Perioperative Workflow Variation Across Hospital Sites During Lower Extremity Angiographic Procedures Measured Using Ambient Computer Vision

Background Workflow variation contributes to differences in efficiency, outcomes, and cost. Despite growing interest in standardization, temporal data about perioperative phases are rarely captured systematically. We used an ambient computer vision system to quantify time spent in perioperative workflows across five hospital sites performing high-volume lower extremity interventions. Methods We conducted a retrospective, multisite observational study of perioperative workflows for lower extremity angiographic procedures across five hospitals within a single academic health system. Cases were identified using an ambient computer vision platform that integrates electronic health record data with automated event detection from operating room video. Each case was segmented into standardized perioperative phases, including anesthesia induction, patient preparation, final preparation, active procedure, postoperation, patient exit, room cleanup, and room setup. Phase durations were summarized using medians and interquartile ranges and compared across sites using the Kruskal-Wallis test. Multivariable linear regression models were used to estimate adjusted site level differences, controlling for surgeon, case timing, case classification and operational factors. Results A total of 916 cases were analyzed across five sites. All perioperative phases demonstrated significant intersite variation (p<0.05). The greatest variability was observed in patient preparation (median 9.0-23.0 minutes) and active procedure duration (40.0 – 75.0 minutes). Turnover phases varied substantially, with room cleanup ranging from 10.0 to 25.0 minutes and room setup from 20.0 to 28.5 minutes. Site C demonstrated prolonged preparation and patient exit times, while adjusted analyses revealed shorter active procedural duration relative to the system-wide mean. Sites B and E demonstrated longer adjusted total case durations (+15.1 and +17.9 minutes, respectively), whereas Site C and D demonstrated shorter adjusted total durations (-20.2 and -14.8 minutes). These differences translated into daily operating room time variation of approximately -26 to +27 minutes across sites. Conclusions Perioperative workflows vary substantially across hospital sites, even within the same health system. The greatest differences occurred in patient preparation, active procedure, and turnover phases. These findings suggest that inefficiencies extend beyond procedural performance and are driven in part by modifiable operational processes. Automated workflow analysis provides a scalable approach to identify phase-specific bottlenecks and support targeted improvements in operating room efficiency.