Predicting depression in old age: Combining life course data with machine learning.

Area of Science:

• Gerontology
• Computational Social Science
• Public Health

Background:

• Aging populations worldwide necessitate proactive strategies to address age-related health issues like late-life depression.
• Understanding the multifactorial determinants of depression in older adults is crucial for effective healthcare planning and cost reduction.

Purpose of the Study:

• To develop and compare machine learning algorithms for predicting the risk of depression in old age.
• To identify key life course factors and childhood conditions that contribute to late-life depression.

Main Methods:

• Utilized the Survey of Health, Ageing and Retirement in Europe (SHARE) dataset.
• Implemented and evaluated six supervised machine learning algorithms, focusing on sequence data for life course representation.
• Employed Shapley Additive Explanations (SHAP) for pattern identification.

Main Results:

• Machine learning models demonstrated comparable predictive abilities across different algorithms.
• The highest predictive performance was achieved using semi-structured life course representations via sequence data.
• Identified age, health status, childhood conditions, and low education as primary predictors.
• Discovered novel predictive patterns related to life course instability and low dental care utilization.

Conclusions:

• Machine learning, particularly with sequence data, effectively predicts late-life depression risk.
• Life course instability and reduced dental care access are significant, previously underappreciated, risk factors for depression in older adults.
• Findings can inform targeted interventions and public health policies for aging populations.