Association of Physical Activity Levels with Severe Multimorbidity Among Middle-Aged and Older Adults: Evidence from CHARLS and a Community Sample

Background: Multimorbidity imposes a substantial burden on older populations, yet research neglects how physical activity (PA) relates to severe disease burden in established cases. Moreover, the efficacy of combining low-cost PA assessments with routine indicators to identify high-burden patients in resource-constrained settings remains unclear. This study explores associations between PA levels and severe multimorbidity (≥4 conditions), evaluating PA’s utility for auxiliary status identification via interpretable machine learning. Methods: This cross-sectional study included the CHARLS 2018 sample (n = 10,130) and a supplementary community sample (n = 315). Multivariable logistic regression evaluated associations between PA and severe multimorbidity. An XGBoost model assessed routine indicators’ classification performance, utilizing SHAP values to analyze feature contributions to the model. Results: PA was inversely associated with severe multimorbidity specifically among those meeting the WHO criteria (MPA > 300 or VPA > 150 min/week) for additional health benefits (OR = 0.80, 95% CI: 0.73–0.88, p < 0.001), especially in adults ≥60 years. Although models using solely routine indicators showed modest discrimination (AUC = 0.673), SHAP analysis revealed PA as the foremost modifiable behavioral feature in classification, with local patterns aligning closely with regression findings. Conclusions: Among middle-aged and older adults with multimorbidity, PA levels exhibit significant categorical associations with severe multimorbidity. In community management, low-cost PA evaluation serves as an auxiliary indicator to preliminarily classify high-burden individuals. Given potential reverse causality, future longitudinal cohorts are required to clarify causal relationships, providing direct evidence for daily activity management.