Between March 2023 and February 2025, a public record took shape around a question that had mostly been discussed in the abstract: what happens when a health plan's claim review runs on software. The record is short — an investigation, a class-action complaint, a federal guidance memo, a state statute, a physician survey — and parts of it are unfinished. But every entry is a document anyone can read. This piece collects the record, states what each document says, and goes no further. Allegations are labeled as allegations. Where a company answered publicly, its answer is quoted.
“an algorithm that determines coverage based on a larger data set instead of the individual patient's medical history, the physician's recommendations, or clinical notes would not be compliant with § 422.101(c)”4
CMS, memo to all Medicare Advantage organizations and Medicare-Medicaid plans, February 6, 2024
The record
Five public documents · 2023–2025
Investigation
ProPublica and The Capitol Forum publish the Cigna PXDX investigation
The reporting describes an internal Cigna review system, PXDX, that “allows its doctors to instantly reject a claim on medical grounds without opening the patient file.” Citing corporate documents, it states: “Over a period of two months last year, Cigna doctors denied over 300,000 requests for payments using this method, spending an average of 1.2 seconds on each case.”1
Cigna called the reporting “biased and incomplete” and said the process was created to “accelerate payment of claims for certain routine screenings.”1
Complaint
Estate of Gene B. Lokken et al. v. UnitedHealth Group Inc. et al. is filed
Estates of Medicare Advantage enrollees file a proposed class action in the U.S. District Court for the District of Minnesota (No. 0:23-cv-03514) against UnitedHealth Group and its subsidiary NaviHealth.3The complaint alleges the companies used a model called nH Predict in determining post-acute care coverage, and alleges the model has “a 90% error rate” — a figure the complaint calculates from the share of payment denials reversed through internal appeals or by administrative law judges. It further alleges that only about 0.2% of policyholders appeal denied claims.2
STAT News, reporting the filing, quoted UnitedHealth's response: the tool “is used as a guide,” coverage decisions “are based on CMS coverage criteria,” and the suit “has no merit.” 2 These are allegations. The court has ruled on which claims may proceed — most recently in a February 13, 2025 opinion on a second motion to dismiss — not on whether they are true; class-certification filings run into late 2026.3
Case documents · Georgetown Health Care Litigation Tracker ↗
Regulator
CMS issues its FAQ memo on algorithms under the CY2024 Medicare Advantage rule
Interpreting final rule CMS-4201-F (in effect since January 1, 2024), CMS states that a tool “can be used to assist MA plans in making coverage determinations,” but that “for inpatient admissions, algorithms or artificial intelligence alone cannot be used as the basis to deny admission or downgrade to an observation stay; the patient's individual circumstances must be considered.” 4
On post-acute care, the memo provides that a predicted length of stay “alone must not be used as the basis to terminate post-acute care services” — the patient's eligibility “can only be determined by re-assessing the individual patient's condition” before a termination notice issues.4
Statute
California enacts SB 1120, the Physicians Make Decisions Act
Chaptered as Chapter 879, the statute amends Health and Safety Code §1367.01 and Insurance Code §10123.135. For California-regulated plans and insurers using AI in utilization review, it provides that the tool “shall not deny, delay, or modify health care services based, in whole or in part, on medical necessity,” and that a medical-necessity determination “shall be made only by a licensed physician or a licensed health care professional competent to evaluate the specific clinical issues.”5
The tool's determinations must rest on the enrollee's medical history and “individual clinical circumstances as presented by the requesting provider,” and may not be based “solely on a group dataset.” 5
Survey
The AMA releases physician survey results on AI in prior authorization
The American Medical Association reports that “three in five physicians (61%) are concerned that health plans' use of AI is increasing prior authorization denials,” and that nearly half of physicians (49%) ranked oversight of payers' use of AI in medical-necessity determinations among the top three priorities for regulatory action.6
Reading the record
Three kinds of document sit in that timeline, and they carry different weight. The statute and the regulator's memo are operative law: SB 1120 binds plans and insurers regulated by California; the CMS memo states how existing Medicare Advantage regulation applies. The investigation rests on corporate documents and interviews with former employees, and the company disputes its characterization. The complaint is the weakest form of record — allegations drafted by one side, still untested; nothing in the docket so far establishes what nH Predict did or did not do.
The record also does not establish why any of these systems was built, and this piece does not guess. What the documents establish is narrower: the specific conduct two regulators have now put in writing as impermissible, the specific figures two companies were asked about in public, and what each company said in reply.
The standard that is missing
Read together, the two legal documents converge on one requirement from different directions: a medical-necessity determination must engage the individual case. CMS states it as a data requirement — a decision based on “a larger data set instead of the individual patient's medical history” is non-compliant.4California states it as a personnel requirement — only a licensed physician or a licensed health care professional “competent to evaluate the specific clinical issues” may make the determination.5 Both answer the questions of who decides and on what.
Neither fully reaches the adjacent question: how a decision shows its work. Fragments exist. CMS requires Medicare Advantage internal coverage criteria to be publicly accessible — the memo specifies they “must be accessible via a website and cannot be behind a paywall or require a subscription for access” — and, on post-acute terminations, requires “a specific and detailed explanation” including “a description of the applicable coverage criteria and rules.”4 But no document in this record states that standard generally.
Verifiable citation of coverage criteria
In the record
CMS requires MA plans to make internal coverage criteria publicly accessible, not behind a paywall; on post-acute terminations, the plan must supply a specific and detailed explanation including a description of the applicable coverage criteria (Feb. 2024 memo).
Not in the record
No document here requires an individual adverse determination — wherever automation assisted it — to cite the specific criterion it applied, in a form the receiving clinician can check against the published text.
Qualified clinical review
In the record
SB 1120 reserves medical-necessity determinations to a licensed physician or a licensed professional competent in the specific clinical issues (California-regulated coverage); the CMS memo requires individual re-assessment before an MA termination.
Not in the record
Within the documents collected here, no equivalent named-reviewer requirement covers commercial coverage outside California, including self-funded employer plans.
That is the state of the record as of this writing: automation may assist; the individual case must decide; and the two mechanisms that would let anyone verify the second condition — a named criterion and a qualified reviewer — appear only where the newest documents happen to reach. The next entries in this record will be written by courts, legislatures, and regulators. The documents above are where they will start.
Sources
- 1ProPublica & The Capitol Forum — How Cigna Saves Millions by Having Its Doctors Reject Claims Without Reading Them (PXDX; 300,000 denials in two months; 1.2 seconds per case; Cigna's response) · Mar 25, 2023
- 2STAT News (Ross & Herman) — UnitedHealth faces class action lawsuit over algorithmic care denials in Medicare Advantage plans (complaint's alleged 90% error rate and 0.2% appeal figure; UnitedHealth's response) · Nov 14, 2023
- 3Georgetown Health Care Litigation Tracker — Estate of Gene B. Lokken et al. v. UnitedHealth Group Inc. et al., No. 0:23-cv-03514 (D. Minn.): filings and procedural history · filed Nov 14, 2023; tracked through 2026
- 4CMS — Frequently Asked Questions related to Coverage Criteria and Utilization Management Requirements in CMS Final Rule (CMS-4201-F), memo to all MA organizations (PDF) · Feb 6, 2024
- 5California Legislature — SB 1120 (2023–2024), Chapter 879: amendments to Health & Safety Code §1367.01 and Insurance Code §10123.135 · chaptered Sep 28, 2024
- 6American Medical Association — Physicians Concerned AI Increases Prior Authorization Denials, Worsens Patient Harms & Waste (61% concerned; 49% rank AI oversight a top-three regulatory priority) · Feb 24, 2025