March 12, 2020: We all accept that if you drive downtown during rush hour there will be high traffic and it can take you longer to get there. Similarly, health systems have just come to accept hinders and clogs which slow down the rate at which patients can be assisted. Our guest today is CEO of LeanTaaS, Mohan Giridharadas, and he joins us on the show to have a great chat about how machine learning can help alleviate this problem. LeanTaaS helps patients get improved access to health systems while simultaneously allowing those systems to reduce the cost of healthcare delivery. It does this through a fleet of AI-driven products designed to assess statistics around patient behavior and hospital support capacity to schedule appointments based on a supremely superior mode than the simple person-to-person calendar check we are all used to. Essentially, LeanTaaS make tools for optimizing the utilization of expensive and constrained hospital resources. In our conversation today, Mohan weighs in on how the constraint-based algorithms in his products work, the amazing sales package they offer which completely removes any risk on the client side, and the specifics of how each of their iQueue products assist both providers and patients. Make sure you catch this interview because Mohan is doing some revolutionary things with AI that should help healthcare centers run to previously unfathomed levels of efficiency.
Key Points From This Episode:
Optimizing Constrained Resources with LeanTaaS CEO Mohan Giridharadas
Episode 195: Transcript – March 12, 2020
This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.
[0:00:04.8] BR: Welcome to this week’s health events where we amplify great thinking with interviews from the floor. You keep listening so we’ll keep producing HIMSS shows for this year. Special thanks to our channel sponsors, Starbridge Advisors, Health Lyrics, Galen Healthcare VMware, Protalent Advisors for choosing to invest in our show. My name is Bill Russell, healthcare CIO coach, creator of This Week in Health IT, a set of podcasts, videos and collaboration events dedicated to developing the next generation of health leaders.
It’s crazy but with all the news this week, we haven’t looked at the steady and even accelerated progress of things like machine learning, predictive analytics and AI among the great advancements that are going on. Today though, we’re going to do that, I’m joined by Mohan, the founder and CEO of LeanTaas and Mohan, I apologize for not saying your last name because I don’t want to do a disservice to you by doing that.
We’re going to address some of these really cool tools for optimizing the utilization of expensive and constrained hospital resources.
[0:01:00.1] BR: Good afternoon Mohan and welcome to the show.
[0:01:03.1] MG: Thank you Bill, it’s nice to be here.
[0:01:04.4] BR: Yeah, I appreciate you coming on. You know, one of the things I missed about not doing the show this year is walking through the booths and having conversations with people like you who are showcasing real solutions that are really going to change the way we approach age old ways of doing things.
Let’s just start there. I mean, give us an idea of what it would have been like if I had come up to your booth. What would you guys be talking about and what would you be showcasing?
[0:01:26.8] MG: Great. LeanTaas helps patients get improved access to the health systems by opening up access while simultaneously allowing healthcare systems to reduce the cost of healthcare delivery. We do this by sophisticated algorithms that help match the demand for care would be available capacity for providing that care.
It turns out this is a very complicated problem mathematically because on the demand side, it’s almost impossible to predict how many patients will need care or what type of treatment many days into the future, how long it will take, and all of the other variables around it so on the demand side, it’s highly unpredictable and highly volatile.
On the supply side, it’s highly constrained and highly interconnected. For example, to deliver an infusion treatment, you need the chair, the nurse, the pump and the drugs to all be available at the same time in the same place. Those are hard to predict, therefore, getting the supply demand curve to match is a very hard math problem and we’ve spent four years and tens of millions of dollars solving it and we now have a 200 person company dedicated to doing just that.
[0:02:43.4] BR: Wow, usually with these shows, I have a bunch of notes but we’re literally doing this as if I just walked up to your booth and so expensive, highly constrained spaces or equipment, what are we talking about, what are the areas that you guys are focusing on?
[0:02:59.7] MG: Three commercially available products. We call the entire product family iQueue and our first product was iQueue for infusion centers. We optimized the delivery of infusion treatments, largely for chemotherapy but not oncology infusions as well. We currently are contractor with 300 infusion centers, controlling about 7,000 chairs which is well over 10 or 15% of infusion capacity in the United States.
