For patients and families fighting to keep a loved one in rehabilitation care, an algorithm most have never heard of has quietly been deciding how long that care lasts. A federal court case in Minnesota is now forcing that algorithm into the open. On March 9, 2026, a Minnesota federal judge handed UnitedHealth Group one of its most damaging legal setbacks in years, ordering the insurance giant to produce internal documents about its nH Predict algorithm by April 29 — documents plaintiffs say will show the tool systematically cuts off coverage for rehabilitation care that patients still need.
What nH Predict is accused of doing
The case, known as the Lokken litigation, was originally filed in November 2023 by patients and the heirs of former UnitedHealthcare Medicare Advantage enrollees. Their central allegation is that nH Predict denies rehabilitation care even though UnitedHealth is allegedly aware the algorithm carries what the complaint describes as a 90 percent error rate — a figure the plaintiffs say refers to the rate at which nH Predict’s denials are reversed when patients successfully appeal them. UnitedHealth has not conceded that characterization, and the figure remains a plaintiffs’ allegation rather than an independently verified finding, pending the document review the court has now ordered.
Why this case has moved further than most
Algorithmic claim-denial lawsuits against insurers are not new, but few have reached the discovery stage this forcefully. The magistrate judge’s order compelling document production, later extended slightly after plaintiffs agreed to a reciprocal 45-day extension on their own expert-declaration deadline related to an expected motion for class certification, puts internal UnitedHealth records about how nH Predict was built, tested and deployed directly in front of plaintiffs’ attorneys — exactly the kind of internal evidence that past insurer lawsuits have struggled to obtain.
It isn’t the only insurer facing this exact fight
UnitedHealth is not alone. In Kentucky, a separate lawsuit, Barrows v. Humana, accuses Humana of using AI tools to improperly deny Medicare Advantage claims for post-acute care; that case received a scheduling order in July 2026 and remains active. In California, a class action against Cigna, Kisting-Leung v. Cigna, alleged the insurer’s PxDx algorithm violated the duty of good faith and fair dealing; a separate secondary report citing ProPublica claims Cigna’s PxDx system denied more than 300,000 claims in just two months of 2022, though that case saw a dismissal order in August 2026 whose underlying resolution is not yet fully clear from public filings.
The regulatory backdrop insurers now face
By late 2025, 23 states and Washington, D.C. had adopted the National Association of Insurance Commissioners’ AI Model Bulletin, a framework intended to govern insurer use of AI in coverage decisions, and California’s SB 1120 — which one source describes as requiring that physicians, not algorithms alone, make final medical necessity determinations — took effect January 1, 2025. Patient advocates argue these rules are catching up to, rather than getting ahead of, years of algorithmic denials already in patients’ rearview mirrors.
The insurer’s side of the argument
UnitedHealth has consistently maintained that nH Predict is used as a decision-support tool for clinicians rather than a final arbiter of coverage, and that licensed medical professionals, not the algorithm itself, make the actual denial decisions — a defense echoed by Humana and Cigna in their respective cases. Insurers broadly argue that predictive tools like nH Predict help standardize care decisions and flag cases warranting closer clinical review, rather than functioning as an automated rejection machine, though plaintiffs counter that in practice, clinician review of algorithm-flagged cases has been cursory at best.
What’s next
The document production ordered for April 29 will be the pivotal moment in the Lokken case: whether the internal records support the 90 percent reversal-rate allegation, and whether they show UnitedHealth knew about accuracy problems while continuing to use the tool. If plaintiffs’ allegations hold up under the released documents, the case could become a template for discovery in the Humana and Cigna suits, and could accelerate pressure on federal regulators to close the gaps that Georgetown’s health policy researchers say the current Medicare Advantage rule left unaddressed around AI-driven prior authorization. Patient advocacy attorneys following the case say they expect UnitedHealth to seek a protective order limiting public disclosure of whatever the documents show, a fight that could itself take weeks to resolve and delay when, or whether, the public ever sees the underlying data behind the algorithm’s track record.
Photo: Engin Akyurt / PEXELS via Pexels
