OBJECTIVE

Maintenance hemodialysis (MHD) patients often suffer from sarcopenia, affecting lower limb muscle strength and increasing the risk of falls and mortality. This study aims to develop an auxiliary screening model for sarcopenia in MHD patients based on machine learning methods, utilizing lower limb muscle strength indicators, while paying attention to the gender difference and exploring its value in sarcopenia screening.

METHODS

This cross-sectional study collected data from MHD patients at a hemodialysis center in China. Sarcopenia was assessed using the 2019 Asian Working Group for Sarcopenia update. A self-developed lower limb muscle strength testing device was used. Other physiological indicators, including basic information and lab findings, were collected. Participants were grouped into sarcopenia and control groups, with gender-specific binary classification models developed. Stratified shuffling and synthetic minority oversampling techniques were used to build screening classifiers.

RESULTS

Data from 164 MHD patients were ultimately collected, including 83 males (41 with possible sarcopenia or sarcopenia) and 81 females (53 with possible sarcopenia or sarcopenia). Gender-specific binary classification models were developed using lower limb muscle strength indicators, with the male model having an AUC of 79% and the female model an AUC of 80%, respectively. Combining lower limb muscle strength with other physiological indicators improved the female model's screening capability, achieving an AUC of 90%.

CONCLUSION

This study demonstrates that the auxiliary screening model for sarcopenia, developed using machine learning methods, highlights the significant value of lower limb muscle strength indicators in identifying sarcopenia in MHD patients. The gender-specific screening models show good discriminatory ability across different genders, providing effective tools for the early screening and management of sarcopenia in MHD patients.

TRIAL REGISTRATION

Chinese Clinical Trial Registry (ChiCTR2100051111), registered on 2021-09-13.