How AI is already evolving, and enabling diverse healthcare improvements

Real-world examples from Premier Inc. showcase how machine learning and natural language processing are helping providers use structured and unstructured data to drive more efficient operations, innovate care delivery and improve patient experience.
By Andrea Fox
11:20 AM

Photo: Panuwat Dangsungnoen/EyeEm/Getty Images

Hospitals and health systems are using artificial intelligence to advance patient selection for clinical trials, avoid healthcare supply chain disruptions, uncover early-stage disease, decrease patient wait times – and those are just a few of the myriad AI use cases already in action, according to a virtual congressional briefing hosted this past week by Premier Inc.

Diversifying and standardizing data

Premier, Inc. is a digital transformation company that works with 4,350 U.S. hospitals and health systems and more than 300,000 other providers and organizations, according to its website. The company finds that hospital systems are currently looking to AI to support analyses of patient data and workflow optimization, according to Mason Ingram, director of payer, policy and government affairs.

In its Congressional advocacy road map announced in August, the company said it believes that while "AI can and should play a critical role in advancing healthcare and spurring innovation," it also believes "AI cannot and should not replace the practice of medicine."

However, connecting underserved patients to clinical trials is an area AI can support and refine, according to the company.

"The healthcare industry, in general, is continuously seeking high-quality data, but the data needs to be standardized, connected, representative of large, diverse communities and reflect the state of clinical practice in the general population," Ingram said.

Such data can lead to finding the most appropriate patients for real-world evidence and clinical trials, according to Denise Juliano, group vice president of life sciences.

"This structured and unstructured data and the use of AI, machine learning and natural language processing can supplement our work in research," she said.

Working with unstructured data

With both structured and unstructured data, Premier has accelerated its efforts in getting "the right patient with the right trials," said Juliano.

With structured data, the company employs predictive analytics and AI to accelerate the work of clinical trials. But she said that the work with the unstructured data – narratives – is an area AI can really make an impact. "Because this is very, very powerful information."

Discharge summaries and clinician -patient conversations are often rich with information that can help advance the care for those patients, said Juliano. Use cases in radiology and histology showcase what AI can do for healthcare.

"It will never replace the clinical practice, but it can help really innovate and supplement the practice,” she said before diving into the first use case.

Analyzing COVID-19-era lung scans

Based on a hunch that a large number of patients during the COVID-19 pandemic would be getting lung scans, the company worked with the largest health information exchange in the country and partners in the life science industry, using AI, ML and NLP techniques.

She said the researchers initially hypothesized they would find between 10,000 and 20,000 patients that had an indication of early-stage lung disorder.

They found that 152,000 patients had pulmonary nodules within their lungs, said Juliano, noting that the analysis for 6.3 million patients took three weeks. 

"Without this technology, the human eye or the human person could never actually comb through 60 million documents with that rapid speed," she said. "So this was really revolutionary. It was the largest observational trial of its kind."

Nodules are often referred to as an incidental finding, an "incidental pulmonary nodule," but by the literature, one-third could potentially have lung cancer, Juliano said.

"We actually could help find patients early on that potentially could have lung cancer," she explained. "It offers the power to identify patients early on in their potential disease state and get them on a care plan.

For phase two, Juliano said Premier is working with some New York health systems to bring in patients, identify them and get them on a treatment plan.

"In this particular study, 12% of the study population was black Americans, which is again a lot higher than otherwise you would have found, again, with the human eye for this work," she said.

Predicting Alzheimer's, improving dementia care 

In the next example, Premier leveraged both the structured data of more than one million patients across 275 sites and the unstructured data in early patient identification for Alzheimer's Disease and dementia.

What she said Premier calls data ontology helps predict whether a patient would go on to a definitive diagnosis.

Using predictive analytics and unstructured data can support early intervention with early Mild Cognitive Impairment, Juliano explained. The AI helped "find that needle in the haystack" and sussed out the terms that could be indicators of disease at a later stage.

They also use these techniques to accelerate clinical trials that are underway "because 80% of U.S. clinical trials fail to meet their recruitment timelines," she said, noting that it takes 90 months to complete a clinical phase three or phase four trial.

