Study: AI improves surgical case length predictions

Editor's Note

A recent study developed and validated an artificial intelligence (AI) model leveraging natural language processing (NLP) and machine learning to significantly improve prediction accuracy for surgical case length. Published November 29 in the journal Surgery, the findings show promise for using AI as an alternative to current methods that often fall short, researchers write.

Researchers applied three machine learning models: linear regression, CategoricalBoost, and feed-forward neural networks, each integrated with embeddings from B'ERT. Case data was divided into training, testing, and hold-out validation groups to ensure robust evaluation. Data included more than  125,000 elective inpatient surgeries performed between 2017 and 2023 at a quaternary care hospital.

“On average, the estimate improved by 62%,” researchers write. The AI model achieved a mean absolute error of 46.4 minutes, compared to error of 120.0 minutes (P < 0.001) electronic health record (EHR)-based estimates. In terms of prediction accuracy, the AI model correctly estimated case lengths within ±20% of the actual duration for 48% of cases, compared to 17% for the EHR system. This improvement represents a nearly 2.8-fold increase in accurate predictions.