Tim Vasil is the Co-Founder & Chief Technology Officer at Hospital IQ, an operations management platform that uses data to deliver turnkey machine learning-based AI solutions for swift, sustainable operational improvements.
What initially attracted you to computer science?
Baby books. As an undergrad unsure of what career to pursue, I explored a part-time web developer position at BabyZone.com. The experience was incredible! One of my first projects was to take a physical medium, baby books, and bring it into the digital age, complete with sounds, transition effects, and an interactive guest book. I wrote some code, clicked a button to upload it to the web site, and suddenly thousands of parents had a way to connect with distant friends and relatives.
That e-baby book app revealed computer science as a field where I could be a bit of an artist, engineer, and maybe even magician, and use those skills to improve many people’s lives. I saw I could write code once and have a lasting impact everywhere. Wow!
Could you tell us the genesis story behind Hospital IQ?
The co-founder Rich Krueger and I teamed up to explore areas neglected by technology. You’d think hospitals would not be one of those areas, given the billions of dollars they spend each year on medical equipment, electronic medical records, and the like. Those areas are certainly well covered. What we saw, though, was a whole other side of hospitals, the operational side. This side is about figuring out when to schedule surgeries, where to transfer patients, which tests to prioritize, how to best plan for tomorrow, and so on. These are especially challenging problems and traditional medical software just doesn’t touch it.
To explore the opportunity, we met with hospital leaders and frontline staff. We saw heroics every day. We saw nursing schedulers making non-stop calls and fielding questions to dispatch staff to the most needed locations. We saw OR managers with sticky notes and whiteboards trying their best to divide up operating room time among surgeons. We saw operational excellence leaders with massive spreadsheets attempting simulations to figure out how many hospital beds to reallocate. In short, we saw so much manual effort being applied to problems because the software tools fell short, and we wanted to help.
Like many startups, our product development journey was not a straight path. Our early “help” came in the form of strategic tools that we thought would solve the toughest problems, but they required a lot of data and a lot of math. The mechanics of it seem impressive: we could assemble models automatically to simulate the inner working of a hospital and make recommendations as to whether that surgical schedule should be changed, or whether that new wing should be built. Yet while the questions they answered were big, they were also infrequently asked.
The true genesis of Hospital IQ as it exists today isn’t some inspired path by Rich or me, but by our entire team working hands on with our customers and realizing that our most important role is not in helping hospitals answer big, infrequent questions, rather it’s the seemingly small, frequent ones. Those are the questions that determine what everyone’s experience is like, both the patient going into surgery and the care team guiding them through that journey.
Could you discuss how the software enables healthcare systems to achieve and sustain peak operational performance?
Our software is all about meeting healthcare workers where they are today, in their daily workflows. Rather than expecting them to do something radically different, such as running a simulation, or interpreting a forecast, or optimizing staff schedules from scratch, we embrace the familiar steps they take every day in two ways. We digitize them, so they can communicate more efficiently with each other, and then we layer on predictions and recommendations. This lets healthcare staff work more efficiently, and more effectively. Best of all, it frees them up to spend more time on patient care.
Let’s take one example: staff scheduling. Figuring out how many nurses need to be in each unit for each shift is a challenge. Some staff call in sick. An unexpected surge in demand could arise. Nurses who can “float” across units need to be allocated fairly. Everyone’s skillsets, qualifications, and preferences also need to be taken into account. Put it all together and you can see why the phone in a typical staffing office rings off the hook. Yet in the same day as a go-live with Hospital IQ, the phones go silent. Much of the work is the same, yet with all the information centralized in the Hospital IQ platform, all of the considerations have moved from spreadsheets, whiteboards, and post-it notes, to sleek communications tools, automated analysis of variance, and suggestions for staff balancing. Nurse schedulers can perform their jobs more efficiently, and enjoyably, than ever before. Sustaining this performance is easy, too, because the tools are built to support the existing workflow. We’re not a consulting company that comes in to change the way work happens, only to see it drift back to status quo.
What are some of the different machine learning technologies that are used?
Our data science team employs whatever methods we need to get great results for our customers’ use cases. We’ve used statistical analyses to understand OR usage, ARIMA models to forecast surgical volume, Prophet to forecast census, random forests to classify inpatient status, neural networks for readmission scoring, and much more. Our data science team keeps abreast on the latest research, data sources and tools with ongoing “journal club” meetings, and regularly innovates on their own too. With such a wide-open field, there are so many compelling use cases and interesting data sets to explore and weave into the Hospital IQ platform.
One of the special challenges for us is to handle the uniqueness we see with each of our hospital customers. They serve different demographics. They have different specializations. The clinical and operational data at each hospital comes from different software configured in different ways with its own shortcomings. If we were to build a comprehensive model across all our customers, or even across all campuses within a single health system, it wouldn’t fit very well. Yet building manually customized and one-off solutions is not a scalable or robust approach. Instead, we rely on understanding the distinct characteristics of each customer’s data, developing generalizable models, and have built tooling to automate model building, ongoing training, and accuracy measurement and monitoring for individual campuses.
