Proponents of artificial intelligence and machine learning have touted for years how algorithms hold significant promise to streamline healthcare administration and delivery, freeing up nurses and physicians to practice at the top of their licenses.
But COVID-19 has shifted the priorities for that technology, speakers said at HIMSS21 in Las Vegas this week.
Before the pandemic, Rush Medical College's major goals were training staff and students, while promoting health equity. Of its artificial intelligence pilots, "many didn't go anywhere," said Bala Hota, Rush's chief analytics officer, on Tuesday.
But as COVID-19 hit, "for the first time, our organization craved the numbers from the top to the bottom … the liquidity of data that we wanted became the No. 1 thing," Hota said.
Speaking on the same panel, Tanuj Gupta, VP of Cerner Intelligence, noted that the EHR vendor's clients want it to do the same tasks — but want more AI getting the job done.
EHR companies help with five main jobs, Gupta said: capturing healthcare information, quickly summarizing that information, coordinating care, securing reimbursement and making critical decisions.
Whether five or 50 years from now, those jobs will still need doing, but "people are asking for different technology to do those jobs,” Gupta said, whether its ambient voice technology or natural language processing to autochart, voice recognition for info summaries, robotic process automation in care coordination or machine learning in clinical decisionmaking.
"The jobs are the same. The tools are evolving. So you're going to see a lot of digital health companies evolving hopefully to bring you those tools," Gupta said.
Cerner, for example, for eight years has only deployed about three to four clinical models to clients. In 2020, after reshuffling investments and ramping up needed infrastructure, Cerner deployed two models 35 times.
And this year, the vendor is on track to deploy 10 models 350 times, Gupta said.
Hota also said that AI and ML tools were becoming more mature, saying Rush is building predictive models to try to bolster its social determinants of health screening platform in a bid to connect in-need patients with resources in their community.
Meanwhile, payers are also taking note of AI's potential.
Anthem Chief Data and Analytics Officer Ashok Chennuru said Tuesday it's "embraced AI" to transmit structured and unstructured data into "proactive, predictive and personalized insights."
But "it's not as simple as developing a model, throwing it in there and then letting it go," Gupta said. Models need to be frequently calibrated, which takes time and money, and developers need to make sure they're not building on dirty datasets. Additionally, it's preferable to end users like clinicians if any output is explainable, which helps with adoption down the line.
And machine learning models are often inconsistent when applied in the real world, according to a Monday presentation by Mujeeb Basit, associate chief medical informatics officer at the University of Texas Southwestern, on an algorithm to predict the likelihood a patient becomes critically ill during a hospital stay.
"There is just not enough data, and models generally aren't designed to deal effectively with that missingness," Basit said.
And the future of AI is iffy, as many healthcare leaders remain uncertain on the ROI of algorithms. One statistic from a survey conducted by management consultancy the Chartis Group released Wednesday found 7% of healthcare executives said they believe AI and ML are the top priority for the future, while the same percentage called them a "distraction."
"There's a general belief that AI/ML is going to create new operational capabilities, but it's not really clear how it fits in the agenda," Tom Kiesau, senior partner with the Chartis Group, said.
One new frontier for AI/ML, however is emergency forecasting in event of further public health crises, experts say.
"Health systems need to predict catastrophes that are going to overwhelm both labor and non-labor supply, in advance," Gupta said. "We should be able to do it with reasonable certainty. We do it with earthquakes, we do it with forest fires, we do it with weather."
Researchers at UnitedHealth's health services arm Optum started studying the COVID-19 pandemic while tracking a flu outbreak in the U.S. in 2019, and began using AI and ML to create infectious disease forecasting, Optum Director of Research Danita Kiser said Wednesday.
The computational epidemiology system — which Kiser said was as accurate as current weather forecasting — can make a global impact by text messaging caregivers early warnings of outbreaks so they can isolate patients.
Currently, Optum is sharing this information internally, and is developing plans to share it with third-party providers.
"If we can forecast a week or two in the future, we can put prevention measures in place to save people's lives," Kiser said.