Clinicians need to gain confidence AI will support care
Healthcare organizations that have successfully implemented the technology say early results must demonstrate tangible assistance in the care process.
Recent studies suggest that the prospects for using artificial intelligence in healthcare are improving, but much work lies ahead in fostering clinician understanding and support for the technology.
In a presentation from the HDM KLASroom, industry experts discussed data from a recent KLAS Research report on AI in healthcare. The panelists contend that the most effectual predictive models are bolstered by clinician understanding on the back end.
“Organizations said that it's great that people can implement predictive models, but they really recommended having change management in place so that organizations can get to outcomes and really help patients or clinicians make better decisions.”
Jennifer Hickenlooper, insights director at KLAS Research, begins by outlining popular applications of AI in healthcare today. Research completed by Orem, Utah-based KLAS Research show that many provider organizations are focused on investing in proven use cases, but that only a few vendors are doing a good job in guiding customers to achieve desired outcomes.
Meeting clinicians’ needs
“These (healthcare) organizations are really taking a problem-first approach as they are implementing AI, rather than asking what cool model they can implement,” Hickenlooper says.
John Halamka, MD, president of the Mayo Clinic Platform, agreed that simply implementing AI tools isn’t enough. “We need to ask, ‘Where should there be human oversight?’ ” he says. “There's got to be context here—an understanding of the patient and their disease state, and human decision making— not just technology for technology's sake.”
If human physicians are to effectively apply AI, those tools that use AI must be curated for physicians, says Christer Johnson, chief analytics officer at Healthfirst.
“We need to learn how to leverage AI in such a way that doctors want to consume the data because it's explainable, understandable and timely. We can't give a doctor 20 different predictive results – it’s got to be one result or one recommendation that is actually meaningful.”
Gaining familiarity with AI
How can provider organizations optimize their AI tools to provide applicable data? They should start by familiarizing themselves with the predictive models and how they will be used, says Hickenlooper.
“Early adopters are sharing that providers need to really make sure that their predictive models don't drift — that users must understand what's going ‘into the box’ so they know what's coming out of the box. Providers need to understand how the models work so that when they’re applying them, the providers really know what they’re doing and don’t make missteps.”
In addition to consumable data and clinicians who understand the origins and applications of that data, successful AI users also focus on change management, Hickenlooper says.
“Organizations said that it's great that people can implement predictive models, but they really recommended having change management in place so that organizations can get to outcomes and really help patients or clinicians make better decisions,” she said. Early adopters feel that this is key to being successful with this technology.”
Johnson of Healthfirst predicted that, over time, predictive models with learnings from whole populations will be able to inform solutions for each individual patient. “We have to get down to the personalized cohort of one and really start talking about the individual patients. I think that's going to help accelerate our ability as a society to bring greater health equity across the entire population.”
In some instances, that’s already occurring, Hickenlooper contended. “We are already starting to see some of these outcomes. Care managers are able to reach out to the right patients to provide some of that preventative care that they need and are not getting. Physicians are starting to see some AI insights in their workflows. As organizations see more outcomes, they'll be more likely to continue to adopt new algorithms and technology and continue to push toward new outcomes.”