AI in Medicine: An AI-based solution for scheduling clinician workload

Founder and CEO Suvas Vajracharya, Ph.D., had the opportunity recently to join Dr. Emma Nichols for an episode of the AI in Medicine podcast to discuss Lightning Bolt’s AI-based solution for scheduling physician shifts. Enjoy their conversation below or listen here.

Dr. Emma Nichols: Hello. This is the AI in Medicine weekly podcast. I’m Dr. Emma Nichols, and today I’m talking to Dr. Suvas Vajracharya who is CEO and founder of Lightning Bolt Solutions. And they were founded in 2002. The company’s dedicated to creating a balanced and sufficient workforce through optimized physician scheduling.

Across the U.S., Canada, Japan, and Australia, Lighting Bolt manages over 3 million shift hours each month, scheduling 20,000 clinicians at over 400 healthcare organizations. 

I talked to Dr. Vajracharya about the technology and how it works in practice. So here is our interview.

Suvas Vajracharya: Thanks for having me, Dr. Nichols.

Dr. Emma Nichols: So tell me, first of all, how does this scheduling software work?

Suvas Vajracharya: So Lighting Bolt uses artificial intelligence to help hospitals and healthcare organizations optimize their schedules for clinicians. And we use a technology called combinatorial optimization, and I’ll have more to say about that. But essentially, it allows to you to build a schedule that optimizes work-life balance for clinicians and also provides better access for patients through flexible scheduling.

Dr. Emma Nichols: OK. Great. So tell me when did this start? When did you develop the idea? And tell me about the history of the company a little bit.

Suvas Vajracharya: Sure. The company was founded in 2002. But even before then, when I was working at Los Alamos National Labs as a staff scientist, my job there was to optimize the scheduling of not people but computers. We employed supercomputers with thousands of processors meant to solve a complex problem like forecasting the weather. So that requires all these processors to work together to solve the problem in a reasonable amount of time. So that meant balancing the workload of the processors so that they complete as soon as possible. And it turned out that some of the requirements for that problem are similar to what’s required for scheduling doctors where you’re looking to see that no one doctor is overworked. You’re balancing the workload.

And so at the time that I was working on that project, one of my friends from high school who became a doctor later was asking me to help schedule a group of family practice internal medicine doctors that he was a part of. And the problem was very challenging. And it wasn’t intended to be a business venture at that point. I was looking to help Leo build a schedule for his group.

And many years later I thought if it’s helpful for Leo, it could be potentially useful for other clients. And sure enough we had customers from day one, and we’ve been bootstrapped since then. And now we have over 400 customers across the country using the system, benefiting from the system, to retain doctors, reduce turnover, and reduce burnout as well as improve patient access through better utilization of clinicians.

Dr. Emma Nichols: OK. That’s great. And it obviously is really applicable to the healthcare situation because many doctors are burned out. But have you extrapolated this to other industries? Or you do think about doing that?

Suvas Vajracharya: Well, we’re focused in healthcare. I think there’s a lot of value in understanding the domain of a particular industry. Although the technology that we use, it certainly has been employed in other industries, such as airline industries and wherever you need to schedule staff. The general requirements to match demand and supply are pretty common. And so we’re focused in healthcare because, as you mentioned, a physician shortage. And this is where, in some ways, the need is greatest and the doctor resources are very expensive. So it makes sense to optimize those resources as much as possible.

Dr. Emma Nichols: Absolutely. So can you tell me about some of the benefits? Like if a practice goes out and implements Lightning Bolt, what kinds of advantages does the product provide?

Suvas Vajracharya: So one benefit is that patient volume fluctuates throughout the year, and it may depend on flu season, for example, or during winter as soon as there might be more patients than we can expect during summertime. So if you can forecast what the patient volume is across the year, we have a forecast of the demand. But that’s not the only part of the story. We also need to make sure that we schedule enough clinicians to ensure that there are enough doctors to see the patients that are forecasted to come in on any given day.

So if there’s no match between demand and supply, you have either one of two situations. One is the doctors are not having enough to do and twiddling their thumbs, which is a very rare situation these days. But the other situation is that they’re overworked. They’re trying to see too many patients within the hour, overwhelmed with the amount of work to do. And that needs to be a priority when considering the schedule to make sure there’s enough capacity.

You also need to consider many other variables, such as when doctors are on vacation. They may have certain requirements about how many hours. Increasingly the workforce of clinicians requires a very flexible schedule where they can only work a certain number of hours or can only take certain types of shifts or only certain types of days. And so the goal then is to look at the collective availability of doctors and see how you can produce a schedule that meets their work-life balance even as you try to match the demand and supply. And it is possible to achieve both.

And so the reason we need to use AI or combinatorial optimization is that it turns out that this problem is very difficult to solve for people to do by hand, and even for computers, it can take a while to come up with a schedule. And so artificial intelligence has different meanings to different applications. But in our case, what we’re trying to do is to look at all the possible schedules and choose one that’s best. And so you can think of it as in trying to play a chess game. When you play a chess game, the intelligence is to make the right choice amongst a sea of options and intelligence. When we play chess, we don’t think about every possible choice. We narrow down the choices to what wins the game.

And it’s similar in scheduling because when you’re creating a schedule, there’s actually a number of schedules possible and say if it’s a 20-person group, it can be as many as the number of atoms in the universe. So you can’t expect a person or a computer to look at every possible option. So you need to be very intelligent. And people are very good at this. You know, they don’t look at all the possible choices. They look at choices that lead to a good solution. And so artificial intelligence, in this case, is like people making choices, not examining every possible option because if you did that, even a computer wouldn’t be able to finish within our lifetime to produce a reasonable schedule.

Dr. Emma Nichols: Right. Right. That’s so interesting. So when a healthcare practice or whatever facility implements this software, is there a lot of input required on the part of the user?

Suvas Vajracharya: There are a number of inputs, yes. So that includes the provider preferences on when they’re available to work as we talked about earlier. Their preferences on when they want to be on vacation. And since there are typically 20, 30 doctors in a group’s schedule, the number of preferences can add up. In addition, there’s staffing requirements, how many bodies you need at various specialty groups. And many specialty groups have a lot of subspecialties. Cardiologists and radiologists have subspecialized doctors that only see certain types of films or only do certain types of office procedures, so it’s really important to have the right staff to see their expected patients that are coming in.

Dr. Emma Nichols: Sure.

Suvas Vajracharya: There are many inputs into the system, and that’s what makes it really challenging to do this in your head.

Dr. Emma Nichols: Yeah, absolutely. Absolutely. So do you have any other comments for our clinician audience or anybody that might be interested in using this solution?

Suvas Vajracharya: Yes. I would say that it’s important for us to focus on today’s problem. Artificial intelligence, you know, when I was attending HIMSS earlier this year, a lot of it was discussing solutions about diagnosis and predicting, which is a really amazing what that promises to do. But a solution in large part in my view, in the future, it may not happen for a decade or so. And so we’re more focused on the current issue of the problems that we have now. Right now, doctors are overworked, they are burned out. They’re leaving the practice and encouraging their children not to go into medicine. So this is a problem that we need to address now, and so our goal is to use AI to solve today’s problem.

The interview originally aired June 13, 2018 on AIMedicineNews.com. The transcript has been lightly edited for brevity and clarity.

2018-08-23T13:22:50+00:00 July 10th, 2018|0 Comments

Leave A Comment