AI Algorithm allegedly leads to Medicare Advantage Denial of Claims
Industry has been talking about AI (Artificial Intelligence) for a while now. And of course, industry has been looking to reduce the amount of labor input with AI. Having used computerized systems such as MRPII, ERP, etc., I have always followed the wisdom of people such as Oliver Wight and George Plossl. The system is as accurate as the input to it. Which means, you have to check your work. A system can also be used to intentionally mislead in its results. That is the topic of this commentary.
A Law Suit alleges the company uses an algorithm to “prematurely and in bad faith” halt payments for healthcare services.
UnitedHealth AI algorithm allegedly led to Medicare Advantage denials, a lawsuit claims, Healthcare Finance, Jeff Lagasse, Associate Editor
A class action lawsuit filed this week alleges that healthcare giant UnitedHealth Group unlawfully used an artificial intelligence algorithm to deny rehabilitative care to sick Medicare Advantage patients.
The suit claims that UnitedHealth knew the AI algorithm had a high potential for error.
AB: This plays into the scenario which others have discussed about Medicare Advantage plans deliberately denying claims. In “Medicare Advantage Artificial Intelligence can Limit Care for Patients.” – Angry Bear, MedPage Today features an article by KFF’s Susan Jaffe. Using naviHealth, a care management company it bought in 2020 it was able to deny care. naviHealth is one of several businesses using computers helping insurance companies make coverage decisions. So here we are again discussing the same issue.
The Story
The algorithm in question is nH Predict, which was developed by UnitedHealth subsidiary NaviHealth, which the former acquired in 2020. One of the allegations is that UnitedHealth uses nH Predict to evaluate claims for post-acute care, including in-home care and extended stays in skilled nursing facilities. The company uses the algorithm to “prematurely and in bad faith” halt payments for healthcare services.
A STAT investigation, which was cited in the lawsuit, suggests UnitedHealth pressured employees to use the algorithm to issue payment denials to those on Medicare Advantage plans, setting a goal for employees to keep patient rehabilitation stays within 1% of the length of stay predicted by nH Predict.
The lawsuit alleges elderly patients are being prematurely kicked out of facilities, or forced to dip into their family savings, to continue to receive care. The suit states . . .
“The fraudulent scheme affords defendants a clear financial windfall in the form of policy premiums without having to pay for promised care.”
When federal administrative law judges hear appeals on these coverage denials, about 90% are reversed, according to the complaint. However only a small percentage of patients actually file appeals.
Claiming that use of the algorithm violates patient contracts and various state insurance laws, the lawsuit alleges that nH Predict decided claims without properly evaluating them and seeks a court order to stop the use of the algorithm and to award monetary damages.
In a statement, UnitedHealth said nH Predict is not used to make coverage decisions, but instead is a “guide to help us inform providers, families and other caregivers about what sort of assistance and care the patient may need both in the facility and after returning home.”
The company added that coverage decisions are based on the terms of the members’ health plans and on criteria laid out by the Centers for Medicare and Medicaid Services.
UnitedHealth said the suit has no merit.
According to STAT, Humana and several regional health plans use the Nh Predict algorithm.
The suit was brought by the families of two deceased Wisconsin residents who had Medicare Advantage coverage through UnitedHealth.
According to STAT, Humana and several regional health plans use the Nh Predict algorithm.
The suit was brought by the families of two deceased Wisconsin residents who had Medicare Advantage coverage through UnitedHealth.
The Larger Trend
In July, Cigna was sued for allegedly using algorithms to deny claims.
The lawsuit, filed in federal court in California, claimed Cigna developed an algorithm known as PXDX to enable its doctors to automatically deny payments in batches of hundreds or thousands at a time for treatments that did not match certain preset criteria.
A Cigna Healthcare spokesperson, responding by statement, said the vast majority of claims reviewed through PXDX are automatically paid and that the process does not involve algorithms, AI or machine learning, but a simple sorting technology that has been used for more than a decade to match up codes.
UnitedHealth faced another lawsuit in June over allegedly denying claims. The U.S. Department of Labor brought the lawsuit against UnitedHealth Group subsidiary UMR, claiming it incorrectly denied emergency room and urinary drug screening claims for thousands of patients, thereby failing to comply with the requirements of the Affordable Care Act and the DOL’s claims procedures regulation.
According to that complaint, UMR is UHG’s third-party administrator, providing benefits services to more than 2,000 self-funded employer health plans.
Comcast, our cable, internet & phone service provider, currently provides over-the-phone customer service using a fairly sophisticated AI. They did so previously with an un-sophisticated one. The difference, it seems to me, is that the old one would turn you over to a ‘live agent’ if you insisted. The new one will not do that under any circumstances, or so it seems. We had some phone service problems a month or so ago, which require calling them.
After an extremely frustrating 20 minutes, I ended the call, hopped into my car & drove over to the nearby support center, and they fixed us right up. They smirked when I remarked about how bad the phone support was. Then, a couple weeks back, the sign went up that ‘This Office is Closing on Nov 30.’
So it goes. Bad Service? Get used to it.
AI entities simply have to SEEM intelligent. That fulfills the requirements of the Turing Test for artificial intelligence. That apparently does not include actually being ‘smart’.