Machine Learning Algorithms for Predicting Injurious Fall Risk Among Older Adults With Depression: A Prognostic...

Area of Science:

• Gerontology
• Data Science in Healthcare
• Public Health

Background:

• Falls and related injuries (FRI) represent a significant health burden for older adults experiencing depression.
• Current prediction models lack the ability to adapt to dynamic health changes over time, limiting proactive intervention.

Purpose of the Study:

• To develop and validate machine learning algorithms for predicting 3-month FRI risk in older adults with depression.
• To compare the performance of elastic net, random forest, and gradient boosting machine models.

Main Methods:

• Utilized a national cohort of fee-for-service Medicare beneficiaries aged 65+ with depression diagnoses.
• Employed 261 time-varying predictors updated every 3 months to forecast subsequent 3-month FRI risk.
• Assessed model performance using c-statistics and risk stratification.

Main Results:

• The random forest model achieved a c-statistic of 0.68, capturing 68.9% of FRI cases within the top risk deciles.
• Key predictors for FRI included frailty, age, previous FRI history, and antidepressant dosage.
• Low-risk individuals (bottom seven deciles) exhibited minimal FRI incidence (<1.7%).

Conclusions:

• A practical, time-varying prediction model effectively identifies older adults with depression at high risk for falls and related injuries.
• This dynamic approach supports clinical decision-making and optimizes the deployment of fall prevention resources.
• The model's ability to be updated over time enhances its utility for ongoing risk management.