The free, publicly accessible tool COVID-19 Regional Forecast Dashboard has more than 76,000 users from hundreds of hospitals. What precisely is this tool?
When we first built the COVID-19 Regional Forecast Dashboard in March 2020, we were worried the U.S. might run out of available hospital beds, and wanted to provide an early warning system, not just to our customers but to all hospitals. To make that happen, we searched for data sets like staffed bed capacity by county, likely transmission and fatality rates of COVID-19 by age group, and dozens of other things. We even built a SEIR model to predict the trajectory of the virus on a county-by-county basis, and tried to provide as much context as possible, including the moment when ICU and med/surg capacity would be breached, how many people would recover, and even how many would die. Our aim was to assemble a complete county-by-county perspective from various reliable data sources.
Hospitals have used our dashboard as a tool to make key decisions, like when to open surge units or when to throttle back elective surgeries to make room for upcoming waves of infected patients. Interestingly, even individuals at home have found some use and even comfort from the tool, as it added a bit of clarity to a very scary and novel global pandemic.
In providing a public tool, we know we have an important duty to collect and analyze data faithfully and frequently, and to select the best data sources available. Sometimes that means swapping in better models as they become available. In the case of our own SEIR model, we eventually brought in, with permission, the Institute for Health Metrics and Evaluation (IHME) state-level model as it becomes a recognized standard by the White House and other sources. We found a way to put those predictions in the context of specific counties, as well as specific hospitals within these counties, to give hospitals hour-by-hour guidance on COVID-19’s continuing impacts.
Hospital IQ data scientists and engineers often participate in hackathons, what are some of the interesting ideas or projects that have come from these?
Each month, we encourage members of our data science and engineering teams to take a day off in order to foster their professional development and ignite ideas for innovation, whether it be attending an industry conference, taking an online course to learn a new skill, or any other activity that bolsters them professionally.
As part of this, several engineers and data scientists choose to spend their professional development day participating in Hospital IQ’s hackathons. Hackathons require participants to be scrappy, innovative, and in a single day, push a hard-to-transform idea into working software. In the days leading up to our most recent hackathon in October 2020, participants formed three teams and crowdsourced ideas from the entire company. No topic was considered off-limits; ideas that weren’t relevant to the company’s platform, or even the healthcare space, were perfectly acceptable. As it turns out, though, all three teams ended up choosing ideas that are now being implemented in the real-world.
The first team – Team Cara – focused on hospital readmissions and set out to build a solution that could predict which patients are at risk of readmission before they’re even discharged from the hospital. Hospital readmissions cost the healthcare system billions of dollars each year, so a predictive and proactive solution, like this one, would arm discharge nurses and care managers with the additional insight needed to reduce the risk, cut costs and know what each patient needs to stay out of the hospital. Team Cara took data from Hospital IQ’s operations management platform, and, using a patient-specific machine learning framework previously developed by the data science team, built a predictive model. For each patient in the hospital, the model assigns a score indicating the likelihood of readmission. Initial results from the model showed a high degree of accuracy.
The second team – Team Burt Reynolds – set out to build a regional surveillance solution that visualized layers of data on a map. The team wanted to integrate maps into Hospital IQ’s existing platform pivot table infrastructure, offering a way to plot a metric of interest arranged by latitude and longitude coordinates using the leaflet.js library. For their proof of concept, they used hospital transfer center data to highlight which affiliates were admit sources and at what volumes. The results showed transfer cases in a whole new light and clarified which geographies most patients were drawn from, as well as opportunities for growth.
The third team – Team Raptor Strikeforce – looked to develop a solution showcasing the return on investment (ROI) Hospital IQ’s operations management platform provides. The team built an interface to customize various inputs into financial models, such as average margin per elective procedure, and used the inputs to track changes in the financial health of a hospital over time. These visualizations tell a compelling story of how significantly operational effectiveness initiatives, and the investment in the Hospital IQ platform that enables them, pay off.
The three solutions developed for the hackathon showed that they could provide greater value to our customers. As a result, Hospital IQ has incorporated all three solutions within the existing platform, and they are being utilized by hospitals today.
Is there anything else that you would like to share about Hospital IQ?
Hospital IQ’s big, bold, audacious goal is to improve the efficiency and happiness of every healthcare worker every day. We’re proud of the impact we’ve had on healthcare so far, but our journey is just beginning. For any compassionate, mission-driven data scientist or engineer out there who’s interested in tackling one of the world’s toughest challenges – improving healthcare efficiency – we’d love to have you join us!
Thank you for the great interview, readers who wish to learn more should visit Hospital IQ.
- Attention-Based Deep Learning Networks Could Improve Sonar Systems
- Cerebras CS-1 System Integrated Into Lassen Supercomputer
- Deepfaked Voice Enabled $35 Million Bank Heist in 2020
- Facebook: ‘Nanotargeting’ Users Based Solely on Their Perceived Interests
- IBM Announces AI-Driven Software for Environmental Intelligence