Predictive model development for possible sarcopenia in community-dwelling older adults: a cross-sectional machine learning approach using the Korean frailty and aging cohort study.

BACKGROUND

Sarcopenia, an age-related decline in muscle mass and physical function, is a major risk factor for frailty, a condition associated with negative health outcomes and increased disease burden in older adults. Providing simple, accurate community-based screening is essential for early prevention and improving the health and quality of life of older adults. Employing screening criteria for possible sarcopenia can broadly identify individuals at risk for sarcopenia and enhance early diagnosis and preventive measures. However, there is a lack of possible sarcopenia prediction models. This study developed and evaluated a model for predicting possible sarcopenia among community-dwelling older adults and identified key predictors.

METHODS

A supervised machine learning approach was used, with data from the 2022-2023 Korean Frailty and Aging Cohort Study (n = 1,761). Individuals were classified as having possible or no possible sarcopenia based on the 2019 Asian Working Group for Sarcopenia criteria. Logistic regression, random forest, support vector machine, and extreme gradient boosting machine learning models were developed, and their predictive performance was assessed using accuracy, precision, recall, F1-score, and receiver operating characteristic curve-area under the curve. Feature importance was analysed applying Shapley additive explanations.

RESULTS

The final sample comprised 500 individuals with possible sarcopenia (mean age: 83.0 ± 3.76 years; 34.4% men) and 1,261 without possible sarcopenia (mean age: 81.0 ± 3.40 years, 51.6% men). Logistic regression demonstrated the best predictive performance among the four models, with the highest recall of 0.700 and F1-score of 0.654. The most influential predictors for possible sarcopenia were lower body mass index, walking aid use, cognitive impairment, older age, and exhaustion.

CONCLUSIONS

Multidomain geriatric indicators including anthropometric status (body mass index), walking aid use, cognitive function, age, and exhaustion can guide pragmatic, community-based screening for possible sarcopenia. Simple, accessible assessments of these predictors may facilitate earlier identification and referral, and should be considered in sarcopenia screening and prevention strategies.

CLINICAL TRIAL NUMBER: Not applicable.