Class Action Lawsuit Alleges UnitedHealth Uses AI with 90% Error Rate to Deny Medicare Advantage Claims, Causing Patient Deaths

On November 14, 2023, the families of two deceased Medicare Advantage beneficiaries filed a federal class action lawsuit alleging that UnitedHealth Group knowingly uses a faulty artificial intelligence algorithm with a 90% error rate to systematically deny elderly patients coverage for medically necessary post-acute care. The lawsuit claims that UnitedHealth’s nH Predict AI tool overrides physicians’ medical judgments to cut off coverage prematurely, leading to patient deaths and deteriorating health. When patients appeal these AI-driven denials, they win over 90% of the time—proving the denials were medically unjustified—but the arduous appeals process results in many patients dying or suffering permanent harm before coverage is restored.

The nH Predict AI Algorithm

UnitedHealth subsidiary NaviHealth, acquired in 2020, developed the nH Predict algorithm to predict how much post-acute care Medicare Advantage patients “should” require following hospitalizations. The system:

Database: Analyzes a database of 6 million patient records compiled over years to predict care needs based on diagnosis, age, living situation, and physical function.

Automated Denials: Generates predictions for the precise moment when UnitedHealthcare should cut off payment for skilled nursing facility care, rehabilitation services, and home health care—regardless of individual patient circumstances or physician recommendations.

Overriding Medical Judgment: The AI’s predictions override treating physicians’ determinations about medical necessity, substituting an algorithm’s statistical analysis for doctors’ individualized clinical assessments.

Coverage Cutoff Timeline: The system pinpoints exactly when to terminate coverage, forcing patients to either pay out-of-pocket for continued care (often thousands of dollars per day), leave facilities before medically ready, or appeal denials while their health deteriorates.

The 90% Error Rate

The lawsuit’s most damning allegation is that UnitedHealth knows nH Predict has a 90% error rate but continues using it because the appeal process is so arduous that few patients successfully challenge denials:

Appeal Success Rate: When patients appeal AI-driven denials, they prevail more than 90% of the time through either internal appeals or federal administrative law judge rulings. This demonstrates the AI’s predictions are medically wrong in the vast majority of cases.

Systematic Wrongful Denials: A 90% error rate for denials is not a “bug” in the system—it represents UnitedHealth’s deliberate business model. The company knows most denials are medically unjustified but issues them anyway because most patients won’t successfully appeal.

Profitable Illegality: The strategy is profitable because:

• Only a small fraction of patients appeal (many lack energy, knowledge, or support to navigate the process)
• Appeals take months, during which many patients pay out-of-pocket, give up, or die
• Even when appeals succeed, UnitedHealth has already avoided paying for weeks or months of care
• Penalties for wrongful denials are minimal compared to the savings from denied claims

Lack of Human Review: The lawsuit alleges UnitedHealth issues denials based on AI predictions without meaningful human review of individual patient circumstances, medical records, or physician assessments—making the process a rubber-stamping of algorithmic determinations.

Deaths and Deteriorating Health

The lawsuit describes specific cases where AI-driven denials led to patient deaths and permanent harm:

Premature Facility Discharge: Patients receiving skilled nursing or rehabilitation care have coverage abruptly terminated based on AI predictions, forcing discharge before they can safely return home. These premature discharges lead to falls, inability to manage medications, malnutrition, and other complications.

Care Interruption: When coverage ends, patients must either immediately leave facilities or begin paying $300-500+ per day out-of-pocket. Most cannot afford this, forcing them to leave even when physicians determine continued care is medically necessary.

Appeal Process Deaths: The appeals process typically takes weeks to months. During this time, patients often deteriorate significantly, suffer permanent loss of function, or die. Even when appeals ultimately succeed, the delay causes irreversible harm.

Family Financial Devastation: Families facing AI-driven denials must choose between bankrupting themselves to continue care or accepting premature discharge knowing their loved one will likely suffer or die. This constitutes a form of financial terrorism against vulnerable populations.

