Recommendations emerging from a data analytics project have helped OR leaders at Vanderbilt University Medical Center in Nashville, Tennessee, better anticipate daily surgical case volume and share that information with their managers. Data gleaned from the project are now being used to predict staffing needs for the OR, anesthesia department, and pathology department. Data are also being used for case cart preparation and inpatient bed management.
“It took us a little bit to get to a place where people believed our predictions were good,” says David Wyatt, MA, MPH, BSN, RN, CNOR, NEA-BC, associate operating officer and associate nursing officer, “but now there is a ‘hunger’ for us to help them with their planning and to help them understand their needs.”
Vikram Tiwari, PhDPredicting daily surgical volume was one of the first analytics projects tackled at Vanderbilt. There were a lot of problems in matching staffing needs with surgical volume, says Vikram Tiwari, PhD, assistant professor of anesthesiology & biomedical informatics and director, surgical business analytics, at Vanderbilt. “Staffing is planned weeks or months in advance, and the number of staff that are actually needed on a certain day may be completely different from what was planned.”
David Wyatt, MA, MPH, BSN, RN, CNOR, NEA-BCFor example, he noted that during a week in November, cases moved from a shortfall of 12 on Monday to an excess of 23 on Thursday, for a swing of 35 cases within 4 days.
Tiwari says he and his colleagues looked at this and questioned whether there was a signal in the schedule as it was building that could have told them Monday would be a lower-volume day and Thursday a higher-volume day.
Their analysis found that most of the time there is enough signal in the schedule as it’s building to predict future case volume.
They did regression modeling, using the number of cases booked as of a certain day for a future date as the independent variable and the final volume as the dependent variable. They now have 30 regression models, one for each of 30 days before the day of surgery.
“Even at 30 days out, we can see a linear trend,” says Tiwari. “The more cases booked 30 days out, the more cases there will be in the final volume. The closer to the day of surgery, the better the predictive value of the model,” he says.
These predictions now are compiled into a daily case report so that managers can start seeing from 14 days out what the final volume will be (sidebar above).
Tiwari says the next step was to see if they could use the same model to determine where a shortfall was coming from. “Looking at the daily case report, the manager can see low volume days 10 days out, see which rooms have no cases booked in them, and investigate that,” he says (sidebar, p 8).
There are key points to keep in mind when approaching any project that involves analytics, says Tiwari.
• Improve the analytical techniques to fine-tune and extend the initial recommendations. “Because it’s not an academic exercise, you will not find the best method right off the bat,” he notes. “Use the best techniques you have, one at a time, and fine-tune them for the next round.”
• Deliver information in a manner that enhances decision making and minimizes data exploration. “Doing the analysis is half the battle. The bigger challenge is how to get the analysis to the decision makers so they can use it. That means putting it in a format that is easily understood without overwhelming them. This involves the coding process and then visually representing the data,” he says.
• Increase engagement of frontline staff and managers in the specification phase of the analytics projects. “The people on the ground who are doing all the work need to feel they are embedded in the analytical approach, like the idea came from them. Otherwise there will not be buy-in,” says Tiwari.
Wyatt notes that when he first started seeing the daily case reports, he didn’t know what he was looking at. He had to learn what the graphs were showing him.
“I was used to looking at reports of averages and total frequencies and knowing what the report was going to tell me,” he says. “I wasn’t used to looking at data in a way that was going to inform my decisions.”
One thing managers at Vanderbilt are doing differently because of the data is that, instead of trying to fill low-volume days by shifting cases from high-volume days, they are looking at the high-volume days and trying to reproduce them. “That means we actually want to function at a higher level,” says Wyatt.
“It took a bit to get to a place where people believed the prediction was a good one,” says Wyatt. “But when the graph was added that showed them where the volume was, it increased confidence.”
Vanderbilt has a tool called “Provider Time Away.” Surgeons use it to inform everyone at least 6 weeks in advance if they are going to be out of town at a conference or on vacation or using personal time. The intention of the tool is to release the block time of the surgeon who is away.
“We wondered if we could look at the tool, understand which providers would be away, and use that information to predict case volume at the provider and service level,” says Tiwari. They had already found there was a direct correlation between the number of providers that were away during a week and the final volume for that week (sidebar, p 9).
Tiwari says they came up with a simulation model that used providers’ away information and their historical probability of operating on any day of the week to determine their theoretical probability of being in the OR any day of the week. Then they looked at how many cases the surgeons do when they are in the OR.
Because of error in the data (eg, the system says surgeons are away but they are actually in the OR), Tiwari says they used a technique to account for the probability of error. That probability is entered, and the data are run through a simulation model that then gives the number of cases that are expected to be done at a future date (eg, 42 days in advance) by a specific surgeon. All surgeons’ cases in a specific service can then be grouped together to find the service volume.
Using this model, Tiwari says they can predict a daily case volume for a particular service 82% of the time within plus or minus three cases. “The model definitely has a huge benefit,” he says.
“The predictive model has resulted in cost savings, but we haven’t been able to use the predictive model to increase revenue, which is frustrating,” says Wyatt.
The reason is that although the model is useful to manage and detect what cases were scheduled and predicted to be scheduled, it was not an effective way to change the surgeons’ scheduling patterns.
“Just because Wednesday had 120 cases and the capacity to do 150 or 160 didn’t mean the surgeons would schedule cases on that day if they typically worked on Mondays and Thursdays,” says Wyatt. “To them it was irrelevant.”
Wyatt says they are now devising a systematic way to identify excessive low-volume days and high-volume days (and the services/surgeons driving those trends) far in advance. Another challenge is to figure out how to persuade surgeons and their office staff to be more flexible about which days the surgeons work in the OR.
Wyatt says working with analytics has taught him there are patterns in virtually everything. “We may not always be able to detect them, but the patterns are there,” he says.
For example, says Wyatt, “We have false beliefs that there aren’t patterns and that people come in randomly to surgeons’ offices, and cases are randomly moved from there, but those patterns are very evident if you take the time to figure out what they are.”
Though some hospitals may not employ their surgeons or have a provider time away calendar, OR managers should not be deterred by such barriers. These should not prevent managers from doing any kind of thinking about what patterns there may be in the information they have, Wyatt says.
“That is what we are doing at Vanderbilt,” he notes. “We are proactively looking at things and planning rather than responding and reacting in a ‘way-too-late’ fashion.” ✥
Tiwari V, Wyatt D. Developing and sustaining successful outcomes for data analytics projects. Presented at OR Manager Conference 2015.