Development of a Predictive Hospitalization Model for Skilled Nursing Facility Patients.

Area of Science:

• Healthcare Analytics
• Clinical Informatics
• Predictive Modeling

Background:

• Skilled nursing facilities (SNFs) face challenges in identifying patients at high risk for hospitalization or death.
• Accurate risk prediction is crucial for quality measures, financial penalties, and efficient clinical staffing.

Purpose of the Study:

• To develop and validate a predictive model for identifying SNF patients likely to be hospitalized or die within 7 days.
• To compare the model's performance against clinician judgment.

Main Methods:

• Retrospective multivariate prognostic model development using electronic health record (EHR) data from 5,642,474 patients across 8440 US SNFs.
• A machine learning model was trained on vital signs, diagnoses, lab results, food intake, and clinical notes.
• Model performance was compared to predictions made by SNF nurses and hospital case managers.

Main Results:

• The developed machine learning model achieved an area under the receiver operator curve (AUC) of 0.75.
• The model demonstrated a sensitivity of 35% and specificity of 92%.
• Clinician judgment showed higher sensitivity (61%) but lower specificity (73%) and positive predictive value (10%) compared to the model.

Conclusions:

• Machine learning models can accurately predict 7-day hospitalization or death risk in SNF patients.
• These models offer a valuable tool for readmission reduction programs without increasing SNF staff workload.
• Proactive, targeted monitoring of high-risk patients can be facilitated by these predictive tools.