Use machine learning to help identify possible sarcopenia cases in maintenance hemodialysis patients

Abstract

Background Maintenance hemodialysis (MHD) patients often suffer from sarcopenia, which is strongly associated with their long-term mortality. The diagnosis and treatment of sarcopenia, especially possible sarcopenia for MHD patients are of great importance. This study aims to use machine learning and medical data to develop two simple sarcopenia identification assistant tools for MHD patients and focuses on sex specificity. Methods Data were retrospectively collected from patients undergoing MHD and included patients’ basic information, body measurement results and laboratory findings. The 2019 consensus update by Asian working group for sarcopenia was used to assess whether a MHD patient had sarcopenia. Finally, 140 male (58 with possible sarcopenia or sarcopenia) and 102 female (65 with possible sarcopenia or sarcopenia) patients’ data were collected. Participants were divided into sarcopenia and control groups for each sex to develop binary classifiers. After statistical analysis and feature selection, stratified shuffle split and Synthetic Minority Oversampling Technique were conducted and voting classifiers were developed. Results After eliminating handgrip strength, 6-m walk, and skeletal muscle index, the best three features for sarcopenia identification of male patients are age, fasting blood glucose, and parathyroid hormone. Meanwhile, age, arm without vascular access, total bilirubin, and post-dialysis creatinine are the best four features for females. After abandoning models with overfitting or bad performance, voting classifiers achieved good sarcopenia classification performance for both sexes (For males: sensitivity: 77.50% ± 11.21%, specificity: 83.13% ± 9.70%, F1 score: 77.32% ± 5.36%, the area under the receiver operating characteristic curves (AUC): 87.40% ± 4.41%. For females: sensitivity: 76.15% ± 13.95%, specificity: 71.25% ± 15.86%, F1 score: 78.04% ± 8.85%, AUC: 77.69% ± 7.92%). Conclusions Two simple sex-specific sarcopenia identification tools for MHD patients were developed. They performed well on the case finding of sarcopenia, especially possible sarcopenia.

Machine Learning Applications in Sarcopenia Detection and Management: A Comprehensive Survey

“…While machine learning techniques show promise for improving sarcopenia definition and diagnosis [ 2 ], several key challenges remain:…”

“…However, applying machine learning for sarcopenia detection faces obstacles like data accessibility, imbalance, and feature selection. High-quality, representative training and testing data are needed but scarce [ 2 ]. Imbalanced class distributions can skew predictions.…”

Metaheuristic-Based Feature Selection Methods for Diagnosing Sarcopenia with Machine Learning Algorithms

“…Machine learning has emerged as a potent tool for enhancing sarcopenia definition and diagnosis, uncovering patterns beyond human observation, and gaining deeper insights [ 9 , 10 ]. Ongoing research explores machine learning’s role in diagnosing and understanding sarcopenia and promising results have been achieved using various datasets and methodologies [ 11 , 12 ].…”

“…However, limitations for traditional studies, particularly in machine learning methodologies, are being debated. The issues include limited data access, imbalanced data, and feature selection complexities that pose challenges, especially with excess features impacting diagnostic accuracy [ 6 , 9 ]. Furthermore, related work also has certain limitations.…”

Correction: Use machine learning to help identify possible sarcopenia cases in maintenance hemodialysis patients

“…Following publication of the original article [ 1 ], we have been informed that authors Hualong Liao, Yujie Yang, Ying Zeng, Ying Qiu, Yang Chen, Linfang Zhu, Yu Chen and Huaihong Yuan were incorrectly affiliated. Also, we have been informed that the spelling 'F1 Sore' from Fig.…”