A Machine Learning Solution to Predict Surgical Case Duration in a Multi-Hospital Health System

• Activity Number: 197 - SPAAC Poster Competition
• Type: Topic Contributed
• Date/Time: Monday, August 8, 2022 : 2:00 PM to 3:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Eric D Hixson* and Zachary Yachanin and Joseph Dorocak and Michael Lewis
• Companies: Cleveland Clinic and Cleveland Clinic and Cleveland Clinic and Cleveland Clinic
• Keywords: machine learning; healthcare; hospitals; surgery; scheduling; efficiency

Abstract

Surgical scheduling and staffing benefit from accurate surgery duration prediction. Current estimates (i.e. ‘intuition model’) exist but lack operationally utility. Inpatient and ambulatory surgeries were included from a single health system in 2019–2021 (sites=22, n=543,092); and machine learning models developed to predict surgical duration. Features include site, surgeon, surgery, demographics, diagnoses, physiology, active medications, and lab results. The data set was split 50:50 train and validate. Training was further partitioned 50:50 to transform multi-dimensional features in one and then embed as learned features in the second for final training. Gradient boost models were developed for inpatient and ambulatory surgeries and performance compared to the intuition model. The inpatient model R2=0.79 and was 23% better than intuition (MAE 51 vs. 40 mins, K-W p< 0.001). Ambulatory R2=0.78 and was 24% better (MAE 24 vs. 19 mins, K-W p< 0.001). Automated procedures score and report out subsequent weeks’ scheduled cases at all locations. More accurate planning ensures optimal utilization of scarce physical and human resources and efficient operations.