Clinician involvement key to optimizing AI in healthcare
Reducing bias and increasing trust is crucial in making AI an effective component of a data management strategy.
End users can play a critical role in developing, refining and deploying tools that use artificial intelligence in healthcare organizations, and that is essential in successfully deploying the technology.
Judy Wawira Gichoya, MD, assistant professor, Emory University School of Medicine
“Forget about AI. Data science is what’s going to save our lives.”
Indeed, AI is no silver bullet, and imposing solutions on providers is not the best route to success, said experts on technology deployments in a recent presentation during the HDM KLASroom.
“AI has risen up as one of the key priorities for health systems. And it's been cited in our research as one of the most exciting emerging areas for health technology,” said Alex Nixon, senior research analyst for UPMC Enterprises and the Center for Connected Medicine.
“But despite all of this promise, AI has been sort of slow to be implemented across healthcare,” he added. There are a number of reasons for that. But one of the primary challenges that we found in our research was the lack of trust and buy-in from many across healthcare, especially clinicians.”
Judy Wawira Gichoya, MD, assistant professor in the Department of Radiology at the Emory University School of Medicine, and Oscar Marroquin, MD, chief healthcare data and analytics officer at UPMC, outlined several areas with relatively high adoption and buy-in within AI, such as radiology and stroke diagnosis. They also acknowledge some reasons for low clinician trust in AI, such as biased algorithms and overly generalized tools.
“We know that technology hasn't always been nice for physicians,” Gichoya noted. “But this is an opportunity, really, for us to redesign and rethink the healthcare system. This happens every time that we have a new technology to embrace.”
Marroquin advised healthcare leaders to engage the clinicians whenever the organization intends to build a predictive model.
“Everybody must understand what the strengths and weaknesses of the data are. What are the limitations? Are there issues with the veracity of the data? We must refine these questions with the people who are going to be the end users of the solution at the table,” he said. “That not only increases trust, but also makes it less likely that we're going to choose a data set that will make the algorithm implicitly biased from the beginning.”
Both Gichoya and Marroquin also emphasized that AI should not be viewed as an end unto itself. “We need to think of AI as one piece that is part of the continuum of how we utilize data,” Marroquin said.
“Forget about AI,” added Gichoya. “Data science is what’s going to save our lives.”
Gichoya urged healthcare leaders to focus more on data management than on AI tools. “If you don’t invest in the data, it doesn't matter what tools you buy or what you bring to your enterprise; they are never going to be used. So you have to invest in the data, and you have to invest in the people. You have to think broadly in terms of who is going to support your initiatives.”
Marroquin also urged leaders to be introspective about the data being used. “Come up with a strategy as to how you're going to be more thoughtful in using the totality of the data you have and worry less about the techniques or the tools that you’re going to use,” he noted. “For some things, AI is going to help us accomplish a goal. But for some other things, the solution is right in front of us — it is just using some of the insights we already have and applying them better for our patients.”
To health system leaders wanting to capitalize on AI but wary of complications, Gichoya and Marroquin offered encouragement and resolve. “I know there are many risks,” Gichoya said. “But please don’t miss out because you’re obsessed with the fear that you’re going to make a mistake.”
Watch the Optimizing AI session video or explore the full HDM KLASroom on-demand video series.