Worsening Conditions: The complaint states that “plaintiffs suffered worsening injury, illness, or death as a result of the denials”—establishing causation between AI-driven coverage terminations and patient harm.

Post-Acute Care Targeting

The AI system specifically targets post-acute care—the period following hospitalization when patients need skilled nursing, rehabilitation, or home health services to recover:

High-Cost Services: Post-acute care represents some of the most expensive Medicare Advantage coverage, making it a prime target for denials to maximize insurer profits. Skilled nursing facility care costs hundreds of dollars per day, creating strong financial incentives to cut coverage as early as possible.

Vulnerable Populations: Post-acute care patients are elderly, recently hospitalized, and often cognitively or physically impaired—making them least able to navigate complex appeal processes. This targeting represents predation on the most vulnerable beneficiaries.

Physician Determinations Overridden: Treating physicians assess patients daily and make individualized determinations about when they’re ready for discharge. The AI system overrides these medical judgments with statistical predictions, substituting corporate profit motives for clinical care.

Senate Investigation Findings: A Senate investigation found that between 2019 and 2022, UnitedHealthcare’s post-acute services denial rate increased from 8.7% to 22.7%—a more than doubling that coincided with NaviHealth’s acquisition and nH Predict deployment.

Skilled Nursing Denials Ninefold Increase: UnitedHealth’s skilled nursing facility denial rate increased ninefold from 2019 to 2022. In 2019, the insurer denied 1.4% of skilled nursing admission requests; by 2022 (the first full year of NaviHealth management), denials reached 12.6%—a dramatic spike directly attributable to AI-driven denial systems.

NaviHealth Acquisition and Deployment

UnitedHealth’s acquisition of NaviHealth in 2020 and subsequent deployment of nH Predict reveals corporate strategy:

Acquiring Denial Technology: Rather than developing AI internally, UnitedHealth acquired a company specializing in claim denial systems—demonstrating that systematic denial is a core strategic priority, not an operational efficiency tool.

Rapid Deployment: Following acquisition, UnitedHealth rapidly deployed nH Predict across its Medicare Advantage plans, causing the dramatic spike in denial rates from 2019-2022. This speed suggests insufficient testing for accuracy and patient safety.

Vertical Integration: The combination of UnitedHealthcare insurance, Optum physician networks, and NaviHealth denial systems creates a vertically integrated apparatus for systematic claim denials—controlling medical records, coverage decisions, and appeals processes internally.

Profit Optimization: The timing suggests UnitedHealth acquired NaviHealth specifically to reduce post-acute care costs as Medicare Advantage enrollment grew. Rather than competing on better care, the strategy focuses on denying care to maximize profits.

UnitedHealth’s Defense: AI Is Just a “Guide”

UnitedHealth responded to the lawsuit by claiming nH Predict “is not used to make coverage decisions, but instead is a guide to help us inform providers, families and other caregivers about what sort of assistance and care the patient may need.”

This defense is contradicted by:

Pattern of Denials: The ninefold increase in denial rates coinciding with nH Predict deployment proves the tool drives coverage decisions, not merely “informs” them.

Overriding Physicians: If nH Predict were just a guide, treating physicians’ determinations would take precedence. Instead, the AI’s predictions override medical judgments, proving it controls coverage decisions.

90% Appeal Success Rate: If denials were based on legitimate medical review rather than algorithmic predictions, appeal success rates would be far lower. The 90% rate proves decisions are algorithm-driven, not medically justified.

Lack of Individualization: A true “guide” would incorporate individual patient circumstances, medical history, and physician input. The AI’s statistical predictions ignore these factors, demonstrating it dictates rather than informs coverage decisions.

Legal Claims and Federal Court Proceedings

The lawsuit asserts claims for:

Breach of Contract: Medicare Advantage plans constitute contracts to provide medically necessary care. Using AI to deny such care breaches these contractual obligations.

Breach of Implied Covenant of Good Faith and Fair Dealing: Insurance contracts include an implied covenant that insurers will fairly evaluate claims. Using an AI with known 90% error rates violates this duty.