Our second product is iQueue for operating rooms. That product runs 1,200 ORs in 125 hospitals across 29 leading health systems in the United States. It’s a pretty successful product as well. We recently launched iQueue for clinics which aims at improving patient flow, initially in specialty clinics, eventually the primary clinics as well. On the drawing board, we’ve got iQueue for inpatient beds which we expect the launch by the end of this year, in fact, we are now in alpha trials with minimum viable products we launch even as we speak.
We’ve got other plans on the table for things like radiation oncology, imaging and lobotomy labs.
[0:04:02.2] BR: You know, this is a really pragmatic solution when I think about it. Off the record, I had a conversation with the CEO and I was saying to him, “If I could do one thing for you, what would it be?” And he goes, “Fill my OR for every minute of the day.” He just said, “That would help me immensely because it alleviates my cost problem, it generates a lot of revenue and we’re able to do a lot of the things that quite frankly create constraints.” That OR, those infusion centers, those clinics, I mean, keeping the throughput is so critical for these health systems.
Are you finding a pretty good response from health systems as you have conversations around these challenges?
[0:04:42.8] MG: Yes we do, it takes a while because I think the reality is people have just accepted that this is just the way life is. We all accept that if you drive downtown during rush hour, traffic is a lot and it can take you longer to get there.
Similarly, health systems have just come to accept the infusion centers are crowded in the middle of the day and patients will need to wait. The OR’s will be short and we need to run into overtime and have extra anesthesiology, teams and nursing teams late into the night.
People just accept that this is a reality because the understand it’s a very hard problem. What is eye opening to them is realizing how elegantly and in a sophisticated manner that this can in fact be solved. If you step back a minute and say, think about how every appointment in a health system is made? Two people chat.
It could be the provider and the patient, the provider and the scheduler or the scheduler and the patient. Two people chat, they look at the calendar and they say Bill, Wednesday, 10:00. Whether it’s for your procedure or for your infusion or for your doctor’s appointment.
They looked at the calendar and they made the appointment. There was no map involved in that decision, nobody ran the stochasticity of the demand signal or that constrained optimization availability of the supply elements into it. They just look at the empty calendar. This makes as much sense as imagine Bill, if you and I were solving the jigsaw puzzle and we’re sitting in an empty table, we pick up a puzzle piece and I say, “Hey Bill, where do you want to put this piece.”
You say, “That corner of the table looks empty, let’s go put it there.” We do that with each piece as we go through, right? What is the chance that the puzzle is solved when we are done with the pieces? Zero. Why? In this model, each puzzle piece is like a patient and the empty table that we started out with which becomes full as the day progresses is the calendar upon which we’re making appointments. Just as we would not be surprised that the puzzle doesn’t solve, how can we expect that the bolus of patients that we’ll pick on any given day actually makes sense in terms of who needs to come in what order or what service?
There was no optimization in there. When we challenge health systems with that, they get it and their first reaction is often, “Well, it’s hard to predict a demand and the supply is uncertain and people don’t show up on time and nurses call in sick and pumps are out of service and robots go down and so things happen.”
Our come back to that is, “Sure, let me give you an example where it works. Think about Uber, Uber has no idea how to predict the demand, right?” Because, in the last month alone, I’ve taken an Uber in New York, LA and in south Dakota and I don’t see any of those places. The demand is hard to predict, the supply is even harder, the drivers don’t even work for them. These are free-lance agents who wake up and decide to drive today or not.
What does Uber do? It builds sophisticated map models. It has continuously modeled demand for cars in every zip code, every minute of every day, everywhere in the world. It factors in whether time of day, political events, sporting events, et cetera. And therefore gets really accurate demand.
Then it looks at the driving pattern of all their drivers and figures out who drives when and where. It tries to match the demand and signal, supply signals. When they are off kilter, it proactively pings the drivers and tells them to come out for high incentive or to drive west six blocks and drive north six blocks just to start getting to where the action is.
If the demand is too high, it does search pricing. It does crazy mathematics with dynamic real time adjustment and machine learning to get this supply, demand signal to match in a world where the demand is volatile and the supply is unpredictable. Healthcare can do that. That’s kind of what we’ve done.
[0:08:27.6] BR: That’s interesting. Every client you go into, you’re going to have to build that model out so you come in to my health system and I say, “Yeah, let’s definitely do this for OR and infusion centers and we’ll get going down the road.” This is where machine learning comes in, you’re going to take a whole bunch of historical data and run it through your algorithms and essentially come back with models that work for my health system is that how we’re going to stand this up?
