Nonelective Rehospitalizations and Postdischarge Mortality: Predictive Models Suitable for Use in Real Time

Area of Science:

• Health Informatics
• Clinical Prediction Models
• Healthcare Management

Background:

• Hospital discharge planning is challenged by the absence of effective predictive tools.
• Accurate prediction of patient outcomes post-discharge is crucial for resource allocation and patient safety.

Purpose of the Study:

• To create predictive models for nonelective rehospitalization and postdischarge mortality.
• To ensure models are compatible with commercially available electronic medical records (EMRs) for real-time application.

Main Methods:

• Retrospective cohort study involving over 360,000 adult patients across 21 hospitals.
• Utilized split validation on a large dataset from an integrated health care delivery system.
• Developed four models predicting a composite outcome of nonelective rehospitalization and/or death within 7 or 30 days post-discharge.

Main Results:

• The best performing model (30-day discharge day) achieved a c-statistic of 0.756.
• Key predictors included a composite acute physiology score and end-of-life care directives, explaining 54% of the model's predictive power.
• Incorporating diagnoses did not enhance model performance due to real-time availability issues.

Conclusions:

• Robust predictive models for nonelective rehospitalization and postdischarge mortality can be developed.
• These models are suitable for real-time implementation within commercial electronic medical records (EMRs).
• The findings support the integration of predictive analytics into routine clinical workflows to enhance discharge planning.