ERISA Violations: For employer-sponsored Medicare Advantage plans, the denials may violate the Employee Retirement Income Security Act’s requirements for fair claims procedures.

In February 2025, a federal judge allowed the lawsuit to proceed, rejecting UnitedHealth’s motion to dismiss. The judge ruled that the plaintiffs’ breach of contract claims could move forward because they are not preempted by federal Medicare law—a significant victory allowing the case to advance to discovery.

Class Action Status: The lawsuit seeks class action certification to represent all Medicare Advantage beneficiaries whose post-acute care claims were denied based on nH Predict predictions. This could potentially include hundreds of thousands of patients.

Prior Authorization as Denial Strategy

The nH Predict system exemplifies how prior authorization has evolved from utilization review to systematic claim denial:

Original Purpose: Prior authorization was introduced to prevent unnecessary care by requiring insurer approval before expensive procedures. The stated goal was ensuring medical appropriateness.

Actual Function: Prior authorization now serves primarily to deny or delay medically necessary care to reduce insurer costs. The process is designed to be so burdensome that physicians and patients give up rather than completing appeals.

AI Acceleration: Automated systems like nH Predict allow insurers to dramatically increase denial rates without hiring additional medical reviewers. The AI can deny thousands of claims per day, overwhelming physicians and patients with appeals.

Physician Burden: Prior authorization requirements force physicians to spend hours seeking approval for treatments they’ve already determined are medically necessary. This burden often leads physicians to prescribe less expensive (but less effective) alternatives to avoid the authorization process.

Senate Investigation and Findings

A Senate committee investigation into Medicare Advantage insurers’ use of AI for claim denials found:

Industry-Wide Problem: UnitedHealthcare, Humana, and CVS all deployed “predictive technology” to deny post-acute care at far higher rates than other types of care, suggesting industry-wide adoption of AI denial systems.

Diminished Access: Between 2019 and 2022, the denial rate increases resulted in “diminished access to post-acute care for Medicare Advantage beneficiaries” compared to traditional Medicare, where such denials are less common.

Racial and Socioeconomic Disparities: The report suggested AI-driven denials may disproportionately affect minority and lower-income beneficiaries who have less ability to navigate appeals processes or pay out-of-pocket during appeals.

Regulatory Gaps: The Senate investigation highlighted that CMS lacks effective oversight mechanisms to detect or prevent AI-driven systematic denials, allowing insurers to deploy these systems without regulatory approval or review.

Medicare Advantage vs. Traditional Medicare

The nH Predict case exposes fundamental differences between privatized Medicare Advantage and traditional Medicare:

Profit Motive: Medicare Advantage insurers maximize profits by denying claims. Traditional Medicare operates on a non-profit basis with no financial incentive to deny medically necessary care.

Claim Denial Rates: Traditional Medicare has far lower claim denial rates than Medicare Advantage plans. Traditional Medicare’s denial rate is approximately 2-3%, compared to 15-20%+ for many Medicare Advantage plans.

Algorithmic Denials: Traditional Medicare doesn’t use AI systems to override physician judgments. Coverage decisions are based on medical necessity as determined by treating physicians, not corporate algorithms.

Appeal Success: The 90% appeal success rate for nH Predict denials demonstrates that the AI is rejecting care that would be covered under traditional Medicare—proving that privatization reduces rather than enhances benefits.

Regulatory Capture and Enforcement Failure

The nH Predict scandal reveals regulatory capture enabling AI-driven denials:

No Pre-Approval Required: UnitedHealth deployed nH Predict without seeking or receiving approval from CMS. Medicare Advantage insurers can implement any denial systems they choose without regulatory review.

Minimal Oversight: CMS lacks systems to detect when insurers use AI to systematically deny medically necessary care. The agency relies on aggregate statistics that don’t reveal AI-driven denial patterns.

Weak Penalties: Even when systematic denials are identified, CMS penalties are minimal compared to the savings from denied claims. This makes illegal denials profitable even when caught.