[0:08:50.6] MG: Yeah, exactly. The beauty of what we’ve done is it’s one code base. It’s a multi-tenant, SaaS product that runs in the Amazon cloud. Sure, we configure it for every health system, where this is not a customized solution built for each health system. Let’s take infusion as a starting point, that’s an easy one to experiment and switch to OR and explain how that works.
For infusion, we start out exactly as you said Bill. We take historical data. From the historical data, we build very sophisticated forecasts for the volume of patients by day of week. Infusion is a day that we give. Because different oncologists practice different days of the week and they tend to have a different propensity for standing their patients on to infusion, depending on the kinds of patients they have with the disease group that they cover. We predict the volume.
We then predict the mix, how many patients are going to need a one-hour treatment, a two-hour treatment on a Tuesday, on a Wednesday. Having predicted the volume and the mix, we grade the curacy of that health system in estimating the duration.
How do we know? We got the historical data. We got a thousand instances where it told us it was a three hour appointment and so we can construct the bell curve of how accurate they are around the three hour prediction. Based on that we adjust the expected duration of each appointment. Having done that, we then have constraint-based optimization algorithms that factor in hours of operation, chair availability, nurse staffing, nurse roster, nurse cooperation, nurse specialization, pharmacy hoods, pharmacy hours, pharmacy distances.
We factor all of that in and come back with the suggested template. No longer is it open up the calendar and first come, first serve, free for all, give them appointments, the dialogs switches slightly to, “Mr Russel I see you need an appointment on Wednesday, your treatment calls for a three hour infusion. On Wednesday’s I can offer you a three hour infusion, at 7:10, 7:40, 8:20, 8:50 or 9:30. Can we make any of those work?”
By steering very intelligently and very gently, it doesn’t damage the patient experience. Patients in every walk of life are used to this concept of, “Let me ask for the time I want and the facility will adjust and give me the closest they can to that and I’ll live with that,” this is true when you go get a haircut or make a dinner reservation. We take that optimization and put it back into the EHR.
[0:11:11.7] BR: Interesting. How do you factor it – there’s two factors I’m struggling with here. Equipment failure and staffing, right? Again, I guess you’re looking at historical data, you know that things fail at a certain rate and you know that people don’t show up at a certain rate and patients not showing up at a certain rate, you factor that into the algorithms but again, it’s complex, it’s not as simple as just saying, “Well, x percent of these people aren’t going to show up,” you really know.
[0:11:38.7] MG: It’s not that simplistic at all. What you have to do is we run discreet events simulation algorithms because we’ve got a historical data. We know what they were aiming for and what they got. Now, we don’t know why people are running late, whether the Starbucks line is back up or there is no parking in the hospital that way. But we just know their propensity to run late for every minute of every day.
So we build our templates to be resilient. Our templates have shocks built into them because we expect stuff will happen, patients will be late, nurses will call in sick, et cetera. We expect all of that to happen. Having built the resilience schedule, that’s like your best shot, that being said, stuff will still happen and when that happens, we can guide them. so for instance, if a nurse calls in sick, you can take one nurse out of the roster for the day and rebalance your algorithm.
We allow dynamic real time support. What’s interesting is, we don’t need to get dragged down into the nuances of each patient, John Doe is got a comorbidity and so when it’s a three-our chemo, it will probably take five hours for John, we don’t need to do stuff like that because we’ve built our schedules to be resilient.
The easiest way to explain it is the following. If you go on to Google maps right now and say, you need to drive from New Jersey to Manhattan on Monday April 15th or whatever that spectators, way into the future at eight AM, Google maps will predict how long that ride will take you.
Now, it has no idea who is going to be driving on that day, it’s not trying to adjust the algorithm and saying Bob’s driving, Bob’s a slow driver and hogs the middle lane, I better add 10 minutes to the commute. It’s not doing any of that. What is done is it’s done probability event distributions for every .1 mile of the journey, it gets some right, it gets some long, it gets some short but because the mathematics is robust, it’s able to give you a remarkably accurate forecast.
Even though that’s 20 days out from now. Does that make sense?
