Development and validation of a machine learning-based model for predicting intraoperative blood loss during burn...

Area of Science:

• Surgical Oncology
• Artificial Intelligence in Medicine
• Medical Informatics

Background:

• Intraoperative blood loss is a significant concern in burn patient care.
• Early identification of patients at high risk for substantial blood loss is crucial for surgical planning and patient safety.

Purpose of the Study:

• To develop and validate a predictive model for identifying patients at risk of substantial intraoperative blood loss (>750 mL) during burn surgery.
• To compare the performance of six machine-learning algorithms for this prediction task.

Main Methods:

• Collected demographic, laboratory, and surgical data from 395 burn surgeries.
• Developed six machine-learning models (logistic regression, decision tree, random forest, K-nearest neighbor, support vector machine, extreme gradient boosting) for predicting blood loss >750 mL.
• Validated models using internal and external cohorts, assessing performance with multiple metrics and creating a web-based calculator.

Main Results:

• The random forest model demonstrated superior predictive accuracy, achieving the highest overall performance scores in internal and external validation.
• Key predictors identified include initial hemoglobin, time to surgery, initial platelets, percentage of total body surface area excised/grafted, and surgery duration.
• A web-based application was created for accessible prediction of intraoperative blood loss.

Conclusions:

• The random forest model is an effective AI tool for predicting significant intraoperative blood loss in burn surgery.
• The developed web-based platform facilitates further data validation and model optimization.
• Identifying key clinical predictors enhances the ability to manage blood loss risks.