A federal class-action lawsuit that began with the deaths of two Medicare Advantage patients has become one of the most closely watched tests of how far insurers can lean on artificial intelligence to decide what care gets paid for. In Estate of Gene B. Lokken v. UnitedHealth Group, filed in the District of Minnesota, plaintiffs allege that UnitedHealthcare and its subsidiary naviHealth used an AI tool called nH Predict to cut off coverage for extended-care facility stays, overriding physicians’ own judgment about how long patients needed to remain under care. The case has survived a motion to dismiss, cleared a major discovery fight in March 2026, and is now forcing UnitedHealth to hand over internal documents about exactly how the algorithm works.
The Number at the Center of the Case
The lawsuit’s central statistic is stark: roughly 90 percent of nH Predict’s denials, when appealed to a federal administrative law judge, are ultimately reversed. Plaintiffs argue that figure is not a rounding error but evidence the tool systematically denies care that Medicare rules actually require insurers to cover, and that UnitedHealth kept using it anyway because most patients and families never appeal—meaning the algorithm’s errors go uncorrected far more often than they get caught. The named plaintiffs’ estates say their family members paid tens of thousands of dollars out of pocket for care that was cut off by nH Predict, only to have those denials overturned on appeal after the money was already spent.
What the Judge Has Allowed to Proceed
In February 2025, Judge John R. Tunheim dismissed five of the original seven counts in the case, ruling that claims like unjust enrichment and bad-faith insurance practices were preempted by the federal Medicare Act. But he allowed the breach-of-contract and breach-of-the-implied-covenant-of-good-faith-and-fair-dealing claims to move forward—narrower legal theories, but ones that get directly at whether UnitedHealth violated the terms of its own Medicare Advantage plans by using an algorithm to make coverage decisions. Then, in a March 9, 2026 order, the court granted or partially granted the plaintiffs’ discovery requests across six of seven categories, ordering UnitedHealth to produce a broad set of internal documents about nH Predict’s design, validation, and use.
A Pattern, Not an Isolated Case
The Lokken case is not happening in isolation. A parallel lawsuit accuses Cigna of using an algorithm called PxDx to automatically deny patient claims in batches of hundreds or thousands at a time, allegedly without any human reviewer opening individual patient files before the denial went out. Data cited in that case suggests roughly 80 percent of PxDx denials are overturned on appeal—a strikingly similar pattern to nH Predict’s reversal rate, suggesting the issue may be structural to how insurers deploy claims-denial AI broadly rather than a defect unique to one company’s software.
UnitedHealth’s Defense
UnitedHealthcare has consistently maintained that nH Predict functions only as an informational guide for case managers, not as the actual mechanism that decides whether a claim is approved or denied, and that final coverage decisions are still based on Medicare coverage criteria and the specific terms of each patient’s plan. The company argues that treating an internal planning tool as if it were the decision-maker mischaracterizes how the software is actually used inside its claims process, and that physicians and case managers retain the authority to override the tool’s output.
Why Patient Advocates Are Skeptical
Patient advocates and some former UnitedHealth case managers who have spoken publicly argue that in practice, case managers face pressure to align their decisions with what the algorithm recommends, making the “just a guide” characterization more theoretical than real. They point to the discovery order itself as telling: if nH Predict genuinely played no role in denial decisions, they argue, there would be little reason for UnitedHealth to resist turning over documents about how the tool was built and validated in the first place.
What’s Next
With discovery now underway, the case is likely to produce internal UnitedHealth documents over the coming months that could reveal, for the first time in detail, how the company validated nH Predict’s accuracy before deploying it—or whether it validated the tool’s real-world performance at all. Legal observers expect the outcome to shape how insurers across the industry use AI in coverage decisions going forward, regardless of which side ultimately wins, since a ruling that AI-driven denial patterns can constitute breach of contract would create a legal template for challenging similar tools at other insurers. For the millions of Medicare Advantage enrollees whose care decisions increasingly run through some kind of predictive algorithm, the practical stakes of this case are about as concrete as health policy gets: whether an AI tool with a 90 percent reversal rate on appeal can keep making the first call.
