Study: AI, ML improve surgical control time estimation

Editor's Note

AI and machine learning (ML) models show significant promise in enhancing preoperative estimates of surgical control time (SCT), which are frequently wrong, according to a study published September 10 in Perioperative Care and Operating Room Management.

The longitudinal study examined differences between predicted and actual SCTs, broken down by surgical specialty, for more than 14,000 surgeries at a US military hospital. Surgeons consistently underestimated surgical control time (SCT), leading to scheduling inefficiencies, increased costs, and staff burnout, researchers write.

Among 13 specialties represented in the study, researchers observed the largest discrepancies in orthopedics, neurosurgery, and pain management. Only two—ear nose and throat (ENT) and pediatric dentistry—overestimated their SCTs, with the rest underestimating. According to researchers, underestimates can cause higher labor costs, canceled procedures, patient dissatisfaction, and staff burnout. Meanwhile, overestimations can result in idle OR time, reduced surgical volume, and lower productivity. Underutilization also increases cost per case by spreading fixed costs across fewer procedures.

“Unfortunately, the dynamic and unpredictable nature of surgical care makes accurate predictions difficult,” researchers write. “In fact, some researchers have suggested strictly using historical data and disregarding surgeons’ predictions for services that consistently underestimate case duration. Moving forward, evolving technologies such as artificial intelligence (AI) powered tools and machine learning (ML) models may help surgical administrators predict surgical case duration more accurately and optimize resource allocation.”

As an example of AI capability, researchers cite enhanced prediction of high-risk patients, which helps better forecast outcomes and guide personalized interventions. Additionally, ML accuracy has improved sufficiently in recent years to rival other methods in predicting surgical times and cancellations, particularly when incorporating institutional-, surgeon-, and patient-specific data. Short-term needs forecasting is also an opportunity. “For example, Jiao and colleagues demonstrated how a real-time neural network model can successfully use real-time intraoperative data to predict the case end times,” the researchers write. “This model performed significantly better than both surgical times forecasted by surgeons and a Bayesian predictive approach. A program powered by this model could be a powerful tool for OR managers to plan future staffing needs by the hour rather than by the day.”