Medicine requires a strong evidence base for the actions that physicians take, but there are substantial lacunae in existing data analysis.
In a bid to address these gaps, Jay A. Redan, MD, FACS, a minimally invasive surgeon from AdventHealth Celebration in Florida, and Rohan A. Joseph, MD, FACS, a general surgeon from HCA Florida Healthcare in Tallahassee, comoderated yesterday’s Panel Session, “Correlating Video-Based Analytics with Surgical Outcomes.”
The session sought to define video metrics and explain how surgeons might record, interpret, and implement video data in their own ORs. The premise included the acknowledgment that such efforts are not new but are growing with the rise of artificial intelligence (AI).
Two presenters offered insights based on videos from wall-mounted cameras inside ORs.
Dr. Redan’s presentation highlighted discrepancies between electronic health records and what can be seen on video. He noted that hospital staff can inaccurately estimate timing data; in 30% of case-length data, total OR time was underestimated by more than 60 minutes. Because OR schedules are based on such data, errors create inefficiencies when ORs sit empty or overbooking triggers delays.
Dr. Redan identified several other areas of potential improvement, including late entry of the surgeon to the OR and delayed access to surgical materials. All are areas where AI-based video analysis can aid hospitals to improve OR use. “Once we start collecting accurate information, we can all work together to be efficient,” Dr. Redan said.
Linnette Johnson, RN, MSN, the chief clinical operations officer of the AdventHealth Central Florida Division, offered further insights based on her hospital system’s current pilot of video-based analysis of deidentified data captured via cameras on OR walls.
Johnson noted some resulting data offered insights impossible to casually observe, such as an 11-minute median variation in timestamps. In an OR with five cases a day, that variability suggests a total lag of 55 minutes, enough to serve “another patient our care could have impacted.”
She also shared that 70% of variation in OR turnover could be explained by the hour of the day, with lunchtime and shift-change times of peak inefficiency that could be addressed with flexible workflows. Her hospital also has identified other uses for video analysis, including training employees and enhancing safety by identifying behaviors associated with infections and other complications.
Two presenters offered insights based on videos showing close views of surgical technique.
Daniel Donoho, MD, a neurosurgeon at Children’s National Hospital and George Washington University in Washington, DC, and founder of the nonprofit Surgical Data Science Collective, focused his presentation on AI usage, saying, “I strongly believe that AI-based analysis can address the missing link between process and outcome by automating process measurements.”
He offered an example: carotid artery injury management. Such injuries occur in one in every 200 pituitary surgeries, are life-threatening, and involve working with sightlines compromised by blood flow. “We don’t create these intraoperatively to train surgeons, which means the first time you see this is often the first time you have to treat it,” he added.
“These pieces of data are profoundly useful if you want to predict blood loss and if you want to know if someone is going to be successful at a task,” he said, citing his own video-based research. “If you have all those underlying kinematic data, maybe you can develop some kind of predictive system” to help surgeons better address bleeding.
The same kind of videos could be used for training, including integration into simulation-based learning programs, he said.
Dr. Donoho also noted the unmet need for surgical care worldwide, combined with population pressures that will reduce the surgeon workforce in coming decades, create a compelling case for video-based training in low- and middle-income countries.
Like Dr. Donoho, Filippo Filicori, MD, a minimally invasive surgeon and AI specialist at Northwell Health in New Hyde Park, New York, presented on capturing surgeons’ fine movements. His research has included calculating variables such as movement arrest period ratio and speed peaks, which correlate with case difficulty and can help surgeons with “knowing when you are struggling automatically without even asking for help.”
Finally, Bruce Ramshaw, MD, a former chair of surgery at The University of Tennessee Knoxville and current chief medical information officer of Caresyntax, took a philosophical tack, remarking on the fragmented and imperfect nature of medical data and the need for humility in adopting more refined objective data analysis toward improved performance.
Throughout the session, all panelists suggested the importance of the human element to the successful use of analytics, whether via executives’ buy-in, surgical team training, or computer-human collaborations. During the Q&A, Dr. Redan offered a comment that summarized this aspect of the session: “The key word is collaboration.”