[0:13:35.4] BR: Yeah, that makes sense. I want to ask you about, let’s assume I implement this within my health system, what kind of returns are we looking at? What are health systems finding in terms of those constrained resources?
[0:13:48.6] MG: It varies by product. For infusion, we are finding the returns are on order of $20,000 per chair per year. We build our products to have anywhere from a five to 15 times return on investments. We want to make sure that if it complete no brainer from an economic standpoint for the health system.
In fact, we are so confident, we do two things that just throws health CIOs by surprise, catches them by surprise. The first is we guarantee the product. If 90 days after the go live you’re not feeling it, we’ll return all the money you paid up until that point, they have never seen that from any IT vendor that they’ve ever dealt with.
The second thing we do is we tell our customers that we do not confuse customers as hostages, you do not need to sign a multi-year contract and be locked into something that’s not working for you. All our contracts are written with a cancelling option. Even after the refund guaranteed period is gone, if you feel that the impact is you know, flattening out and you could do just as well without it, cancel it any time and our churn is as close to zero as it’s possible to get.
Which means that every single day, every one of our customers had the right to cancel their contract and choose not to at the after day that is the reality of it. Some of our customers have been on our product for four years and five years and so they had 1,500 opportunities that 300 days a year times five years to cancel and they have and so it’s very clear that it is delivering in almost value to them.
[0:15:18.8] BR: Your solution by the way is – by the way that’s sales. The way you are going about selling to health systems that’s as unique as I have ever heard and really fascinating. I mean you have taken all of the risk out of me moving down this path or introducing you to my clinicians to see if this is a great solution for our system. I mean what’s the risk? The risk is the time I guess spent on doing the pilot if nothing comes of that.
But I assume you also have a fairly good list of reference-able clients that are seeing these kinds of returns. So there is almost no risk in the sales process.
[0:15:54.6] MG: It’s no risk at all because we understood that this is a bit of a disruptive technology. It is going to surprise health systems therefore we needed to put into place all of these things. In terms of the time invested on their side, we have minimized that as well. We have built the scripts to drive the data extraction. We don’t need integration with the EHR. We are HIPAA compliant and SOC compliant. We mostly don’t need DHI and so we have eliminated the security risk, the implementation risk, the implementation effort, the financial risk so we are de-risking it as well as we can.
On the reference ability side, we’ve got the who’s who of institutions. So if you start down the academic medical centers and you rattle up the names of the academic medical centers you respect most starting with the east, Columbia, Cornell, John Hopkins, UPMC, Penn, Duke, Emory come down in the middle of the country, North Western, Rush, MD Anderson.
On the West Coast, Paul Rudder, Utah, Seattle Cancer Care, UCSF, Stanford, UST, UCSD are all customers. Many non-academics as well. Dignity prior with the merger with CommonSpirit deployed our product across 39 hospitals and 255 OR’s. Many, many systems have deployed the OR product across the entire asset based, Duke across all 110 OR’s. Oregon has been signed across all hundred plus OR’s similarly with UC Health.
And so we are getting system wide deployments of each of our products and they usually start with a narrow field. So slow and catering on infusion stock as to the one set that we now run all of the centers. Similarly with MD Anderson, Duke started us big bang across all of their OR’s and so we are open to whichever way we structure it.
[0:17:42.5] BR: So this is how these conversations usually go with me and the room is, it starts with, “Hey, this is what your product does” I get it, the sales model makes sense and all of that but then I come back and say, “Does your product help me do some other things?” So it is interesting, you’re really looking and optimizing that calendar but can you come back and help me to optimize my maintenance schedules around specific equipment or scheduling of resources for those infusion centers and OR’s. Can we do it in reverse and sort of start to design better?
[0:18:14.1] MG: We could do some of those and those are natural progressions of what we’re talking about because right now, we are optimizing the asset by matching the demand and supply signals. Improving the supply by making greater availability of the pump or the robots etcetera is a natural extension but the moment and the reason we are doing that is that starts to limit scalability because now suddenly you have to deploy a team at each institution to work your way through what are exactly the equipment supply chain issues or the exact maintenance regiments in place etcetera, etcetera.
We are building up product to be scalable SaaS products. So for instance, our infusion product is now already live in a 180 infusion centers. In 120 of them we have never set foot on the property even once, right? We have gone live in a 120 centers without setting foot on the property.
