Establishing thresholds of handgrip strength based on mortality using machine learning in a prospective cohort of Chinese population

Background: The relative prognostic importance of handgrip strength (HGS) in comparison with other risk factors for mortality remains to be further clarified, and thresholds used for best identify high-risk individuals in health screening are not yet established. Using machine learning and nationally representative data from the China Health and Retirement Longitudinal Study (CHARLS), the study aimed to investigate the prognostic importance of HGS and establish sex-specific thresholds for health screening.Methods: A total of 6,762 participants from CHARLS were enrolled. A random forest model was built using 30 variables with all-cause mortality as outcome. SHapley Additive exPlanation values were applied to explain the model. Cox proportional hazard models and Harrell's C index change were used to validate the clinical importance of the thresholds.Results: Among the participants, 3,102 (45.9%) were men, and 622 (9.1%) case of death were documented follow-up period of 6.78 years. The random forest model identified HGS as the fifth important prognostic variable, with thresholds for identifying high-risk individuals were < 32 kg in men and < 19 kg in women. Low HGS were associated with all-cause mortality [HR (95% CI): 1.77 (1.49-2.11), p < 0.001]. The addition of HGS thresholds improved the predictive ability of an established office-based risk score (C-index change: 0.022, p < 0.001).Conclusion: On the basis of our thresholds, low HGS predicted all-cause mortality better than other risk factors and improved prediction of a traditional office-based risk score. These results reinforced the clinical utility of measurement of HGS in health screening.