Appeal Process Burden: CMS requires beneficiaries to navigate complex appeals processes rather than proactively auditing denial patterns. This places the burden on the most vulnerable patients rather than on regulators to prevent abuse.

Industry Lobbying: Medicare Advantage insurers spend hundreds of millions on lobbying to prevent regulations that would require transparency about AI systems, limit denial rates, or impose meaningful penalties for wrongful denials.

AI and Algorithmic Discrimination

The nH Predict system raises broader concerns about AI in healthcare decisions:

Black Box Decisions: Patients and physicians don’t know how the AI generates its predictions, making it impossible to challenge the algorithm’s logic or identify bias.

Lack of Individualization: The AI uses population statistics to predict individual patient needs, ignoring that patients may have unique circumstances requiring longer recovery periods.

Embedded Bias: AI systems trained on historical data may embed racial, socioeconomic, and geographic biases, potentially causing discriminatory denial patterns.

Accountability Vacuum: When an AI denies coverage, who is responsible—the programmers, the corporation, the AI itself? This accountability vacuum allows systematic harm without clear liability.

Automation of Cruelty: AI systems allow corporations to deny medically necessary care at massive scale without individual humans making case-by-case decisions, automating cruelty while diffusing responsibility.

Impact on Physician-Patient Relationship

AI-driven denials fundamentally alter the physician-patient relationship:

Undermining Medical Authority: When corporate algorithms override physicians’ medical judgments, it demonstrates that insurers—not doctors—control patient care.

Physician Frustration: Doctors spend increasing time fighting denial systems rather than treating patients, leading to burnout and reducing time available for patient care.

Erosion of Trust: Patients lose trust in physicians when doctors cannot deliver recommended care due to insurance denials, damaging the therapeutic relationship.

Defensive Medicine: Physicians may avoid recommending optimal treatments knowing they’ll be denied, leading to defensive medicine that prioritizes insurer approval over patient welfare.

The Business Model: Denial as Profit

The nH Predict lawsuit exposes health insurance business model fundamentals:

Deny-Then-Pay-If-Appealed: Issue systematic denials knowing most are wrong, then pay only the small fraction that successfully appeal. This strategy maximizes profits by avoiding payments to patients who lack resources or stamina to fight.

Attrition Strategy: Make the appeals process so difficult and time-consuming that patients give up, deteriorate beyond treatability, or die before coverage is restored—eliminating the need to ever pay denied claims.

Information Asymmetry: Insurers possess far more information about denial rates, appeal outcomes, and AI accuracy than patients or regulators, enabling systematic abuse without easy detection.

Legal Impunity: ERISA preemption prevents most lawsuits for wrongful denials, and Medicare Advantage regulations impose minimal penalties, creating near-immunity for systematic denial practices.

Systematic Corruption

The nH Predict case exemplifies systematic corruption in American healthcare:

Profit from Death and Suffering: UnitedHealth’s business model depends on denying medically necessary care, causing preventable deaths and suffering to maximize corporate profits.

Regulatory Capture: CMS allows Medicare Advantage insurers to deploy AI denial systems without approval, oversight, or meaningful penalties, demonstrating that regulators protect corporate profits rather than patient welfare.

Privatization Failure: Medicare Advantage was promoted as providing better care than traditional Medicare through private sector efficiency. In reality, it extracts taxpayer dollars while providing worse care through systematic denials.

Legal System Failure: Even with evidence of 90% wrongful denial rates causing deaths, the legal system provides minimal remedies. The class action lawsuit, if successful, will likely result in a settlement that represents a tiny fraction of profits from the denial scheme.

Political Protection: Despite Senate investigations exposing AI-driven denials, Congress has passed no legislation to ban or regulate these systems, demonstrating insurance industry political power prevents reforms even when systematic patient harm is documented.

The nH Predict lawsuit proves that Medicare Advantage insurers use artificial intelligence not to improve care but to systematically deny medically necessary treatment to elderly and disabled Americans, causing preventable deaths and suffering to maximize corporate profits while regulatory capture ensures minimal consequences.