So we have sold the product over a Web X. We have done the set up over a Web X, CDs or WebXs over a six to eight week period and we have been live and supporting them every single day for many years now and for example, we got a client in Montana and I have never in my life set foot in Montana let alone on this particular location and so this is a goal for us to make it scalable and therefore we are very judicious about where exactly we extend the process improvement mindset because that immediately goes down to a facility by facility deployment, which limits scalability in a pretty dramatic way.
[0:19:41.6] BR: Yeah and you have just addressed the last challenge, which is when you talk about somebody like Dignity or some of these other systems. I mean you’re talking about national deployments of this. So because you have chosen to focus in on that one area, which is an acute problem within health systems and you’ve decided to just focus in on that problem, you have created a very scalable solution across the board.
You know to be honest with you, I mean if I am walking through the booths I see all of these things I am seeing referencable accounts, I am seeing a solution that is really going to help me from an operational and a cost standpoint, I see very little risk in implementation. There is part of me that’s a pessimist and I am like there has to be something here, something wrong and I am not seeing it. So probably what I would do is just pick up the phone, call some of those academic medical centers where I know the CIO’s and just have a conversation with them I think.
[0:20:34.1] MG: Please do and for us – I mean we have gone very far right? Because the OR product was launched in 2017. So at the start of 2017, it is working in one hospital at 35 OR’s. It is now in 1,200 ORs across a 120 hospitals. So we are scaling as fast as we possibly can and it is a matter of getting in front of the executive teams of large systems. It is about building a nationwide scalable sales post that can scale and position and that is kind of what we are in the process of doing.
We haven’t talked about the OR product. Let me give you the two minute version of what the OR product does. So what has historically happened is every health system and I am sure you are very familiar from your CIO days has block schedules in place and the reason they have block schedules is surgeons and service lines need a guarantee. I am a surgeon I get Monday and Thursday block. So I know I’ve got two full day blocks of an OR available to me.
And so all the patients I see when they need surgery I steer them into a Monday or a Thursday into the future, into one of the blocks I’ve got, okay? So now let us just think of that from a supply-demand perspective. What we have just done is create static pre-allocated reserve supply. Now think about the demand side. The demand side is crazy volatile. In order for me to fill a particular Monday eight Monday’s out, it will depend on all of the patients I see in my clinic.
What percent of them will need surgery, what percent of them need surgery that fits a following Monday or Thursday timeframe and what the lengths of their procedures will be and whether all of them will add up to be exactly the eight hours I’ve got. There is no chance of all of that falling in place and so what happens a lot is blocks are reserved, surgeons are still looking for time and they scramble and at the last minute they do add on cases and emerging cases.
Which are not really add on and not really emerging and so it throws the whole resourcing off the health system into a bit of chaos to find a match supply and demand of an OR. OR capacity is precious. It is $300 a minute to have an OR and so to have it allocated in this rough map sort of static supply doesn’t make a lot of sense. So all of the efforts in improving utilization rely on trying to cut down these minutes of we call it grains of sand. First case on time starts, turnover time etcetera.
Those are saving five minutes here and 10 minutes there and 20 minutes in another place, that’s great, but you can’t squeeze in a case in those. So instead of starting 20 minutes later if I started on time, nothing really changes, all that would happen is I’d end 20 minutes earlier than I otherwise might have. You just shifted the entire day forward and back. You haven’t effectively raised capacity. What we have come up with is a patented concept called collectible time.
We are able to mine the patterns of surgeons and figure out who leaves large blocks of time either in the mornings or the afternoon. Who releases blocks systematically ahead and above and beyond what you should normally recommend, who cancels and abandons blocks, from that we can reallocate blocks more efficiently to surgeons of service lines. Even when that’s done, in the moment the match won’t happen and so what we have done is we have created a concept like OpenTable for OpenTime.
Which matches the need for you to get a dinner table with the restaurant’s availability and ability to offer you a table before. So we have created exchange, which allows surgeons and service lines with two clicks on their mobile phone to request a block or get a block. You know we see Colorado Health when we started with them said, “We are 95% blocked out. We have no room for new surgeons.”
