Predicting Patient Mortality for Earlier Palliative Care Identification in Medicare Advantage Plans: Features of a...

Area of Science:

• Gerontology
• Health Informatics
• Machine Learning

Background:

• Machine learning (ML) offers enhanced precision for predicting end-of-life and palliative care needs in Medicare beneficiaries.
• Previous ML studies have not fully explored feature impacts or the role of social determinants of health.

Purpose of the Study:

• Develop a binary classification ML model to predict 1-year mortality in Medicare Advantage members aged 65 and older.
• Examine the key features influencing the predictive accuracy of the ML model.

Main Methods:

• A light gradient-boosted trees model was developed and validated using 5-fold cross-validation.
• The model utilized 907 features from claims and administrative data, including demographics and social determinants of health.
• Training data comprised 80% of cases (n=255,020), with 20% (n=63,754) held out for validation.

Main Results:

• The model achieved an AUC of 0.84 (95% CI 0.83-0.85) in predicting mortality.
• The top predictive features included patient demographics, diagnoses, pharmacy utilization, costs, and social determinants of health.
• The model accurately predicted 44.2% of expirations among the highest-risk 1% of patients.

Conclusions:

• The developed ML model effectively predicts end-of-life for Medicare Advantage members.
• The model leverages routinely collected data, enabling earlier identification for palliative care services.