When Algorithms Decide Which Patients Deserve Care -By Fransiscus Nanga Roka

Usually, recovery time can be approximated by an algorithm. It cannot evaluate an anxious patient, make a judgement about an unexpected complication or take moral responsibility for causing irreversible injury through withdrawal of care. But a federal lawsuit contends that UnitedHealth Group permitted just such a scheme to influence when elderly patients received “enough” rehabilitation.

The case is Estate of Gene B. Lokken et al. v. UnitedHealth Group Inc. et al., a putative class action by the estates of two recently deceased Medicare Advantage members ripped on Nov. 14, 2023 in Minnesota federal court Who is accused? UnitedHealth, UnitedHealthcare and naviHealth. What allegedly happened? NH Predict, an algorithm trained on millions of patient records, was used by the defendants to predict how long patients should remain in skilled nursing (SNF) or post-acute rehabilitation and then purportedly utilized those predictions as coverage limits. Where and when? The dispute is now being tried in the US District Court for Minnesota, spanning UnitedHealthcare’s Medicare Advantage operations.

So what was the alleged system? The expected discharge date from generalized historical data was generated by nH Predict. The complaint states that case managers were pressured to adhere to that prediction even when treating physicians deemed continued care medically necessary. The supposed outcome was a benchmarking calculation done via machine implicating it may substitute individual clinical judgment.

Why would an insurer do this? In their telling, the answer is brutally simple: cost containment. They observe that about 90% of overturned denials were challenged, but a very small percentage of patients appealed. That number is an accusation; not a conclusive court ruling. UnitedHealth has rejected the idea that nH Predict makes coverage decisions. But the business rationale is even more chilling: provided it can survive the claim that it has called a sick, elderly patient a liar, something they are often too frail, disoriented or unwell to take on.

It gets beyond speculation at this stage of the litigation Significant contract and bad-faith claims were allowed to proceed by U.S. District Judge John Tunheim in February 2025. Magistrate Judge Shannon Elkins granted most of the plaintiffs’ motion to compel on March 9, 2026, allowing inquiry into how UnitedHealth internally utilized, evaluated and managed nH Predict. The order does not impose liability, but it acknowledges the algorithm’s internal workings for what they are: relevant evidence, not some sort of inviolable trade secret.

This scandal reveals a regulatory scam disguised as the trendy buzzword “human in the loop.” If employees are formally rated, pressured, or punished for overruling the machine (as they likely would be), then all a nominal human reviewer means is nothing. Human oversight must be independent, competent and accountable.

The Centers for Medicare & Medicaid Services and Congress should ban predictive algorithm based coverage decisions; require that named physicians review every denial; specifically disclose decisive factors, error rates and validation populations as part of the justification of any adverse determination: previously imposed limits on automatic appeals such as enabling forced in-house peer review to belabor their decision must be lifted: and finally guarantee rapid, automatic external review before post-acute care ends. Decisions also need to be justified based on persisting algorithm logs, override histories and internal communications for litigation & regulatory audits.

Free advocates and plain-language explanations must also be provided to patients, as well as continuing coverage while urgent appeals are pending. Courts should allow for punitive damages in cases where evidence shows that an insurer knowingly took advantage of low appeal rates.

AI may assist medical judgment. Digital contract enforcement must never turn into a digital executioner. If code decides who gets care and secrecy guards the code, efficiency plus innovation equals waste automated abandonment.

Fransiscus Nanga Roka

Faculty of Law University 17 August 1945 Surabaya and Managing Partner Law Firm Victorious Indonesia