Soviet constructivist graphic: a mechanical rubber stamp bearing DENIED descends onto geometric paper forms amid industrial gears and bureaucratic machinery in red and black

The Algorithm Recommends Discharge. The Doctor Objects. The Patient Pays.

UnitedHealthcare deployed an AI tool with a documented 90% error rate to determine when elderly Medicare patients had received sufficient post-acute care. The company notes, with some care, that the tool was not making decisions. It was guiding them.

In a development that will surprise no one who has been paying attention, UnitedHealthcare — the largest health insurance company in the United States — deployed an automated tool called nH Predict to determine when elderly Medicare Advantage patients had received, in The Algorithm's estimation, sufficient post-acute care.

The tool was developed by naviHealth, an Optum subsidiary. It estimated, based on historical patterns, how long a given patient "should" require rehabilitation or skilled nursing care following surgery or illness. When the estimate said the patient was ready, coverage for continued treatment was ended.

The treating physicians — the human beings who had examined the patients — sometimes disagreed. Sources within the Medical Community confirm that physician disagreement, in these circumstances, was not treated as dispositive.

The Algorithm was unavailable for comment.

The Error Rate

Even now, a federal class action in Minnesota is working through the precise dimensions of this arrangement. The plaintiffs allege that UnitedHealth Group deployed nH Predict while knowing the tool had a ninety percent error rate.

Ninety percent. To be precise, dear reader: this means that for every ten coverage decisions The Algorithm informed, nine were wrong.

UnitedHealthcare's position is that nH Predict was not making decisions at all. It was, the company explains, a "guide" — a tool to "help inform providers, families and other caregivers about what sort of assistance and care the patient may need." This is the same guide whose recommendations, patients allege, were used to discharge them before their physicians determined they were medically ready.

One presumes the distinction between "guiding a decision" and "making a decision" has lawyers in both camps working very long hours.

The Outcomes

Patients were discharged too early. Some paid thousands of dollars out of pocket to continue receiving care their insurance was supposed to cover. The class action is brought on behalf of elderly patients whose coverage was cut short over physician objections — cases in which, authorities note, the patient's doctor said one thing and The Algorithm said another, and the insurer chose The Algorithm.

In March 2026, a federal court ordered UnitedHealth Group to hand over broad discovery materials on nH Predict's implementation, granting or partially granting six of seven categories of the plaintiffs' discovery requests. The court, it appears, would also like to understand what "just a guide" means in practice.

The case is ongoing. The patients are older now.

The Pattern

And yet — as if this were not enough — UnitedHealthcare is not unique. Seventy-one percent of health insurers surveyed in 2025 reported using AI for utilization management — the process of determining whether a treatment, medication, or hospital stay is "medically necessary." Less than one percent of denied claims are appealed, not because patients believe the denials are correct, but because the appeals process is lengthy, confusing, and difficult to navigate while also being ill.

The Machine denies your claim swiftly and without ceremony. The appeal will take several months. The discharge date, sources confirm, has already been set.

The experts have reviewed the case. The inbox, as ever, is full.

Sources: CBS News — UnitedHealth uses faulty AI to deny elderly patients medically necessary coverage · Becker's — Judge orders UnitedHealth to hand over broad discovery in AI coverage denial case