By deploying exchange, suddenly you created a liquid market place where a new surgeon who have been bought out to the faculty could ask to put up a hand and ask for a certain time in the OR and somehow they could arrive. They have recruited 11 new surgeons who built viable practices without really having guaranteed block time. So the power of things like that is enormous. The product also helps people get the analysis on their phone, they can drill down into how they perform, what the causes of delay were, etcetera, whereas historically they just got a bunch of PDFs once a month and nobody looks at.
[0:24:46.6] BR: Yeah that’s amazing. So it’s iQueue, for those of you listening because infusion centers, operating rooms and clinics, what’s next? I mean what are people pushing you or asking you for because I would imagine there’s other areas that health systems would like for you to do.
[0:25:04.3] MG: Health systems are really concerned about beds, inpatient beds. Virtually every urban hospital is running flat out: 100% bed utilization and for periods of the day 110% bed utilization, which means 10% of the people don’t really have beds, so they’re in the hallway, they are backed up in the back queue or put somewhere in the system. If you compare what happens to an inpatient bed, this is what happens in a hotel.
So think about a hotel, guest check out in the morning, there is time to clean the room and then guest check in the afternoon. It happens the other way around in a hospital. Patients try and check into a bed in the morning because surgical procedures happen in the morning and by the time someone gets discharged from the hospital with rounds and imaging and CT scans and blood work, it becomes late afternoon.
And so you got this inversion period where the incoming is more than the outgoing, late morning, early afternoon and that causes a chronic back-up every day. So hospitals have built sophisticated mechanisms to manage this patient placement people who are trying to play the chess game of which patient gets which bed and which unit at which time. There are people who can stand up surge units when you suddenly need 20 more patient beds and they can open up a temporary unit of staff for the right kind of nurses. It is a really complicated chess game to play.
We are using our algorithm to predict the likelihood of a discharge on time by each unit for each hospital in a fingerprinted, customized sort of a way and based on the prediction of discharge, be able to make more intelligent decisions on bed placement to get more patients into the right unit with the right service at the right time with the minimal amount of wait, right? So it requires playing the chess game two and three moves ahead.
[0:26:48.7] BR: So I’d be remiss and we are overtime so I appreciate you answering my last questions here but I’d be remissed if as a CIO I wasn’t asking you. So you said earlier you are grabbing data from the EHR, you are moving into the cloud, you are running your algorithms out there in the cloud and then providing that information I assume potentially back to an interface of some kind or are you actually delivering data back into the EHR itself?
[0:27:13.5] MG: We are not writing to the EHR. So what we do is we provide information that helps the health system make more intelligent decisions. So it varies by product. So for the infusion product we are providing a template and we will teach them how to modify their Epic cadence template or sonar templates from their current methods to an intelligent thing. They’ve got a block schedule in their EHR already. We replace it with our more intelligent block schedule.
So there is a little bit of an assistant at that point in time just so that from a workflow standpoint, the front line continues to work entirely within their existing systems rather than seeking to replace anything.
[0:27:54.8] BR: So from the time I make a decision to the time I’m up and running, taking contract out of it because some health systems take 60 days for a contract but let’s assume the contract is done, is that a 30, 60, 90 day window? What is that?
[0:28:07.9] MG: Contracts signed to go live is typically eight to 10 weeks, give it plus or minus a week or two here or there and that involves getting the data clean, getting it loaded, getting the models configured, run metric alignment, schedule our training, etcetera, etcetera. We’ve raised a whole lot of money over the last several years and therefore acquired well-funded to grow and expand and that’s kind of the mode we’re in.
[0:28:31.1] BR: That is exciting. Well thanks for taking the time Mohan, I appreciate it. Where can people go if they want to get more information on your solution, the company or even investing?
[0:28:40.6] MG: leantaas.com, all of our products are laid out. Each product has got three or four minute video attached to it so you can get a sense of how the product works. It lists the names of several of our referenced clients. So that would be the place to start.
[0:28:57.1] BR: Fantastic, all right let me close out the show and then I’ll end the recording and then we can have probably follow on conversation. So don’t forget to check back multiple times this week. We are going to be dropping multiple shows from HIMSS and I appreciate everybody who’s sent me notes and I’ve really appreciated the interviews and the coverage. So that is greatly appreciated.
The show is a production of This Week in Health IT. For more great content, you can check out the website at thisweekhealth.com or the YouTube channel. Thanks for listening. That is all for now.