Achieving healthcare efficiency and reducing costs is increasingly a twin goal of hospitals, payers and providers, and many are turning to machine learning algorithms to help achieve these goals, while also publicly saying it will not compromise the quality of care and patient outcomes. But leveraging AI/ML has its own pitfalls, and human judgment can make the difference between reinforcing systemic racism and mitigating it.
This is the message of Dr. Farzad Mostashari, former National Health IT Coordinator and CEO of Aledade, who participated in a panel discussion on public and population health moderated by Hemant Taneja, managing partner of venture capital firm General Catalyst, and collaborated with panelists Caroline Savello, Chief Commercial Officer color, a genomics testing and population health company based in Burlingame, California. [A third panelist, Dr. Robert Wachter, professor and chair of the Department of Medicine at UCSF joined the virtual panel that was part of the annual virtual J.P. Morgan Healthcare conference.]
The conversation has inevitably turned to health equity, a topic of paramount concern for many healthcare stakeholders given how Covid-19 has exposed inequities in the U.S. health system.
Savello describes the health equity challenge primarily as an access issue. Through Color’s experience in the pandemic, she makes the argument that the infrastructure built up as a result of Covid – whether through government efforts or private efforts – can and should be used to address issues in underserved communities access to health care.
Savello describes how Color was conducting HIV screening and testing at the same location while Color was delivering vaccines and Covid testing was being done at historically black churches.
“Color now operates 8,000 health care locations across the country,” she declared. “We’ve started HIV testing at African-American churches next to vaccine and testing sites. We’re seeing 40 percent of people opting to get tested and screened for HIV because it’s there and it’s convenient.”
More telling are the different statistics. More than 60 percent of those who opted in had never had HIV screening in their lives or had not done so in the past year. In other words, this effort to go where the community is will lead to higher levels of engagement.
“When you see Childhood immunizations in schools, or you see HIV testing in historic African American churches, I think it really changes the way individuals think about health care and the nature of where they can get basic health care,” Savelo said.
Turn to Farzad Motsashari, former National Health Information Technology Coordinator, current CEO Aledad, to comment on health equity and the impact of Covid on communities of color, he framed the discussion as a national reckoning of race that coincides with the unrest sparked by the May 2020 murder of George Floyd .
Later, he went on to describe the challenges of achieving health equity, since the system doesn’t even require providers to measure disparities.
“we [Aledade] Taking global risks, so we have 1.7 million patients working with us – more than 1,000 clinics. If we can reduce hospitalizations, reduce adverse events, and improve quality, we get health plan contracts or Medicare contracts that reward those practices,” he explained. “And there’s no requirement that you even look at the difference. So I think that’s where we should start. All of these value-based payment models, should require you to stratify — in fact CMS can even do this for us — to stratify the quality reporting that we already do in usage reports and race cost reports. “
People can’t wait to start measuring based on race, he said. Mostashari’s most interesting comments, however, come from the anecdotes he shared about the dangers of AI/ML in healthcare, crafted word for word below.
I didn’t intend to share this, but it might be interesting in a technical context, especially considering ML/AI applications. To me, this is a great example of the need for human judgment around the questions we ask.
We want our practice to reach patients who need care during this time, who are suffering from a lack of primary care, and Population Health 101 is right to reach patients.
These practices have proven to be very busy and understaffed, making it difficult to reach a cohort of pf patients. So the question being asked was, “Gee, maybe we should prioritize this list, not only based on the risk this person faces, but also on their likelihood of being successful in the practice.”
Seems like a good question, right? Let’s do the most useful work.let’s put people on the list [with whom] You will have a conversion rate. So we did. We have a good ML model that can significantly improve the experience of the practice and thus increase efficiency. Then, thankfully, one of our staff members said, “What impact will this have on racial equality?”
We found that while 15% of our patients were black, only 8% of the patients who were thought to be most affected by phone calls were black.
That’s the definition of systemic racism. why? Because historically when these people get a call, either the phone number doesn’t work, or they don’t update, or they don’t answer the call for some reason, or when they’re on the other end of the phone with them Conversation, resulting in lower engagement rates. So we can have ML/AI predict who is unresponsive and it will tell us “Hey, yeah, don’t call those black people.” It will only exacerbate the disparity we already have in the system.
Or we could ask a different question, “What predicts higher black participation?”
I just thought we were easy [say], “It’s a little random forest, put it in to get the answer, sort the list”, never even thinking about how we can further enforce the differences that exist in our system.
Image credit: Andrii Shyp, Getty Images
