Artificial Intelligence-Driven Solutions to Evaluate and Optimize Operating Room Efficiency in Thoracic Surgery

Abstract

A critical area for improving healthcare spending is the operating room (OR), where inefficiencies contribute to up to 30% of surgical costs. Enhancing OR efficiency requires the identification of key factors that influence timely case completion. This project applies machine learning (ML) techniques to identify and rank key factors linked to surgical delays in thoracic surgery, and establish time benchmarks that improve the rate of on-time completion of surgical cases to optimize overall efficiency.

Data from 3738 lung resections from April 2008-April 2024 were analyzed using descriptive statistics and correlation analyses to determine associations between individual factors and the so-called "Surgical Success Rate" (SSR), defined in our studies as the proportion of on-time case completions over total OR days. The overall SSR was 43.0%. Lower SSR was associated with cases involving male patients with higher BMI, elevated ASA class, and those requiring prolonged Anesthesia Prep Time (APT), Procedure Time, Surgical Finish Time (SFT), or Turnover. SSR varied widely between surgeons and anesthesiologists independent of historical case volume.

Next, supervised ML techniques were used to validate and rank factors influencing efficiency. Key time intervals (APT, Surgical prep time (SPT), Procedure) and select team members (primary anesthesiologist and Scrub nurse #3) were consistently selected as important predictors of SSR. ML modeling of time-related features demonstrated superior predictive accuracy compared to patient or team metrics. Decision tree (DT) modeling was used to generate benchmark time interval scenarios which led to increased predicted SSR.

Finally, a comparative analysis of predictors of efficiency in thoracic surgery versus arthroplasty revealed that mean duration and variability of time intervals was greater in thoracic surgery. Shorter surgical preparation time (SPT) and SFT in arthroplasty indicate a more streamlined, standardized approach to these phases by the team. SSR in arthroplasty was more sensitive to SFT, AFT and turnover than in thoracic surgery.

Overall, this work lays the foundation for developing a prescriptive decision-support system capable of monitoring case progression in real-time and guiding the surgical team to address potential delays, ultimately enhancing efficiency in thoracic surgical care.