Machine Learning-Based Depression Risk Prediction Models for Older Adults Analyzing From the Perspective of the...

This review systematically examined the current research on machine learning-based depression risk prediction models for domestic and international older adults, aiming to provide references for methodological development and application in this field.

A scoping review approach was used, following the Participants, Concept, and Context (PCC) framework and incorporating the Health Ecology Model (HEM) as the analytical theoretical framework.

Based on the PCC principles, the study scope was defined, and a systematic search was conducted in Web of Science, PubMed, Cochrane Library, Embase, CINAHL, CNKI, VIP Database, Wanfang Database, and the Chinese Biomedical Literature Database for literature related to depression in older adults, covering publications up to October 12, 2025. Two researchers independently screened the literature, extracted data, and summarized and analyzed the papers' content.

After screening and full-text review, 15 studies on machine learning-based depression risk prediction models for older adults were included. A total of 90 risk prediction models were covered, with area under the curve (AUC) values ranging from 0.73 to 0.943. Most models were only internally validated, with only 20 models undergoing external validation. Predictors of depression in older adults were mainly at the individual intrinsic traits level (e.g. age, gender, number of chronic diseases, cognitive scores, pain) and the behavioral characteristics level (e.g. sleep duration, physical activity, smoking, drinking) of the HEM. At the same time, interpersonal network, community, environmental, and policy-level factors were less involved. These models are applicable in various settings, including community health service centers, outpatient clinics, long-term care institutions, and home health management.

Research on depression risk prediction models for older adults in China is still in its early stages. Existing models demonstrate good predictive performance, with a manageable risk of bias, and can provide more reliable decision support for healthcare professionals.