44: Developing Models for Mortality and Community Discharge After Skilled Nursing Facility Admission

Music City Center 

Prognostic models for older adults admitted to a skilled nursing facility (SNF) following hospitalization are needed to guide clinical decisions. Using 20% Medicare data (2017-2019), we developed models predicting 6-month mortality and community discharge post-SNF. A hybrid approach combined machine learning (Gradient Boosting, Random Forest, Neural Networks, SuperLearner) for feature selection with Bayesian logistic regression to estimate inclusion probabilities, quantify uncertainty, and compute credible intervals for individual risk predictions. Performance was assessed by discrimination (C-statistic) and calibration. Model outputs include Bayesian odds ratios with 95% credible intervals, which represent the range within which the true odds ratio lies with 95% probability. The final model provides interpretable effect estimates and predictive uncertainty, making it a valuable tool for SNF clinicians.

Prognostic Models

Prediction

Bayesian

Machine Learning

Skilled Nursing Facility 

Section on Statistical Learning and Data Science