Machine learning models for identifying pre-frailty in community dwelling older adults

Area of Science:

• Gerontology and Public Health
• Computational Medicine and Machine Learning

Background:

• Pre-frailty is increasingly recognized as a condition that can manifest in middle age.
• Understanding early indicators of pre-frailty is crucial for developing timely preventive interventions in middle-aged and older adults.

Purpose of the Study:

• To identify factors associated with pre-frailty in community-dwelling adults aged 40-75 years.
• To utilize machine learning models for predicting pre-frailty status.

Main Methods:

• A cohort of 656 adults underwent comprehensive health assessments.
• Machine learning models and correlation-based feature selection were employed to identify pre-frailty indicators.
• Two stages of feature selection were performed, including physiological, anthropometric, environmental, social, and lifestyle variables.

Main Results:

• Machine learning models identified higher BMI, lower muscle mass, poorer grip strength and balance, increased distress, poor sleep quality, shortness of breath, and incontinence as associated with pre-frailty.
• The models achieved significant predictive performance, with Area Under the Curve (AUC) scores up to 0.817 for Fried Frailty Phenotype (FFP) and 0.722 for Clinical Frailty Scale (CFS).
• Feature selection enhanced model performance by up to 7.4% for FFP and 7.9% for CFS.

Conclusions:

• Machine learning is a suitable method for predicting pre-frailty.
• Identified factors provide valuable insights for targeted health assessments to detect pre-frailty in middle-aged and older populations.