Health AI Event Recap: It’s All About the Data

Health AI Event Recap: It’s All About the Data

May 13, 2024
Torso of a doctor in a white coat with abstract data viz overlay

Is it time to pump the brakes—or hit the gas—on health artificial intelligence (AI)?

This question was a focal point of the discussion at a virtual event Mathematica convened last week, featuring Mathematica’s Ngan MacDonald and Noland Joiner and Luminos Law’s Ellie Graeden. The speakers explored how the health industry can use new technologies to drive better outcomes, along with legal and regulatory considerations that health organizations should keep in mind as they innovate.

MacDonald, Joiner, and Graeden agreed that integrating health and AI is a team sport. Because while it’s too late to pump the brakes on health AI, hitting the gas responsibly means bringing together clinicians, technicians, data scientists, and other health industry professionals to effectively steward data and build more comprehensive models that improve patient outcomes.  

Here are some of the highlights from the event: 

On leveraging AI to advance health care: The panel noted that AI is most effective when it augments human expertise. Health organizations should continue to focus on playing to the strengths of machines and humans as they build out AI models. 

On patient safety and AI: Health industry professionals need to ensure that the data used to train models is applicable to their goals and objectives. This is difficult because data access and health care access vary drastically across geographies and populations. Ensuring data privacy starts with good data governance and understanding data provenance—the origins or history of the data coursing through the health care system. 

On data governance: If data inputted into an AI model is inaccurate, incomplete, or biased, the model’s results will be too. Health organizations need to recognize that bias is present in every data set and ensure that it is measured throughout the model’s life cycle. Using available open-source tools is one way to determine a model's effectiveness.

On data-enabled workflow: Health organizations should integrate various data tools across the health system, thereby improving patient experience. For example, they can use utility and emergency notifications to evaluate hospital capacity during crises. To ensure medication adherence, providers might check pharmacy records for what has been ordered, picked up, and refilled. 

View a recording of the full conversation below and learn more about Mathematica’s Health Data Innovation Lab