机器学习风险评估筛查与诊断肌少症:SARCO X 研究
该多中心横断面病例对照研究纳入≥45岁受试者5649人,其中19.4%为肌少症。基于年龄、体重、身高、教育水平、运动状态及高血压和糖尿病等因素构建的梯度提升分类器模型诊断肌少症的召回率为0.979、精确率为0.926、准确率为0.980,且较基础算法减少握力与超声需求分别达38.1%和49.5%。
要点
- 机器学习模型诊断肌少症召回率0.979、精确率0.926、准确率0.980。
- 机器学习增强算法减少握力需求38.1%、超声需求49.5%。
- 年龄、体重、身高、教育、运动及高血压糖尿病与肌少症显著相关。
核心结论
- 基于机器学习的算法可显著减少肌少症诊断中对检测和影像的需求。
- 该算法有助于在初级和二级医疗环境中识别肌少症并减少需转诊进一步评估的人数。
研究设计与样本
- 多中心横断面病例对照研究,纳入年龄≥45岁的受试者5649人(男性1379人,占24.4%)。
主要发现
- 5649名参与者中1097人(19.4%)为肌少症。
- 机器学习模型中梯度提升分类器在留出测试数据中预测肌少症表现最佳。
- 机器学习增强算法诊断肌少症的召回值为0.979、精确值为0.926、准确值为0.980。
- 与基础肌少症算法相比,机器学习增强算法分别减少握力和超声需求38.1%和49.5%。
关键数值与推荐
- 肌少症诊断采用基础算法和基于机器学习的算法,纳入低股四头肌质量/厚度,并结合延长椅子站立试验时长和/或降低握力。
- 与肌少症显著相关的因素包括年龄、体重、身高、教育水平、运动状态以及高血压和糖尿病的存在。
临床怎么用
- 机器学习算法可用于初级和二级医疗环境中肌少症的筛查和诊断,减少对握力和超声设备的依赖。
- 该算法有助于识别需转诊进一步评估的个体,优化医疗资源配置。
局限性
- 本研究为横断面病例对照研究,无法确定因果关系。
- 研究人群年龄≥45岁,结论外推至其他年龄人群需谨慎。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
OBJECTIVES: Sarcopenia imposes significant morbidity and economic burden on health care systems, underscoring the critical need for early/effective screening and diagnosis. This study aimed to develop a machine learning (ML)-based algorithm to facilitate the screening/diagnosis of sarcopenia. DESIGN: A cross-sectional case-control study. SETTING AND PARTICIPANTS: This multicenter study enrolled subjects aged ≥45 years. METHODS: Demographic data such as age, weight, height, education/exercise status, smoking, and comorbid diseases were obtained. Sarcopenia was diagnosed using the basic and ML-based algorithms, which incorporate low quadriceps muscle mass/thickness, combined with prolonged chair stand test (CST) duration and/or reduced hand grip strength (HGS). RESULTS: Of 5649 participants (1379 males, 24.4%), 1097 of them (19.4%) were sarcopenic. Using the ML-based model, significantly associated factors with sarcopenia were age, weight, height, education level, exercise status, and presence of hypertension and diabetes mellitus. Of the various ML models, the Gradient Boosting Classifier (GBC) demonstrated the highest performance in predicting sarcopenia in the holdout test data. For the ML-augmented algorithm, the recall value was 0.979; the precision value was 0.926, and the accuracy value was 0.980 for making the diagnosis of sarcopenia. When compared with the basic sarcopenia algorithm, the ML-augmented algorithm further decreased the need for HGS and ultrasound by 38.1% and 49.5%, respectively, demonstrating its effectiveness in optimizing sarcopenia diagnosis while minimizing testing required for medical device(s). CONCLUSIONS AND IMPLICATIONS: The ML-based algorithm significantly reduces the need for testing/imaging in the diagnosis of sarcopenia. It facilitates the identification of sarcopenia particularly in the primary and secondary care settings and decreases the number of individuals who should be referred for further evaluation.
引用
- Kara M, Ceran Y, Analay P, et al. Screening/Diagnosing Sarcopenia with Machine Learning-Powered Risk Assessment: The SARCO X Study. J Am Med Dir Assoc. 2025. PMID: 40383138. doi:10.1016/j.jamda.2025.105683
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。