Where Geisinger, Security Health Plan execs see promise in AI

2023-05-11

Preview

来源: FierceHealthcare

There’s a lot of excitement about AI these days, but also plenty of concern. Health plan executives highlight the potential of the tech.

When Kaiser Permanente announced earlier this month that it plans to purchase Geisinger Health Plan, industry observers noted that both companies believe artificial intelligence can make healthcare faster, smarter and less expensive.

For instance, take a health system that is looking at 1,000 patients and trying to determine who may have the greatest risk for colon cancer and should be encouraged to get a colonoscopy before they turn 50, the recommended age for low-risk patients.

But what about high-risk patients, who should probably be screened sooner? AI could gather the information that's needed to bring in the patients at greatest risk.

"If you ask a 45-year-old, ‘Hey, are you ready for a colonoscopy?', it’s not something that they’re eager to do," Aalpen Patel, M.D., medical director for artificial intelligence at the Geisinger and Steele Institute for Health Innovation, told Fierce Healthcare. "But if you tell that 45-year-old that you need a colonoscopy because you’re at a very high risk of getting colon cancer, that changes the conversation. They come in much, much earlier.”

AI is a disrupter, looked upon as an exciting opportunity in some quarters and a dire threat in others. Either way, it's a key priority that many payers are already focused on, and, if they aren't already, they will be in short order.

Aalpen Patel, M.D." data-entity-type="media" data-entity-uuid="a87a58c1-bf6a-4f8f-9065-580d2ee3ce66" data-view-mode="thumbnail"/>

“About one in four of our plans report using AI in some form,” Jeff Van Ness, the vice president of communications for the Association of Community Affiliated Plans, a trade association representing 78 not-for-profit health plans, told Fierce Healthcare in an email. “Given that the last time we asked the question was in 2021 and the rapidly growing profile of AI since then, I’d speculate that that number is significantly higher today."

"Tasks reported as AI-assisted include quality analysis, risk adjustment, predictive modeling, medical record review, network monitoring, and data loss prevention, to name a few," he said.

The potential for insurers

AI represents the latest step in the automation of healthcare procedures, a level above robotic process automation, which uses digital triggers to perform office tasks such as extracting data from electronic health records and filling in forms.

There’s some technological overlap, John Bulger, D.O., Geisinger’s chief medical officer, said in an interview.

“It is not a one or the other question," Bulger said. "They work together to help improve efficiencies. RPA mostly replaces repetitive, non-decision-based tasks and AI helps replace tasks that require decisions, but it is a continuum. We currently use and will continue to use all of these tools.”

Security Health Plan, which has about 210,000 members who reside mostly in Wisconsin, hasn’t moved up from robotic process automation to AI, but plans to, Todd Preston, the company’s director of information systems, told Fierce Healthcare.

“With the automations that we’ve built so far, we built those to be able to make decisions so it can say: ‘If this condition is met, do this action. If a different condition is met, take an alternative path,'" Preston said. "So, it has those capabilities, but it’s not self-learning. If we find that a mistake was made, we have to go correct the automation and then redeploy it. Whereas AI would be self-correcting.”

John Bulger, D.O." data-entity-type="media" data-entity-uuid="c33e9a11-0dc6-4706-9c48-d479d03fe014" data-view-mode="thumbnail"/>

A large segment of Security’s membership has come to expect automated interaction when they call the health plan for help, so Preston said he believes that interacting with AI won’t be that much of a cultural shock.

In fact, many Security members already interact with AI provided by a vendor.

“We don’t get all the information that we need with claims data,” Preston said. “So, we’re leveraging a third-party application that uses artificial intelligence to ingest medical records for our members and it’s able to look at those medical records to find things that our coders would need to take a look at.”

The application Security Health uses is Lumanent Retrospective Review and the vendor is Health Fidelity (acquired by Edifecs), Preston said. The application leverages AI capabilities that are part of the product.

Preston added that potentially those “coders” that double-check the information would be AI algorithms as well.

“Right now, we’re using it to scour the very document-heavy unstructured medical records to find the pieces that a coder would need to take a look at," he said. "We still think a human coder is needed to make the final decision, at least for now.”

Preston said that Security uses the RPA UiPath, which is manufactured and sold by Gartner. “That’s really to streamline operations, and automate things such as claim processing, where the steps to adjudicate a claim are well defined,” he said.

Preston said he envisions implementing a conversational AI process that will answer patients’ calls but would hand the call off to a human being if the “question gets too in-depth.”

Health insurers aren't the only payers taking an interest in AI. Employers want to tackle rising healthcare costs and may turn to AI to help.

