To accommodate weekday and seasonal variations in case volume, continual adjustment of operating room resources is imperative.
Unlike other health systems, at Vanderbilt Health the OR planning process taps into predictive analytics running automatically on the back of the OR scheduling system.
With 60 adult operating rooms, the average weekday adult surgical case count at Vanderbilt Health is currently 150, but on any given Tuesday or Wednesday the count might surge as high as 170, and between Christmas and New Year’s or on the Friday after Thanksgiving it might fall as low as 50.
In 2013, a small team from Perioperative Services led by Warren Sandberg, MD, PhD, professor and chair of the Department of Anesthesiology, and Vikram Tiwari, PhD, MBA, who was then a faculty recruitment candidate, gathered to consider whether it would be helpful to model this fluctuation and distribute predictions to managers on a routine basis.
“While much of the monthly volume variability can be explained by annual and weekly patterns that are more or less clear, it seemed likely that historical data could reveal less obvious day-to-day regularities that could inform the process of resource allocation and help make the OR more efficient, and that’s indeed what we found,” said Sandberg, who also holds faculty appointments in Biomedical Informatics and Surgery. “And while we were at it, we landed a critical faculty member who plays a unique role in our system’s success to this day.”
The team evaluated various models created by Tiwari, now associate professor of Anesthesiology and director of Surgical Business Analytics. The approach that worked best was a days-out model using simple linear regression on booked cases. Based on nine months’ worth of case data, Tiwari created five ordered series of 30 linear models, one for each weekday, starting at T minus 30 days and growing more accurate toward T minus 1.
They presented their results in a July 2014 report in the journal Anesthesiology. Within seven days of the day of surgery, their method predicted the adult case count (elective and non-elective cases) to within seven cases 80% of the time. The authors also found that the model allowed improved resource allocation, including a savings of attending anesthesiologist clinical effort translating to 2.8 full-time faculty positions per year.
“To our knowledge no other medical center had mathematized their case volume predictions to any degree. I continue to receive queries from centers around the country that want to emulate our method,” said Tiwari, who also holds faculty appointments in Biomedical Informatics and the Owen Graduate School of Management.
Today the outputs from Tiwari’s separate models for adult and pediatric weekday OR case counts are distributed automatically to selected managers and are posted to the electronic health record system. For a given upcoming OR day, the trend lines defined by successive predictions are cast against the recent historical average for that weekday, allowing readers to easily spot upcoming fluctuations from average. While managers continue to flex staff one or two days in advance based on actual booked cases, the predictions allow advance adjustments that smooth the resource allocation process.
“As early as 14 days out, if we happen to see a prediction that’s way below expectations, we might start selectively closing down operating rooms. And by seven days out we can anticipate what our final staffing levels will be,” said David Wyatt, PhD, RN, vice president, Perioperative Enterprise. Wyatt and his managers use various levers to flex staffing, such as a float pool of experienced surgical nurses and agreeableness among regular staff to switch to an alternate workday on short notice.
While Anesthesiology routinely flexes attendings, residents and nurse anesthetists one day in advance based on booked cases, “as holidays approach the predictions are helpful as we begin weighing additional vacation requests. And throughout the year, whenever anesthesiology attendings request an academic or administrative day within 14 days of the date of service, we can look at the volume prediction to learn whether or not we can honor the request,” said Amy Robertson, MD, MMgt, vice chair for clinical affairs in the Department of Anesthesiology.
Sandberg recently acquired an added measure of confidence in the model’s accuracy as it approaches the 30-day horizon. “It’s a bit like having a time machine,” he said. “Last year, January, February and March were light OR months. This year, we could tell at the beginning of each month that we would actually finish the month ahead of budget, and indeed we have, each time. Regrettably, I lost a small bet to [Surgeon-in-Chief] Seth Karp over volumes. I should have trusted the model more!”
Pharmaceutical Services is among other departments that use the predictions, in their case to gauge upcoming demand for certain compounded medications used by some surgical patients.
