Forecasting Pediatric Trauma Volumes: Insights From a Retrospective Study Using Machine Learning.

Area of Science:

• Pediatric trauma research
• Healthcare resource management
• Machine learning in medicine

Background:

• Pediatric firearm-related fatalities are increasing in the US, straining trauma centers.
• Accurate trauma volume prediction is crucial for resource allocation and preparedness.
• The study investigates the feasibility of daily trauma volume forecasting.

Purpose of the Study:

• To evaluate the accuracy of various machine learning models for predicting pediatric trauma center volumes.
• To compare model performance across monthly, weekly, and daily prediction intervals.
• To assess the real-world applicability of these forecasting models.

Main Methods:

• Retrospective analysis of 12,144 pediatric trauma patient records (2013-2023).
• Data grouped into monthly, weekly, and daily cohorts (21 groups total).
• Evaluation of 14 time-series forecasting models using standard metrics and real-world simulations.

Main Results:

• Monthly predictions generally showed higher accuracy than weekly or daily forecasts.
• The Silverkite model was superior for monthly predictions; 1D-CNN excelled in daily predictions.
• The Prophet model performed best for monthly real-world simulations; weekly prediction results were inconclusive.

Conclusions:

• Monthly forecasting of pediatric trauma volume is the most accurate approach.
• Model performance varied significantly across different data groupings and prediction intervals.
• While monthly predictions show promise, daily trauma volume forecasting needs substantial improvement for clinical utility.