Association member UVA Health System, in Charlottesville, Va., is one of seven finalists selected to advance to the next stage of the Centers for Medicare & Medicaid Services (CMS) Artificial Intelligence (AI) Health Outcomes Challenge. Launched in March 2019, the challenge encourages applicants to use AI tools to predict unplanned 30-day admissions and adverse events at hospitals and skilled nursing facilities.
In stage 1, 25 applicants received a data set of Medicare administrative claims data, including for Medicare Part A (hospital) and Medicare Part B (professional services), to forecast several outcomes, including unplanned admissions related to medical conditions, such as heart failure, pneumonia, and chronic obstructive pulmonary disease; and adverse events, such as hospital-acquired infections, sepsis, and respiratory failure.
A second objective of the challenge asked participants to visually demonstrate and explain AI predictions to front-line clinicians and patients.
“It’s putting all these things together — like the advance deep-learning explaining factors that led to the prediction and the descriptive model — and being able to actually show that to the physicians in a clinical setting and provide them with enough information to really know why this prediction happened and why these particular patients may need this particular intervention,” says Valentina Baljak, PhD, data scientist at UVA Health. “Together it provides a solution that we call Actionable AI.”
At UVA Health, just 3 percent of patients account for 30 percent of readmissions within 30 days of discharge. The health system’s competition proposal, Actionable AI, builds on years of work within the health system to use Medicare claims data — including diagnosis codes, medication, prescriptions, and health care appointments — alongside electronic medical record data that captures social determinants of health, advanced care planning, and chronic condition management, to evaluate readmission risk. The resulting predictions serve as a decision support system for health care providers.
“The goal is to provide better information to them, potentially from data deep into someone’s medical history that might not otherwise be obvious to a physician,” says Sean Mullane, MS, senior data scientist at UVA Health. “It’s very much about empowering our physicians and caretakers with the best data that they can get.”
Armed with that data, physicians can be better equipped to respond to patient needs.
“That can be anything from providing easier access to medication that the patient needs to making sure that they actually have a support network,” Baljak says.
Baljak and Mullane hope that a CMS tool using a broad dataset from the large Medicare beneficiary population can aid health care facilities that might not have the resources to conduct their own data analysis.
“The scale of the data allows trends to be discerned that aren’t really available through the more tightly prescribed methods,” Mullane says. “There [are] things you can get out of that, that you can’t get necessarily with smaller patient populations, especially with rare conditions where the numbers may just be too small to be statistically feasible.”
As a finalist, UVA Health received $60,000 from CMS to refine the algorithm for the second stage of the competition, which was delayed due to the COVID-19 pandemic. In this stage, the team will focus on equity and bias reduction in AI.
Historic health care inequities can result in lack of data for certain groups, and social determinants of health influence patient care and patient outcomes. These trends are reflected in the data and train algorithms to respond accordingly, at times suggesting interventions that might unintentionally harm certain groups.
“The goal is to proactively understand what biases might exist in the data so that we can address those intentionally again, to avoid making predictions and recommendations that turned out to be harmful,” Mullane says.
CMS will announce the grand prize winner, who will receive up to $1 million in prize money, and a runner-up, who will receive up to $230,000 in prize money, in April 2021.