A federal judge in Minnesota dealt UnitedHealth Group one of its most damaging legal setbacks in years on March 9, 2026, ordering the insurance giant to hand over a sweeping set of internal documents about nH Predict, the AI algorithm at the center of a class-action lawsuit alleging elderly Medicare Advantage patients were denied medically necessary post-acute care not by a doctor’s judgment, but by a machine.
A 91-year-old’s broken leg and a lawsuit that followed
The case, Estate of Lokken v. UnitedHealth Group, Inc., was filed November 14, 2023, in the U.S. District Court for the District of Minnesota. Its named plaintiff, Gene Lokken, was a 91-year-old Wisconsin man who fractured his leg and ankle in May 2022. UnitedHealthcare covered 19 days of his rehabilitation before nH Predict flagged him for discharge. His family disagreed with the algorithm’s determination and paid out of pocket for his continued care for nearly a year, until his death. The suit alleges UnitedHealthcare used the tool, developed by Optum subsidiary naviHealth, to systematically override treating physicians’ recommendations and cut off coverage for rehabilitation and skilled nursing care.
What the March ruling actually ordered
The March 9 order largely sided with plaintiffs on their motion to compel discovery, granting or partially granting requests across six of seven disputed categories. Specifically, the court ruled UnitedHealth must disclose internal records detailing whether nH Predict was designed to override clinicians’ independent medical judgment — a question that goes to the heart of plaintiffs’ theory that the algorithm functioned as a cost-control mechanism dressed up as clinical decision support. Legal observers describe the order as unusually broad discovery relief this early in a case involving a major national insurer, and it significantly raises the stakes for UnitedHealth as the litigation proceeds toward class certification.
The number driving the case: a 90% reversal rate
Central to the plaintiffs’ argument is an allegation, drawn partly from earlier investigative reporting, that nH Predict carries roughly a 90% error rate — meaning nine out of ten denied claims are ultimately reversed on appeal. If accurate, that figure suggests the tool’s initial denials were systematically out of step with what independent medical review, and often the treating physician, determined patients actually needed. UnitedHealth has disputed characterizations of the tool as the sole basis for coverage decisions, maintaining that human reviewers retain final authority — a defense the new discovery order will now test against the company’s own internal records.
Part of a broader legal reckoning
The nH Predict case is not an isolated dispute. Cigna faces a parallel lawsuit over its PxDx algorithm, filed after a 2023 ProPublica investigation found the tool was used to reject more than 300,000 payment requests over a two-month span in 2022, with Cigna physicians spending an average of 1.2 seconds reviewing each flagged claim before denial. Plaintiffs in that case argue PxDx was used to replace, rather than assist, clinical judgment — echoing the core allegation against UnitedHealth. Both cases remain active, with judges still weighing motions to dismiss as of 2026.
Regulators are moving too, unevenly
State insurance regulators have not waited for the courts to catch up. By late 2025, 23 states and Washington, D.C. had adopted the National Association of Insurance Commissioners’ AI Model Bulletin, which sets expectations for how insurers must govern algorithmic decision-making. California went further with its Physicians Make Decisions Act (SB 1120), which took effect January 1, 2025, and requires that only licensed physicians — not algorithms alone — can deny claims on medical necessity grounds. But enforcement remains uneven across states, and industry groups have pushed back on characterizations that AI tools are making final denial decisions rather than flagging cases for human review.
What’s next
With discovery now compelled across most of the categories plaintiffs sought, the Lokken case moves into a phase where internal UnitedHealth communications, algorithm design documents, and training materials could become public record — material that could reshape not just this lawsuit but the broader legal exposure facing every insurer using predictive tools in utilization review. A ruling on class certification is expected later in 2026, and plaintiffs’ attorneys in the Cigna case are watching closely, since a finding that nH Predict was designed to override physicians could become a template argument across the industry’s growing docket of AI claims-denial litigation.
