Using Artificial Intelligence to Obtain More Evidence? Prediction of Length of Hospitalization in Pediatric Burn Patients

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Pediatric Burn Care

Background:

• Accurate prediction of hospital length of stay (LOS) is crucial for pediatric burn patient management and counseling.
• Existing heuristic models, like the rule of thumb (1 day per % TBSA), serve as benchmarks for predictive accuracy.
• The study evaluates advanced computational models against this benchmark using real-world data.

Purpose of the Study:

• To compare the prediction accuracy of an artificial intelligence (AI)-based approach and a linear model for pediatric burn patient length of stay.
• To establish a performance benchmark using a common heuristic (1 day per % TBSA).
• To identify factors influencing prolonged LOS and analyze the relationship between TBSA and prediction error.

Main Methods:

• Utilized a large international pediatric burn patient dataset (N = 8,542).
• Calculated mean absolute error and standard error for prediction models: rule of thumb, linear regression, and random forest.
• Analyzed factors contributing to prolonged LOS and the correlation between TBSA and residual error.

Main Results:

• Both the random forest AI approach and the linear model demonstrated statistically significant superiority over the rule of thumb (p < 0.001 and p = 0.009, respectively).
• Residual error increased with rising TBSA across all evaluated methods.
• Total body surface area (TBSA), burn depth, and inhalation trauma were identified as key factors associated with prolonged LOS.

Conclusions:

• AI-based algorithms show significant promise for medical prediction using large-scale international registry data.
• While powerful, the application of AI requires careful consideration of data set quality and inherent model limitations.
• Further development and validation are needed to optimize AI for clinical prediction in complex cases like pediatric burns.