Todd Preston" data-entity-type="media" data-entity-uuid="2537079f-f4c2-4117-8518-73be797298aa" data-view-mode="thumbnail"/>

"PBGH and its members are in constant pursuit of innovative solutions that reduce cost without jeopardizing transparency, care quality or equity, and AI could be one of those avenues; but it is still early, and more work needs to be done to ensure adequate safeguards are in place to provide assurances to large purchasers that individuals don’t slip through the cracks," Randa Deaton, vice president of purchaser engagement at the Purchaser Business Group on Health, an organization comprising about 40 large private and public companies, told Fierce Healthcare in an email.

The major risks in deploying AI

Sounds wonderful? Proceed with caution, experts warned.

“AI-generated fabrications, errors, or inaccuracies can harm patients and physicians need to be acutely aware of these risks and added liability before they blindly rely on unpredictable machine-learning algorithms and tools," Jesse M. Ehrenfeld, M.D., the president-elect of the American Medical Association, said in a statement.

Ehrenfeld noted that the Office of the National Coordinator for Health Information Technology (ONC) for the first time since 2012 issued a proposed rule change (PDF) that would revise clinical decision support certification criteria to account for newer technologies.

The rule would take effect Dec. 31, 2024, and, among other things, said that decision-support interventions must address equity and identify potential biases "to expand the use of these technologies in safe, appropriate, and more equitable ways.”

Kevin Kavanagh, M.D., the president and founder of the patient advocacy organization Health Watch USA, noted that the U.S. Office of Copyright ruled in March that “copyright can protect only material that is the product of human creativity.” He also wondered just how broadly this rule can be interpreted.

“The visit documentation has been historically included in insurance payments and the type of payment is also predicated on a human doctor/patient interaction,” Kavanagh told Fierce Healthcare.

As its use grows, providers could also view AI’s proficiency as a threat to their livelihoods.

For example, a study last month in JAMA Internal Medicine compared answers given by doctors and to those given by chatbot to 195 randomly selected patient questions on a social media forum. The responses from the chatbot were judged to be nearly 10 times more empathetic than those given by real doctors and 3.5 times more likely to be of better quality, according to the study.

Bulger said he agrees that the JAMA article could raise concerns but added the definition of AI is still “pretty broad and very heterogeneous. It’s taking processes that a human had to do over and over again and automating that.”

Where Geisinger is focusing its efforts

Patel said that Geisinger’s tech team works with Tempus, a cardiology healthcare technology company that uses AI, to review electrocardiograms.

“It’s not in clinical practice yet because there are a lot of things that have to happen first,” Patel said.

However, as an experiment, the Tempus AI converted 1.8 million ECGs into voltage, he said.

“And the idea was with these so-called normal ECGs: Is there additional information? And the answer is yes," Patel said. "One of the questions we asked was: ‘What is the probability that somebody’s going to develop atrial fibrillation within the next year?’ If the probability is high, then you can imagine embedding a device to monitor the patient’s rhythm on a much more long-term basis and intervene proactively if it flips into afib.”

Bulger said that when he was in practice, a patient might come in and say they feel fine, but his stethoscope detects afib.

“And you have no idea how long they’ve been in atrial fibrillation, and because of that, they’re at much higher risk for stroke and other complications," he said. "Where AI will come into scenarios like that is that years before that ever happens, we’ll be able to understand who’s at risk for atrial fibrillation and monitor them in a different way.”

AI could also revolutionize claims processing and prior authorization, said Patel.

“Where we get concerned is when a health plan’s system is overwhelmed with claims," he said. "You’re flooded with a bunch of claims, and you need to check those claims to make sure they’re accurate, and to make sure they represent something that was actually done. Those are areas that are ripe for CMS or private payers to use AI.”

In such a scenario, AI would “kick out” claims that seem inaccurate and a human would need to double-check, said Patel.

Patel said that health plans could use AI for prior authorization. Preston echoed the sentiment, noting that many prior authorization requests come in via fax.

By deploying AI, the providers get a “yes” or “no” immediately, rather than waiting at length for the authorization to come through. This would be especially critical in cases where chemotherapy needs to be started or adjusted, for example.

“Mrs. Smith has breast cancer,” said Patel. “Mrs. Smith needs to be treated for breast cancer. Is the treatment I’m going to use for Mrs. Smith for breast cancer on the health plan’s formulary? Because every health plan has a different formulary. Treat Mrs. Smith today, don’t sit and wait a week for it to get through the queue and then treat Mrs. Smith.”

AI would also help when it comes to population health management, said Bulger. The data these tools gather can offer "more nuanced recommendations to the clinicians about where to go next" in treating specific populations.

“There’s just more computing power," he said. "So, you’re able to do things on much larger populations and much faster."