Machine Learning Approaches to Determine Feature Importance for Predicting Infant Autopsy Outcome

Area of Science:

• Medical informatics
• Forensic pathology
• Pediatric pathology

Background:

• Sudden unexpected death in infancy (SUDI) is the leading cause of postneonatal mortality.
• Predicting autopsy outcomes in SUDI cases is crucial for identifying causes of death.

Purpose of the Study:

• To explore the utility of machine learning (ML) in predicting pediatric autopsy outcomes.
• To derive data-driven insights for identifying factors contributing to explained or unexplained infant deaths.

Main Methods:

• Analysis of a large pediatric autopsy database (>7,000 cases, >300 variables).
• Application of machine learning models including decision trees, random forests, and gradient boosting.
• Iterative training and evaluation of models based on autopsy outcome (explained vs. unexplained).

Main Results:

• Utilized data from 3,100 infant and young child (<2 years) autopsies.
• Achieved 80% predictive performance using the XGBoost model.
• Identified age at death, cardiovascular, and respiratory histological findings as critical predictors of explained death.

Conclusions:

• Machine learning is feasible for evaluating complex medical procedures like pediatric autopsies.
• Highlights the importance of standardized clinical data collection for ML applications.
• This ML approach has potential applications in various healthcare scenarios.