How AI denies your insurance claim in seconds

You would expect a denied medical claim to mean that someone looked at your case and decided against it. A doctor, or at least a trained reviewer, weighing your records. For a large volume of claims, the allegation in two active lawsuits is that nobody looked at all.

The tell is the clock

In the class action against Cigna, the allegation is that an algorithm called PXDX was used to batch-deny around 300,000 claims in a two month sample, at an average of about 1.2 seconds per claim, with physicians signing off on the denials without opening the patient records. A human cannot read a medical file in 1.2 seconds. The speed is not a sign of efficiency. It is the evidence that the review did not happen.

That case, Kisting-Leung v. Cigna, is live. A court ruling let parts of it proceed. Everything described in it is an allegation until the court resolves it, and the case file on this site says so on the page.

From triage tool to hard cap

The second case concerns UnitedHealthcare and a tool called nH Predict. The allegation is that a model built to predict how much post-acute care a patient would need was used instead as a hard cap on that care for Medicare Advantage patients, cutting people off when the model said so. The detail that makes the allegation land is the appeal rate: a large share of the denials that patients appealed were later overturned. A denial that is reversed most of the time it is challenged is a denial that was wrong most of the time it was made.

A model that predicts is a tool. The same model used to refuse is a policy. The gap between those two sentences is where the harm is alleged to live.

Why automation makes it worse, not neutral

The usual defence of an algorithm is that it removes human bias and applies the rules evenly. The problem is that evenness cuts both ways. If the rule is wrong, a human reviewer catches some of the harm through judgement, tiredness, mercy, or simply reading the file. An automated system applies the wrong rule to everyone, at full speed, without the friction that used to let some cases through.

It also shifts the burden. When a denial is instant and free to produce, but an appeal is slow and costly for the patient, the system runs on the maths that most people will not appeal. The ones least able to navigate an appeal, the elderly and the very ill, absorb the most harm. That is the pattern both case files describe.

What to take from it

Ask who signed the decision and whether they saw your file. In both cases the accountability question is the same one: a named person put their name to a refusal that an algorithm produced. The law is now testing whether that is a review at all. Until those courts rule, treat every claim here as what it is, a serious allegation on the public record, sourced in the case files linked below.