New models of machine learning could improve the prediction of the duration of surgical cases

The operating rooms are a precious resource. They can account for 50% of a hospital's income and cost up to $ 80 per minute. But determining how much time to allocate to a surgical operation is a challenge that every hospital faces.

OR planning is a $ 5 billion problem. To optimize the operating room, you must answer a fundamental question: how long does each operation take? Under-utilization means fewer patients are receiving surgical care and the hospital has excess capacity. Overuse results in canceled trades and overtime expenses. "

Rajeev Saxena, Senior Author and Resident in Anesthesiology, Faculty of Medicine, University of Washington

In an attempt to improve the prediction of the duration of surgical cases, Saxena, in collaboration with surgical and anesthesia specialists, scientists and computer scientists from the University and the University of Washington, has created models of surgical procedures. automatic learning for each surgical specialty and for each surgeon.

Their study entitled "Improving the Efficiency of Operating Rooms: A Machine Learning Approach to Predict Treatment Duration" was published July 18 in the Journal of the American College of Surgeons.

The researchers collected data on more than 45,000 surgeries performed over four years by 92 surgeons. Their specific models to surgeons could improve accuracy from 30% (according to the surgeon's estimate) to 40%. In the upper third of surgeons, the accuracy has improved by more than 50%.

As more data is entered, the model will improve over time, said lead author Matthew Bartek, chief resident in general surgery at the UW School of Medicine.

"If we can improve the data, we can focus on more accurate estimates," Bartek said. "It's only the first step."

To create their estimates, the researchers collected data from several electronic medical records on the patient, the type of operation, the operation staff, and the operation planning information. Only preoperative data were used.

Of all the data captured, the greatest variability was observed between surgeons.

"Every surgeon has a unique approach to the operation and this data confirms it," said Bala Nair, former board author, former director of the technical center at the UW School of Medicine, who sought to improve care through computer and technological solutions.

"This work is bringing theaters into the 21st century by applying modern methods of data science to improve operations," he said.

"For years, operating rooms have been relying on the surgeon's estimates for operating times, but surgeons generally underestimate the time of the procedures," he said. Lead author John Lang, Clinical Director of Surgical and Perioperative Operations at UW Medicine.

According to Lang, two hospitals of the UW Medicine System are clbadified in the Medicare Case Mix Index for complex surgeries. The UW Medical Center is ranked # 3 and the Harborview Medical Center is ranked # 13. For this reason, accurate prediction of the duration of a surgical procedure is particularly difficult and impactful.

Saxena said, "You can change an entire organizational culture by adopting a data-driven approach and engaging key stakeholders."