How AI is helping fix rural and Native American health challenges
BOSTON – At the HIMSS Machine Learning & AI for Healthcare event on Thursday, William Paiva, executive director of the Center for Health Systems Innovation explained, how his organization is leveraging data and technology to address some fundamental challenges to underserved rural and Native populations.
Based out of Oklahoma State University, CHSI was built in part from an endowment from late Cerner CEO (and Oklahoma native) Neal Patterson – and also from a Cerner-donated clinical database containing de-identified information from some 63 million patients.
Its aims, said Paiva, are threefold.
- Find innovative ways to tackle the inadequate care available to rural and Native Americans in the region;
- Improve that care delivery with help from predictive data modeling and machine learning tools; and
- Eventually "make Oklahoma the Silicon Valley of rural and Native American Health Innovation and predictive Analytics."
Like many similarly underserved areas, rural Oklahoma suffers from challenges related to both supply and demand, Paiva explained: a pronounced lack of "-ologists" and other medical specialists; lack of technology layers; and lack of emergency medicine resources.
At the same time its patient populations have very high rates of chronic conditions (obesity, COPD) and tend to be older, poorer, under- and uninsured.
The Center for Health Systems Innovation was set up between the business school and the medical school at OSU for a reason, said Paiva. In its creation and deployment of new care delivery models, the center is keenly focused on implementation as innovation.
Part of the way it's making good on these tech-enabled models is by the creation of health access networks, practice-based research networks and hospital networks to ensure patients are getting the most out of its innovations.
In its surveys of 250 local providers, CHSI found that a majority of the challenges reported – 70 percent of them were not clinical but operational.
The imperative, said Paiva, is to harness some of the 14 million terabytes of medical data collected nationally each year. That's where Cerner's HIPAA-compliant dataset – with de-identified but time-stamped clinical, financial, pharmacy and lab data of 63 million patients – comes in.
It helps the center answer the pop health questions so many other health systems are facing: "How do I identify the two or three people in this sea of people who are most at risk?"
For instance, it's already helped the CHSI make headway with improving annual exam compliance rates for diabetic retinopathy patients , which typically hover around 10 to 30 percent.
With help from machine learning algorithms, the center was able to identify nine new lab variables that had yet to be associated with DR, improving the sensitivity and specificity of screenings to 95 percent.
That's allowed for creation of a tool that can solve that "ologist problem" by arming primary care physicians with definitive data: They can tell diabetic patients they're at risk for losing their eyesight – something that will motivate most patients to maintain their annual compliance.
Besides the obvious benefit to the patients there, Paiva said rural areas such as Oklahoma offer a good opportunity to solve a common conundrum for new AI and ML tools: the Catch-22 between getting validation and gaining traction.
"Consider rural areas for your validation," he said. "If you're looking for markets for quick adoption, the rural market shouldn't be overlooked."