Applying Supervised Machine Learning to Identify Which Patient Characteristics Identify the Highest Rates of...

Area of Science:

• Health Informatics
• Clinical Data Science
• Patient Outcomes Research

Background:

• Interhospital transfers are common and associated with significant patient mortality.
• Identifying high-risk patient groups post-transfer is crucial for targeted interventions.
• Traditional methods may not capture complex interactions of patient characteristics influencing mortality.

Purpose of the Study:

• To apply supervised machine learning (ML) to identify patient groups with elevated mortality post-interhospital transfer.
• To demonstrate the utility of ML in analyzing complex patient data for risk stratification.
• To pinpoint specific patient characteristics associated with increased post-transfer mortality.

Main Methods:

• Cross-sectional analysis of the 2013 National Inpatient Sample (Health Care Utilization Project).
• Supervised ML: Classification and Regression Trees (CART) for group identification.
• Supervised ML: Random Forest for determining characteristic importance in mortality.

Main Results:

• Identified 21 distinct patient groups, with 13 showing at least double the national average post-transfer mortality rate.
• Key factors influencing mortality included circulatory disorders, coagulopathy, cancer diagnosis, and advanced age.
• ML effectively identified high-risk subpopulations based on combinations of characteristics.

Conclusions:

• Supervised ML provides a computationally feasible approach to identify high-mortality patient groups post-transfer.
• Findings can inform the development of clinical decision support systems for interhospital transfers.
• This approach enhances the ability to stratify risk and potentially improve patient outcomes.