Automated Insurance Claim Denials: Evidence, Regulatory Response, and Patient Rights
Automated Insurance Claim Denials: Evidence, Regulatory Response, and Patient Rights
Abstract
Insurance companies have increasingly deployed algorithmic systems to evaluate and deny coverage requests, often without individualized clinical review. Investigative reporting and federal oversight findings have documented that at least two major insurers used AI models to deny tens of thousands of claims at rates inconsistent with individual patient circumstances. The U.S. Senate Finance Committee, the Department of Health and Human Services Office of Inspector General, and CMS have all identified AI-driven denials as a patient safety concern. Regulatory responses are underway, but currently incomplete. Patients whose claims were denied by automated systems retain full appeal rights, and external review processes that involve independent clinical reviewers have high patient success rates.
Background
Prior authorization, the process by which insurers require advance approval before covering certain treatments, has long involved human clinical reviewers. Over the past decade, a significant share of that review function has migrated to algorithmic systems developed by third-party utilization management vendors and, in some cases, built internally by the insurers themselves.
The vendors that dominate this space include eviCore (now owned by Evernorth, a Cigna subsidiary), Carelon (formerly AIM Specialty Health, acquired by Elevance Health), and Cohere Health. These systems process prior authorization requests by comparing submitted documentation against a database of coverage criteria and returning approval or denial decisions, sometimes in seconds.¹
The efficiency argument for these systems is straightforward: prior authorization volumes have grown substantially, and algorithmic processing reduces per-decision cost. The clinical concern is equally straightforward: an algorithm scoring documentation against a checklist cannot assess the individual clinical circumstances of a patient the way a reviewing physician can, and the design incentives for these systems may not align with patient access to care.
The extent to which automated denials are generating clinically inappropriate outcomes is now the subject of active federal investigation, litigation, and legislation.
What the Evidence Shows
The ProPublica Cigna Investigation
In March 2023, ProPublica published an investigation into how Cigna handled prior authorization decisions using an internal system called PXDX (procedure-to-diagnosis).² The investigation, based on a review of internal documents and interviews with former Cigna employees, found that Cigna's medical reviewers were using the system to mass-reject claims at a pace that made individual case review functionally impossible. According to the ProPublica report, Cigna physicians rejected 300,000 prior authorization requests in a two-month period, spending an average of 1.2 seconds per case.²
The PXDX system automatically compared diagnosis codes against procedure codes and flagged claims that did not match a pre-defined approved combination. When a mismatch was flagged, it generated a denial recommendation that a physician was expected to review. The investigation documented that "review" frequently meant confirming the algorithmic determination without examining patient-specific medical records.²
Cigna disputed some characterizations in the ProPublica report but did not contest the core volume and timing data.
The UnitedHealth nH Predict Litigation
ProPublica and the New York Times both reported in 2023 on litigation and regulatory concerns surrounding UnitedHealth Group's use of an AI tool called nH Predict (developed by NaviHealth, a UnitedHealth subsidiary) to make coverage decisions for post-acute care, specifically skilled nursing facility (SNF) and rehabilitation coverage for Medicare Advantage patients.³
Court filings and investigative reporting described a system in which nH Predict generated discharge timelines for SNF patients, and those timelines were used to deny coverage extensions even when attending physicians stated that patients required continued care. According to reporting by STAT News and others, the algorithm generated denial recommendations at a significantly higher rate than was consistent with individual patient circumstances, and human reviewers at UnitedHealth were following the algorithm's output without independent clinical assessment.³
Multiple class action lawsuits were filed against UnitedHealth alleging that the use of nH Predict to deny Medicare Advantage post-acute care claims violated federal law and harmed patients. As of early 2026, at least one case was proceeding through federal court.
Senate Finance Committee Findings
In 2023, the Senate Finance Committee released findings from an investigation into the use of AI systems in Medicare Advantage coverage determinations.⁴ The committee reviewed internal data from several major Medicare Advantage insurers and found:
The committee findings built on earlier OIG work. A 2022 OIG report found that 18 percent of prior authorization denials in Medicare Advantage plans that the OIG sampled would have been approved under traditional Medicare coverage criteria, suggesting a systematic pattern of inappropriate denial.⁵
CMS 2024 Medicare Advantage Rule on AI Denials
In April 2023, CMS finalized the CY2024 Medicare Advantage and Part D rule, which included specific provisions addressing the use of algorithms and AI in coverage decisions.⁶ The rule codified that Medicare Advantage plans must make coverage determinations based on individual patient circumstances rather than solely on statistical models or population-level data.
