in key brain areas involved in integrating visual, auditory, and vestibular signals, correlating with the decline in balance ability [35].

Interestingly, our subgroup and interaction analyses revealed a socioeconomic paradox:

older individuals with higher SES, as measured by education and income, exhibited amplified risks of concurrent cognitive-physical decline, contradicting the conventional SES-health gradient. This counterintuitive association necessitates a multidimensional interpretation rooted in the Cognitive Reserve Hypothesis. While enhanced neurocompensatory mechanisms in high-SES populations may initially delay clinical manifestations of cognitive impairment [36], prolonged compensatory demands could deplete neural resources, precipitating accelerated physical deterioration once pathological thresholds are breached—a” double-edged sword” effect [37]. Moreover, behavior patterns related to occupational characteristics warrant attention. This group may have maintained an imbalanced “high cognitive load-low physical activity” state throughout their careers, leading to accelerated degeneration of motor function neural circuits [38]. These insights advocate for dual-target interventions prioritizing simultaneous cognitive reinforcement and physical activation in high-SES older adults, particularly during the identified critical window (MMSE 19‐24), to counteract reserve depletion trajectories.

Of course, our study has several limitations. First, as this study uses a cross-sectional design, future research should adopt longitudinal approaches to explore causal relationships and developmental trajectories. Second, due to the feasibility constraints of large-scale studies, many variables were assessed via self-report measures, potentially introducing reporting bias. Third, we did not directly measure occupational type or physical activity levels using validated instruments. Future studies are recommended to further investigate the “high cognitive load–low physical activity” hypothesis. Finally, this study evaluated overall cognitive function but did not assess specific cognitive domains, such as memory or executive function. Future research should investigate these domains to better understand the relationship between cognitive function and physical performance.

This study demonstrates a bidirectional, asymmetric relationship between cognitive function and physical performance, identifies a nonlinear threshold effect, and highlights substantial population heterogeneity. These findings offer novel insights into the interaction between cognitive decline and physical disability, particularly in older individuals, and provide guidance for developing targeted interventions for high-risk subgroups in an aging society.