The following is a summary of “High-performance pediatric surgical risk calculator: A novel algorithm based on machine learning and pediatric NSQIP data,” published in the JULY 2023 issue of Surgery by Bertsimas, et al.

For a study, researchers sought to develop a prediction model for pediatric surgical complications using machine learning based on data from the pediatric National Surgical Quality Improvement Program (NSQIP).

Data from all pediatric NSQIP procedures performed between 2012 and 2018 were reviewed. The primary outcome was defined as 30-day postoperative morbidity/mortality, which was further classified as any, major, and minor. Prediction models were developed using data from 2012 to 2017, and the performance of these models was evaluated independently using data from 2018.

A total of 431,148 patients were included in the training set from 2012 to 2017, and 108,604 patients were included in the testing set from 2018. The prediction models showed high performance in predicting mortality, with an area under the curve (AUC) of 0.94 in the testing set. Furthermore, the models outperformed the ACS-NSQIP Calculator in all categories for morbidity, with AUCs of 0.90 for major complications, 0.86 for any complications, and 0.69 for minor complications.

The study successfully developed a high-performing pediatric surgical risk prediction model using machine learning. The predictive tool had the potential to significantly improve the quality of surgical care for pediatric patients.

Source: americanjournalofsurgery.com/article/S0002-9610(23)00106-X/fulltext