Machine Learning Innovations for Improving Access to Surgical Care

Hospitals are a low margin business. Yet their most profitable resource, their operating rooms, are consistently under utilized. In fact, they typically go unused over 30% of the time despite accounting for almost 40% of a hospital’s revenue. Furthermore, it’s not just hospital balance sheets that suffer as patients typically wait several months, often in pain, for their surgeries and in many cases can have their health deteriorate while waiting. Even surgeons wanting to operate more frequently are not able to find operating room time — and often take cases to other hospitals when they cannot be accommodated.

Unfortunately, in many ways the COVID 19 pandemic has worsened this problem. For months, hospitals around the country either outright canceled or limited the number of scheduled surgeries due to concerns about not being able to admit those patients due to high COVID 19 patient volumes. This led to a number of patients who had delayed operations. Now, this backlog is putting even more pressure on the system as COVID spikes have subsided. These delays have caused even longer lead times for surgeries and hurt hospital finances.

How is it possible that these operating rooms are so under utilized? A number of factors are at play:

• Unused block time. Block schedules are specific times given to either individuals or groups of surgeons. For example, the Ortho group can have Operating Room 7 every Thursday between 8am and 6pm. This helps create a balanced schedule for surgeons but often leads to unused time. For instance, if a surgeon has two, six hour orthopedic cases that need to be scheduled, they will have to be done on separate Thursdays above. And if the surgeon does not have shorter cases to squeeze in, four hours in each of these days can go unused.
• Variable demand. There are some variations in the types of surgeries that occur over time, such as seasonal variation in planned orthopedic surgeries It’s possible that there just are not that many patients who need a particular type of surgery, which can create gaps on the schedule.
• Uncertainty around case length. Patients are not homogeneous and accordingly surgeries aren’t either. Typically each case has a likely distribution for how long it will run and often the time used for scheduling is a conservative estimate.
• Uncertainty around emergency surgeries. Often, hospitals have standard protocols to reserve particular surgical rooms for unplanned emergency surgeries. These can go unused but will mitigate the possibility of not being able to operate on a patient in need.

Given the complexity of these challenges, it’s not realistic to expect zero unused time — but there’s still a massive opportunity to increase access and utilization.

In late 2020, a multidisciplinary team at Qventus started to explore how we could use our technical platform to reduce the amount of unused time in these operating rooms. We have a thorough evaluation process to assess new use cases for our platform, and we believed there was an enormous opportunity to apply our capabilities to address this challenge. It is also worth noting that not all problems are necessarily addressable via technology, but we were able to find some compelling places to start to break the problem down into pieces we could address with our platform.

First, we realized that some portions of unused time can be booked well in advance. For instance, certain blocks of operating room time did not get used well consistently. Because block schedules are set relatively infrequently, differences between surgical demand when they were created and when they are actually used can actually grow over time. To take it one step further, often we would see groups of surgeons with particular temporal patterns in the way they scheduled cases into their blocks. For some types of cases, if they are not scheduled at least two weeks in advance, they will have to be delayed.

With this insight, we started to develop a supervised machine learning model to ultimately predict what parts of the surgical schedule would go unused. By continuing to iterate on features with our health system partners, we were able to achieve significant model performance boosts — and then we layed in the overarching intervention that prompted owners of these block times to give up time far enough in advance so that these unused times could be filled in by other surgical cases.

Second, we realized that many surgeons wanted to operate more often but were not able to find time on the hospital’s OR schedule. In many cases, they would manually call up the OR scheduling team to see if there was any time available in the coming weeks or months and then go through herculean efforts to try to align the available time with their patients’ schedules. Often, these surgeons would think no time was available when in reality scheduled time had actually opened up and would go unused.

Naturally, this seemed like a good place for technology to intervene. To start we developed an interface to surface available time and streamline the workflow between surgeon schedulers who wanted to book time and the operating room schedulers who had to handle these requests. Though simple in theory, in practice this workflow can actually be relatively complex and can vary from hospital to hospital. Fortunately, we were able to push through some of this complexity and even in early MVP sites were able to save hundreds of calls each month for surgeon schedulers

We also realized that not all matches between surgeons and open times are necessarily equal. From a surgeon’s standpoint, they might like to operate on particular days of the week or prefer to use surgical robots. Or, a hospital might prefer to prioritize accommodation for surgeons who have a track record of starting on time and getting open times that are early in the day. There are a number of non-aligning perspectives in this scoring, but once again through quick iteration, our interdisciplinary team was able to build a point of view that defines the best matches through data. We then use this to drive preferences in the interface above as well as proactively reach out to surgeons to ensure they know about time that is available.

These insights helped establish the core of our latest innovation: the Qventus Perioperative Solution. Through continual feedback from our users, we rapidly took this early MVP focused on product validation and developed it into a mature, proven product that we are now scaling to more health system partners. We are excited about bringing this to market and continuing to enhance the solution in order to increase surgical access at hospitals — making it easier for surgeons to schedule time and ultimately drive better outcomes for patients.