# “Could She/He Walk Out of the Hospital?”: Implementing AI Models for Recovery Prediction and Doctor-Patient Communication in Major Trauma

## TL;DR

This study uses AI to predict recovery in major trauma patients and improves doctor-patient communication through a hospital-integrated web app.

## Contribution

The novel contribution is the development and deployment of an AI model (XGBoost) for recovery prediction in trauma patients, integrated into a hospital system.

## Key findings

- XGBoost outperformed traditional clinical scores like ISS and GCS with an AUC of 0.748.
- The AI model was integrated into a hospital information system as a web-based application.
- Clinical use improved efficiency and interpretability, with positive feedback from healthcare professionals.

## Abstract

Background and Objectives: Major trauma ranks among the leading causes of mortality and handicap in both developing and developed countries, consuming substantial healthcare resources. Its unpredictable nature and diverse clinical presentations often lead to rapid and challenging-to-predict changes in patient conditions. An increasing number of models have been developed to address this challenge. Given our access to extensive and relatively comprehensive data, we seek assistance in making a meaningful contribution to this topic. This study aims to leverage artificial intelligence (AI)/machine learning (ML) to forecast potential adverse effects in major trauma patients. Methods: This retrospective analysis considered major trauma patient admitted to Chi Mei Medical Center from 1 January 2010 to 31 December 2019. Results: A total of 5521 major trauma patients were analyzed. Among five AI models tested, XGBoost showed the best performance (AUC 0.748), outperforming traditional clinical scores such as ISS and GCS. The model was deployed as a web-based application integrated into the hospital information system. Preliminary clinical use demonstrated improved efficiency, interpretability through SHAP analysis, and positive user feedback from healthcare professionals. Conclusions: This study presents a predictive model for estimating recovery probabilities in severe burn patients, effectively integrated into the hospital information system (HIS) without complex computations. Clinical use has shown improved efficiency and quality. Future efforts will expand predictions to include complications and treatment outcomes, aiming for broader applications as technology advances.