I³ award brings Assistant Professor Paula Voorheis and an interdisciplinary team together to improve AI-based disease screening with user-centered design and AI-enabled evaluation
A clinically sophisticated, AI-powered tool that could, for example, detect early signs of preeclampsia in a patient’s medical record can’t meaningfully impact care if that information isn’t conveyed to the right clinician at the right time — and in the right way.
Unfortunately, at present, most digital health innovations are either beautiful but lacking in substance, or scientifically sound but hard to use, says Paula Voorheis, assistant professor of pharmacy at the University of Wisconsin–Madison School of Pharmacy.
“They tend to fall at one of two extremes: They either look and feel like Instagram or TikTok but aren’t designed for the realities of healthcare, or they’re clinically rigorous but so clunky that no one wants to use them,” she says.
Voorheis is helping an interdisciplinary team of UW–Madison researchers close the gap between technically sophisticated AI and tools clinicians can use to improve care.
“Technology can throw out every answer you could imagine. We need to continually evaluate the use of AI so that it is human-centered, not technology-centered.”
–Paula Voorheis
With support from a newly awarded Igniting Interdisciplinary Innovation (I³) grant, the team — including Irene Ong, associate professor of obstetrics and gynecology and biostatistics and medical informatics; Maja Waldron, assistant professor of statistics; Daniel Cho, assistant professor of surgery; and Fred Sala, assistant professor of computer sciences — is beginning work on closing the gap for reliable AI-based disease screening.
The project targets three barriers that slow AI tools’ integration into clinical practice: small, fragmented data sets that machine learning tools can’t reliably process; models that don’t fit clinical workflow or earn clinicians’ trust; and risk prediction that looks at a single-timepoint snapshot instead of how a patient’s data changes over time. Voorheis is focused on the second barrier, helping researchers build tools that will actually be used.
Collaborating to advance health
Cho, a pediatric plastic surgeon, and Waldron had already collaborated to build CranioSure, an app that assesses and evaluates infant head shape — a digital advancement designed to detect abnormalities more consistently than traditional tape measurements and visual assessments, both of which can be difficult to conduct on squirming babies.
“They approached me because they really wanted to do user research to understand how clinicians would use their app and whether families would have any concerns with the process,” Voorheis says.
Ong was also independently developing an AI-driven digital warning system to detect early signs of preeclampsia in patients’ electronic health records and alert clinicians to potential risk.
When UW–Madison invited applications for I³ funding to support cross-disciplinary approaches to complex societal challenges, the researchers decided to collectively utilize their complementary expertise in AI development, disease screening, clinical practice, and human-centered design. Their proposal earned them seed funding for one year.
The gap between AI innovation and clinical impact
At present, tens of thousands of health and wellness apps are added to the major app stores annually. Countless other digital health tools are marketed directly to clinicians and healthcare organizations. Most of those tools, Voorheis says, are built and launched without a solid understanding and consideration of users’ needs.
“If they were really working to create clear, consolidated solutions for users, there wouldn’t be a hundred thousand healthcare apps being added every year,” she says.
Voorheis, who leads the BUILD (Behavioral and User-Informed Learning for Digital Health Innovation) Lab at the School, will be working to build Nudgi, an evidence-informed, AI-enabled software platform designed to facilitate user research for digital health innovation.
Instead of digital health innovators juggling study planning, recruitment, user testing, data collection, analysis, and reporting across separate tools, Nudgi would bring these activities into one platform. Before and during implementation, innovators could use Nudgi to simulate user testing, conduct moderated or unmoderated interviews and usability walkthroughs, capture live user behavior and feedback, and use AI to synthesize findings and generate evidence-informed recommendations for improvement..
“Our lab is about advancing both the methods and practice of digital health innovation,” Voorheis says. “In this project, we’ll be creating innovative software that hopefully will allow more digital health teams to use evidence-based methods to gather user research at scale.”
Her co-investigators will be among Nudgi’s first users, embedding its user research into their own innovations to determine whether they are providing the right content via the right channels. For Cho’s project, for instance, Nudgi could help assess whether CranioSure fits pediatric workflows, whether clinicians can easily interpret its outputs, and where the design could be improved. That kind of evaluation is often missing in digital health innovation, and Voorheis hopes that her software could eventually be applied more broadly.
““With Nudgi, my aim is to help innovators understand how their digital health tools are actually being used in practice — who is using them, when and how often — and what users think,” she says. “Ultimately, I want to help innovators build tools that better meet people’s needs.”
Development will unfold in three phases: First, the BUILD Lab will interview digital health innovators, like Cho and Ong, to better understand their needs for using AI to support user research. Then, the team will conduct a usability test on how innovators are able to use Nudgi in simulated scenarios. And finally, the team will pilot the tool next summer.
As AI becomes increasingly embedded in clinical practice, pharmacists will be among the healthcare professionals expected to use and help improve these tools.
“Technology can throw out every answer you could imagine,” Voorheis says. “We need to continually evaluate the use of AI so that it is human-centered, not technology-centered.”
With a front-row seat to pharmacy’s evolution, alum and Epic executive Joel Jones shares insights on how AI is elevating pharmacists’ roles.
With support from a PATCHES Pilot Award, Assistant Professor Paula Voorheis partners with a local Madison company to refine its digital care-coordination app.