The 90% figure in the UnitedHealth lawsuit is not a proven 90% technical error rate. Plaintiffs’ amended complaint alleges that more than 90% of post-acute-care claim denials—and more than 80% of preauthorization denials—were later reversed on appeal. Those are allegations about appeal outcomes, summarized by the court, not an independent validation of nH Predict’s accuracy or a finding that the model itself was wrong 90% of the time.
What the lawsuit alleges
The putative class action, Estate of Gene B. Lokken et al. v. UnitedHealth Group, Inc. et al., was filed in the U.S. District Court for the District of Minnesota. Plaintiffs allege that UnitedHealth Group, UnitedHealthcare and naviHealth used an algorithm called nH Predict in Medicare Advantage decisions involving post-acute care.
Post-acute care is treatment after an acute illness, injury or other major medical event. The dispute includes services such as skilled-nursing facilities, inpatient rehabilitation and home health. A coverage denial can end authorized care, force a patient to pay out of pocket or require an appeal.
UnitedHealthcare denied using nH Predict to make coverage determinations. The court has treated that dispute as unresolved: whether the system was used for coverage decisions is a merits question for later stages of the case.
What the “90% error rate” actually measures
The number comes from the plaintiffs’ amended complaint, filed in 2024, and was recounted in the court’s 2025 opinion. Plaintiffs allege:
- More than 90% of claim denials were reversed on appeal.
- More than 80% of preauthorization denials were reversed on appeal.
A reversal means the denial was overturned through the appeals process. It does not, by itself, measure how often nH Predict’s underlying predictions were inaccurate across every patient or case. No independent technical benchmark, validation study or official model-accuracy statistic establishing a universal 90% error rate was identified in the cited materials.
| Figure | What it refers to | What it does not establish |
|---|---|---|
| More than 90% | Plaintiffs’ allegation that this share of claim denials was reversed on appeal | A validated 90% error rate for all nH Predict outputs |
| More than 80% | Plaintiffs’ allegation that this share of preauthorization denials was reversed on appeal | That every initial denial was caused by the algorithm |
| More than doubled | A 2024 Senate staff report’s reported increase in UHC’s post-acute claim-denial rate after naviHealth and nH Predict use began in 2019, as cited by the March 2026 court order | A judicial finding that nH Predict caused the increase |
What nH Predict is alleged to do
Descriptions reported by Ars Technica, based on the lawsuit, say nH Predict estimated a patient’s expected medical needs, length of post-acute stay and projected discharge date. The alleged method used patient attributes and comparisons with similar cases.
The reporting also described plaintiffs’ claim that the system could override treating clinicians’ recommendations and fail to account for relevant health factors. Those are allegations about the system’s operation and implementation, not confirmed technical documentation or findings after a trial.
Ars Technica relayed a statement that occupational therapist and former naviHealth case manager Amber Lynch gave to STAT: “By the end of my time at NaviHealth I realized: I’m not an advocate, I’m just a moneymaker for this company,” she told Stat. The quotation is Lynch’s characterization, not an official UnitedHealthcare statement and not a fact established by the court.
What the courts have decided—and what they have not
February 2025: partial survival of the lawsuit
On February 14, 2025, the district court dismissed five of seven state-law theories as preempted by the Medicare Act. It allowed breach-of-contract and breach-of-implied-covenant claims to continue and waived exhaustion of administrative remedies on futility grounds.
This was a ruling on motions to dismiss. The court assumed the relevant allegations for purposes of deciding which claims could proceed; it did not determine that the alleged denial practices occurred, that nH Predict made coverage decisions or that UnitedHealth is liable.
March 2026: discovery order
On March 9, 2026, Magistrate Judge Shannon G. Elkins granted in part and denied in part the plaintiffs’ motion to compel. The order required production of specified materials concerning post-acute claims policies, nH Predict’s development and use, relevant government investigations, oversight and certain employees.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →The order did not require production of the underlying data, rules, source code and medical guidelines as part of one request. Discovery rulings determine what evidence must be exchanged; they do not decide whether the allegations are true.
The order cited an October 2024 report by the U.S. Senate Permanent Subcommittee on Investigations. As the court recounted it, the report said UHC’s post-acute claim-denial rate more than doubled after naviHealth and nH Predict use began in 2019. That statistic is attributed to the Senate staff report as described by the court, not presented as an independent finding in the discovery order.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why an appeal reversal is different from a model-accuracy test
An appeal reversal reflects the final outcome of a coverage dispute after additional records, clinical review or other evidence may have been considered. A technical error rate would require a defined prediction target, a representative data set, a prespecified evaluation method and a comparison with the correct outcome. The complaint figures, as described by the court, do not provide those elements.
The figures may still be significant evidence for the plaintiffs. A high reversal rate could support an argument that initial denials were unreliable or that the review process improperly discounted clinical judgment. But the figures alone cannot show how many denials were generated by nH Predict, whether the model directly determined coverage, or what other factors influenced an appeal.
Best Value
What this means for Medicare Advantage members
The case concerns alleged post-acute-care coverage decisions, not a general finding about every UnitedHealthcare Medicare Advantage claim or every artificial-intelligence tool used in health care. Coverage disputes remain governed by the plan’s notices and Medicare Advantage appeal procedures.
- Read the denial or termination notice for the stated reason, effective date and appeal deadline.
- Ask the treating facility and clinicians for records supporting the medical-necessity argument, including progress notes and a proposed care plan.
- Follow the appeal instructions in the notice and keep copies of submissions, decisions and delivery confirmations.
- Request expedited review when delaying the decision could jeopardize the patient’s health or ability to regain function, if the plan’s rules allow it.
The lawsuit does not itself reverse an individual’s denial or guarantee coverage. A member’s rights depend on the specific notice, plan and stage of appeal.
The current bottom line on the allegation
The headline shorthand is misleading if it is read as a proven technical statistic. Plaintiffs allege that more than 90% of claim denials and more than 80% of preauthorization denials were overturned on appeal. The courts have allowed contract-related claims to proceed and ordered some discovery, but neither the 2025 opinion nor the March 2026 discovery order found that nH Predict had a 90% error rate or that UnitedHealthcare used it to make coverage determinations.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →
