UnitedHealth's nH Predict AI wrongly denied 90% of appealed Medicare claims while overriding attending physicians

UnitedHealth's AI Denied Elderly Patients Care. It Was Wrong 90% of the Time.

UnitedHealth Group told investors, regulators, and the public that its nH Predict algorithm improved efficiency in post-acute care decisions. It did not tell them the algorithm overrode attending physicians and was wrong on appeal 90% of the time.

The Federal Order

A federal magistrate ordered UnitedHealth to produce the internal nH Predict documentation by April 29, 2026 — specifically the model's training data, decision logic, and any internal audits tracking error rates. The order followed evidence in a class action lawsuit that the algorithm denied medically necessary skilled nursing care for Medicare Advantage enrollees, often within days of hospitalization.

What the Algorithm Did

nH Predict set a fixed "expected length of stay" for each patient condition. When a patient exceeded that predicted window, coverage was automatically terminated — regardless of what their doctor recommended. Physicians filed appeals. UnitedHealth denied them. When those denials were independently reviewed, 90% were reversed as medically unjustified.

The Business Model

This is not a bug. A system that denies claims at high volume, collects premiums from Medicare Advantage, and processes appeals slowly enough that some patients die or give up before they're resolved is a revenue strategy. The April 29 document deadline will tell us how much of that strategy was deliberate.