CMS stated explicitly: "Coverage criteria used by MA plans, including any prior authorization requirements or utilization management tools, must be based on current clinical criteria and individual patient circumstances. Use of an algorithm or AI model that does not account for individual patient needs does not satisfy this requirement."⁶
The rule did not prohibit the use of AI in coverage decisions but established that algorithmic outputs cannot function as a substitute for individualized clinical review. As of the 2024 plan year, Medicare Advantage plans were required to ensure that coverage decisions could not be made based solely on an AI determination without human clinical review.⁶
Academic Research on Automated Denial Systems
Academic literature on automated utilization management has grown alongside industry adoption. A 2020 study in the Journal of the American Medical Informatics Association examined the characteristics of automated prior authorization systems and found that systems optimized for denial consistency tended to produce higher denial rates for edge cases, patients with comorbidities, and diagnoses that fell outside common presentation patterns.⁷ The study noted that algorithmic consistency is not equivalent to clinical accuracy.
A 2022 analysis in Health Affairs examined the relationship between electronic prior authorization implementation and denial rates across commercial insurers. The study found that insurers who implemented electronic prior authorization systems at scale experienced measurably higher denial rates for specialty medications compared to a matched comparison period, even after controlling for changes in formulary and coverage policy.⁸
A 2023 study in JAMA Network Open examined the characteristics of claims denied by utilization management vendors (often functioning through automated review) versus claims reviewed by in-house insurer staff. The study found that third-party utilization management vendor denials had higher rates of reversal on external review, consistent with the hypothesis that automated screening processes are less clinically accurate than individualized review.⁹
California's SB 1120 Response
California enacted SB 1120 in October 2024, requiring that coverage decisions involving AI or algorithmic tools be reviewed by a licensed physician or other appropriate clinician before becoming final.¹⁰ The law applies to health plans regulated by the California Department of Managed Health Care and represents one of the first state-level statutory requirements for physician oversight of AI-driven coverage decisions.¹⁰
Other states are developing similar legislation, and federal proposals from the Senate Finance Committee and House Energy and Commerce Committee have been introduced but not yet enacted as of early 2026.
What This Means for Patients
If your claim was denied quickly, the speed of the denial may itself be informative. A denial issued within hours of a prior authorization submission is unlikely to reflect individualized clinical review of your specific circumstances.
This matters for your appeal strategy. When an appeal or external review is conducted by a human clinical reviewer, they are examining your case with attention to its particulars. The evidence consistently shows that when algorithmic denials are subjected to human clinical review through appeal and external review processes, the denial is reversed at high rates.
You are not appealing the judgment of a clinician who reviewed your records. In many cases, you are correcting the output of a system that compared codes against a checklist. That is a different kind of challenge, and it is one where the evidence is strongly in favor of the person with the actual clinical history.
The CMS 2024 rule means that Medicare Advantage patients now have a documented regulatory basis for challenging any denial where the plan cannot demonstrate that individual clinical circumstances were considered. That documentation standard is something an appeal can directly invoke.
What You Can Do
Ellen can help you draft your appeal using the same evidence base cited here. Start here
Frequently Asked Questions
How do I know if AI was used to deny my claim?
You can ask your insurer directly. Under the CMS 2024 Medicare Advantage rule, plans must ensure that coverage decisions reflect individual patient circumstances, which means they are accountable for disclosing the basis of the decision. Some state laws, including California's SB 1120, require insurers to disclose when an automated decision-making system was used. You can also request the clinical criteria document used to evaluate your request and ask whether the review was conducted by a licensed clinician.
Can I appeal an AI-generated denial?
Yes. All insured individuals have the right to an internal appeal regardless of how the original denial was generated. Most also have the right to independent external review. The external review process removes the insurer's systems entirely from the decision: an independent medical reviewer examines the clinical merits of your specific case. Federal data shows this process results in patient victories in a substantial proportion of cases.
What is a peer-to-peer review?
A peer-to-peer review is a direct phone consultation between your prescribing physician and a medical officer at the insurer or utilization management vendor. Your physician presents the clinical rationale for the prescribed treatment in real time. Peer-to-peer reviews bypass algorithmic processing and frequently result in denial reversals before a formal appeal is necessary. Your physician can request this process, and it is one of the most efficient options available when a denial appears to be based on documentation deficiency rather than a genuine clinical dispute.