Dr. Ashwin Belle is an analytics architect on the Data Science team at the Michigan Center for Integrative Research in Critical Care (MCIRCC), one of the nation’s most innovative organizations in diagnosing, monitoring, and treating patients with critical illness or injury and a part of the University of Michigan’s Michigan Medicine Health System. MCIRCC is comprised of several teams, faculty members, and physicians who are focused on using interdisciplinary collaboration to discover new and better ways to diagnose, detect, and treat medical problems with technology.
“The reason I wake up every day and come to work is to be able to build and develop analytic solutions that can save lives,” said Ashwin.
The field of medicine is awash in data from doctor visits, diagnostic tests, and routine patient monitoring by machines and nurses alike, presenting a huge opportunity for data scientists like Ashwin. Even though machines are constantly monitoring patients, long-term patterns often go unrecognized due to the fleeting nature of doctors’ rounds and nurse visits and the sheer volume of data that is produced. The challenge lies in capturing and prioritizing which pieces of data are relevant.
MCIRCC is working to harness this data to supercharge their life-saving work by applying data science superpowers to critical care situations.
MCIRCC’s Data Science division is an integral part of their strategy to use big data to discover ways to better mitigate critical medical situations like internal bleeding. The first step was finding a way to capture all the relevant data and store it, which required the development of a special infrastructure. Once that monumental task was complete, Ashwin and his team had to figure out the best way to apply data science to use that data to inform clinical decision-making and improve outcomes for patients.
The data science team at MCIRCC were experiencing issues such as:
- — Overly complex models took too much time to develop and implement
- — Models and methods weren’t easily scalable to adjust to the vast amounts of data being produced by medical monitoring systems
- — Models didn’t provide a high enough degree of accuracy or allow for the use of real-time data
- — A communication gap existed between data scientists, engineers, and medical professionals
After seeing a DataRobot demo, Ashwin saw a possible solution and immediately activated the DataRobot signal.
Impressed by the speed and accuracy of DataRobot’s automated machine learning platform, not to mention the platform’s ability to deliver multiple options for different predictive models, Ashwin knew DataRobot would provide the tools MCIRCC needed to take their lifesaving applications of data science to the next level.
With DataRobot, MCIRCC and Ashwin can now:
Ashwin and the rest of the team are already using DataRobot to address a number of predictive challenges at MCIRCC. One project with life-saving potential is a model that predicts the possibility of Hemodynamic Instability in intensive care patients. The model evaluates thousands of data points produced by medical devices that monitor patients in real time, alerting hospital staff when the onset of hemodynamic compromise is detected well before traditional vital signs. If testing goes well and the model proves accurate in assisting physicians, the next step is to leverage commercial partnership to achieve FDA approval in order to scale the model to save lives across the country.
|Traits & Capabilities|
|Traits & Capabilities||Data Scientist Mortal||Data Scientist Superhero|
|Bottom Line Business Impact|
|Bottom Line Business Impact||$100K+ per year||$M's+ / Year|
|Model Engine Tested|
|Model Engine Tested||No||200,000,000+ times|
|Standardization||When there's time||Every time|
|Multitasking||Single model||Multiple models|
|Package installation and dependency management|
|Package installation and dependency management||Manual||Not needed|
|Risk of coding errors|
|Risk of coding errors||Substantial||None|
|Popularity with business users|
|Popularity with business users||Somewhat||Very|
|Business user engagement|
|Business user engagement||Too busy||Always available|
|Helped with guardrails|
|Helped with guardrails||None||Yes|
|Skills Needed||Significant - PhD-level||PhD not required|
|Model deployment effort|
|Model deployment effort||Significant||Effortless|
|Model iteration||Manual & slow||Automated|
|Helped by other Data Scientists|
|Helped by other Data Scientists||No||85+|
|Algorithms and libraries|
|Algorithms and libraries||A comfortable few||Many|
|Number of models / month|
|Number of models / month||Very Few||Thousands|
|Model Transparency||None||X-Ray Vision|
|Explainable Models||'Black Box'||Automatic|
To be continued…
In the future, not only will DataRobot help people like Ashwin develop predictive algorithms that save lives, but because DataRobot is built to allow non-data-scientists to understand results, and reasons for its predictions, the platform will help MCIRCC speed up the time it takes for its solutions to get through clinical trials.
“The easier it is for us to explain to the FDA how we did certain things and what we have done, the better. The deployment mechanisms that DataRobot has provided and the direction they are headed is really going to make it easier to get our solutions approved, and thereby save more lives in the process.”