Regularized regression outperforms trees for predicting cognitive function in the Health and Retirement Study.

Area of Science:

• Machine learning in healthcare
• Cognitive decline research
• Biostatistics

Background:

• Generalized linear models (GLMs) are favored in healthcare for interpretability.
• Tree-based models (e.g., random forest, boosted trees) excel in predictive performance but lack transparency.
• Clinical applications require interpretable models for patient understanding and actionable insights.

Purpose of the Study:

• To utilize machine learning (ML) for detecting cognitive decline.
• To enable timely screening for cognitive impairment.
• To uncover associations between cognitive decline and psychosocial determinants.

Main Methods:

• Employed data from the 2018-2020 Health and Retirement Study.
• Developed three linear regression and three tree-based models.
• Evaluated model performance using RMSE and R-squared, and interpretability via coefficients, variable importance, and decision trees.

Main Results:

• Elastic net regression achieved the best performance (RMSE = 3.520, R² = 0.435).
• Baseline cognitive function and computer use frequency emerged as the most significant predictors across all models.
• Linear models, particularly elastic net, showed better performance than tree-based methods for this dataset.

Conclusions:

• Elastic net regression outperformed tree-based models in predicting cognitive outcomes.
• Additive linear relationships appear optimal for modeling cognitive decline.
• Elastic net's feature selection balances interpretability and predictive power for cognitive health data.