Sarcopenia feature selection and risk prediction using machine learning: A cross-sectional study.

Area of Science:

• Gerontology
• Medical Informatics
• Nutritional Science

Background:

• Sarcopenia poses a significant health risk in aging populations.
• Identifying key risk factors is crucial for early detection and intervention.
• Machine learning offers potential for developing predictive models.

Purpose of the Study:

• To evaluate the utility of machine learning (ML) for identifying sarcopenia risk factors.
• To develop and assess ML-based predictive models for sarcopenia in older adults.

Main Methods:

• Utilized medical records from Korean adults aged 65+ from the Korea National Health and Nutrition Examination Surveys.
• Employed machine learning algorithms including support vector machine, random forest (RF), and logistic regression for model construction.
• Performed receiver operating characteristic (ROC) curve analysis to evaluate model performance.

Main Results:

• Identified distinct sets of top 10 risk factors for sarcopenia in men and women, including body mass index (BMI), blood counts, nutrient intake, and blood markers.
• Random forest models proved effective for feature selection.
• ML models demonstrated comparable predictive performance across genders, with no significant differences in area under the ROC curve.

Conclusions:

• Machine learning, particularly RF, is valuable for selecting sarcopenia risk factors.
• Combining RF feature selection with expert knowledge and multi-model validation enhances predictive accuracy.
• Further validation of the developed prediction models in additional studies is recommended.