Cost is also a factor here – bringing a new drug to market costs more than $2.6 billion, she said. Trial delays could cost "$600,000 to $8 million per day."

AI is a powerful way to support "getting the right people on the products that are appropriate for their care," said Juliano.

Diversifying clinical trials

Black Americans, Latinos or Latinx are usually underrepresented in clinical trials, she said, so they developed a model to improve the diversity of clinical trials.

If you think about the traditional model, "they usually try to identify a physician investigator in the site first," she said. "Then they hope those patients are in the healthcare systems.

"What we developed at Premier is a model that flips the funnel," she explained. "We actually apply the inclusion and exclusion criteria to both our structured and unstructured data. We find out where those patients are, then we recruit the system, and then we recruit the physician investigator," Juliano said.

Building supply chain resiliency

Angela Lanning, chief operating officer within Premier's healthcare informatics division, shared how the company is leveraging AI, NLP and ML in healthcare administration and supply chain.

In 2019, Premier partnered with UPMC to use machine learning to help guide procurement and recommend formulary management decisions in real time.

But after the pandemic presented the stark reality "how catastrophic supply disruptions are to our ability to provide safe, timely healthcare to our patients," the company leveraged its big data and used AI "to scrub" it for leading indicators – line delays and products that are in shortage.

Once health systems can predict when a product will be three to six weeks from a shortage, "we can do something about it," she said.

"This is significant because we are able to do this for over 80% of the medical and surgical supplies in healthcare today."

She said the company is working with health systems and suppliers to mitigate shortages, either through a supplier increasing production or helping health systems identify substitutes that they can use to minimize supply chain disruptions to patient care.

Automating end-to-end prior authorization

A survey last year by the Council for Healthcare Quality found that only 28% of prior authorization activities were fully electronic, Lanning said.

When a health plan denies a physician-recommended procedure, "you end up in this circular loop with the physician and the payer trying to get this procedure approved."

It's not only difficult for patients, "it's very manual and inefficient," she said.

Physicians have to intervene with faxes and forms or their administrators do. According to Lanning's presentation, the Centers for Medicare and Medicaid Services estimated that physicians and physician group practices spend an average of 13 hours per week on manual prior auth activities.

"We don't have the people in the health system … who should be taking on these non-value-added processes when we have the ability to automate it," she said.

There is a lot of tension in the prior auth process, and AI can help address it and strengthen payer-provider relationships, she said.

"It was important to us to figure out how we could work with our health systems and the health plans to identify patients that are eligible along with the appropriate clinical indications to authorize the procedure in real-time," Lanning said.

Back in 2014, there was to be a mandate that health systems would be required to use clinical decision support to gain compliance around appropriate use. While that mandate has been delayed, Premier "integrated decision support into the electronic health record so that it's within the workflow for the physician."

The evidence is coded in the EHR and it's 100% automated.

"While Medicare did not mandate it, this has helped us with the commercial payers around getting advanced imaging approved," said Lanning.

Premier has also worked with health systems for more than 10 years to embed CDS within EHRs to get reimbursed for radiology tests.

While the agency requires that providers have a significant clinical review for payment approval, Lanning said AI has helped to automate known patient cases in real time.

"What we've done to help providers manage this process is only give them the most complex patients to review from human interaction," she said. "We've also coupled that with radiologists on our team so that they can provide the ordering physician with the best test to provide the best patient outcome."

It's completing 30-minute processes in less than a minute, she said, and patients are getting a much better experience.

Premier, along with HIMSS (parent company of Healthcare IT News) and others, is part of the Patient ID Now coalition. In 2021, the organization introduced a framework for national patient identity, which it says could streamline the flow of patient information, but there has been little legislative movement.

"It will also remove obstacles to care coordination and nationwide interoperability, as well as save millions in associated costs for the healthcare system," said Blair Childs, former Premier Inc. senior vice president of public affairs.
 

Andrea Fox is senior editor of Healthcare IT News.
Email: afox@himss.org

Healthcare IT News is a HIMSS Media publication.

Want to get more stories like this one? Get daily news updates from Healthcare IT News.
Your subscription has been saved.
Something went wrong. Please try again.