An innovative artificial intelligence project from a group of Mayo Clinic investigators was among the top 25 participants selected to advance to Stage 1 of the Centers for Medicare & Medicaid Services (CMS) Artificial Intelligence Health Outcomes Challenge.
The challenge is to develop technology that can be used to predict unplanned hospital and skilled nursing facility admissions, and adverse events within 30 days – using Medicare claims data.
“In addition, teams are expected to develop strategies to explain AI-derived predictions to front-line clinicians and patients, and identify ways to increase use of AI-enhanced data for quality improvement,” says Patrick Wilson, a statistician in the Mayo Clinic Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery.
Along with James Naessens, Sc.D., a researcher in the Mayo Clinic Robert D. and Patricia E. Kern Center for the Science of Health Care Delivery, focused on issues in quality and safety, Wilson co-leads Mayo Clinic’s project that made it through the first cut in the competition.
More than 300 teams submitted entries in response to the CMS challenge. Now 25 are competing for 7 spots in Stage 2 of the Challenge.
Mayo’s advancing project is “Building Interpretable Model Ensembles to Reduce Unnecessary Hospital and SNF Admissions – A Claims-based Learning Framework to Improve Patient Care through Policy.”
“More than 10 years ago, then President and CEO Dr. Denny Cortese and collaborators developed the Mayo Clinic value equation,” says Dr. Naessens. “Value equals quality over cost, with quality defined as outcomes, safety, and service, and cost considered over time.”
“This work is an outgrowth of that value equation – and a chance for us to share our expertise and potentially transfer some of our value to the nation’s health care system.”
Dr. Naessens and Wilson are leading a diverse team to design a claims-based learning framework for developing machine learning models. They are using what the team refers to as ‘ensemble models’ – a methodology that incorporates multiple machine learning algorithms to leverage the diversity of different concepts, expertise and backgrounds.
“These ‘ensemble models’ will predict unplanned hospital and skilled nursing facility admissions and readmissions, as well as other adverse events,” says Wilson. “We are developing the ensemble model to work with the current data delivery and management capabilities of CMS.”
To build a scalable and sustainable artificial intelligence framework with a high likelihood of integrating into a CMS health innovation model, Wilson says the following five design principles are critical. The model must:
Usability is ensured by integrating clinicians throughout the process.
“Our team is leveraging the collective knowledge and experience of several health services researchers and health care policy analysts, as well as potential physician users, and combining it with the power of artificial intelligence/machine learning,” says Dr. Naessens. “This multidisciplinary, team science approach is one of the critical elements of Mayo Clinic’s practice transforming research and value for our patients.”
He and Wilson say that Mayo Clinic’s algorithms will produce clear, actionable insights, and if selected as a winning proposal, CMS could incorporate them into an innovation solution for quality improvement across model participants.
“We view this as the main objective for this project,” says Wilson. “However, the claims-based learning framework we are developing could be leveraged for multiple purposes involving other prediction models, quality improvement initiatives, risk adjustment and testing for subgroups of program treatment response.”
Wish them luck and stay tuned – winners of Stage 1 will advance to Stage 2 in April 2020.
Tags: artificial intelligence, care transitions, Center for the Science of Health Care Delivery, Centers for Medicare & Medicaid Services, health care policy, health care value, health sciences research, James Naessens, medical innovation, News, readmissions