The operational pivot of UnitedHealthcare into the direct management of post-acute care authorization represents a definable shift in actuarial strategy between 2023 and 2026. This transition centers on the acquisition and subsequent weaponization of NaviHealth by Optum. Optum is the health services arm of UnitedHealth Group. The acquisition date was May 2020. The purchase price was approximately 2.5 billion dollars. The integration of this entity allowed UnitedHealthcare to deploy the nH Predict algorithm. This proprietary artificial intelligence model became the primary gatekeeper for Skilled Nursing Facility (SNF) and Inpatient Rehabilitation Facility (IRF) care for Medicare Advantage beneficiaries. The data indicates a systematic replacement of physician judgment with algorithmic probability. This section analyzes the mechanics of this acquisition. It details the statistical denial rates that followed. It examines the financial incentives driving the deployment of nH Predict during the 2023 to 2026 observation period.

1. The Algorithmic Mechanism: nH Predict Implementation

The core asset obtained through the NaviHealth acquisition was the nH Predict software. This tool utilizes a database of six million patient records to generate length-of-stay (LOS) predictions. The system operates by matching a patient’s diagnosis and demographic data against historical discharge patterns. It produces a rigid target discharge date. This date is presented to case managers as a definitive metric rather than a clinical guideline. Internal documents obtained during the 2024 Senate Permanent Subcommittee on Investigations inquiry reveal that Optum management enforced strict adherence to these algorithmic targets. Case managers were required to keep patient stays within a 1 percent variance of the nH Predict projection. Staff members who authorized days beyond this algorithmic limit faced disciplinary action or termination. This performance metric effectively stripped clinical autonomy from the authorization process.

The algorithm lacks the capacity to account for non-standard clinical complications. It does not factor in secondary infections. It ignores slower-than-average physical recovery speeds. It disregards home safety unavailability. The software treats the mean regression of historical data as a prescriptive mandate for individual care. UnitedHealthcare utilized this mechanism to issue denials for post-acute care extensions. The denials were issued regardless of the treating physician’s assessment of medical necessity. The system creates a bureaucratic loop where the computer model overrides the human doctor. The doctor must then appeal to a medical director who is often employed by the insurer and bound by the same utilization management targets.

2. Statistical Denial Metrics: 2023-2025 Analysis

The integration of NaviHealth correlated with a statistically significant increase in denial rates for post-acute care. Data submitted to the Senate Subcommittee indicates that UnitedHealthcare’s denial rate for post-acute care prior authorization requests rose from 10.9 percent in 2020 to 22.7 percent by the end of 2022. The trend line continued upward through 2023 and 2024. Skilled Nursing Facility denial rates specifically saw a nine-fold increase over the reporting period. This escalation is not attributable to a decline in the medical acuity of the beneficiary population. It is directly attributable to the implementation of the nH Predict automated review process.

The most damning metric regarding the accuracy of nH Predict is the overturn rate. Federal data and court filings from the Estate of Gene B. Lokken v. UnitedHealth Group class action highlight a 90 percent error rate. This figure represents the percentage of denials that are reversed when a beneficiary pursues a full appeal through the federal Administrative Law Judge process. A 90 percent reversal rate implies that the initial algorithmic determination is factually incorrect or legally indefensible in nine out of ten cases. However. UnitedHealthcare relies on the "frictional cost" of the appeal process to maintain profitability. Only 0.2 percent of beneficiaries initiate an appeal after a denial. The remaining 99.8 percent either pay out-of-pocket. They deplete their life savings. Or they forgo necessary care and return home prematurely. The disparity between the 90 percent overturn rate and the 0.2 percent appeal rate constitutes the profit margin of the nH Predict system.

3. Financial Incentives and Revenue Attribution

The financial logic behind the NaviHealth acquisition is visible in the earnings reports of Optum Health. Optum acts as a service provider to UnitedHealthcare. It retains a portion of the savings generated through utilization management. This vertical integration allows UnitedHealth Group to capture revenue on both the insurance premium side and the care management side. Optum Health reported revenues of 92.4 billion dollars in the third quarter of 2023. Full-year revenues for 2023 reached 226.6 billion dollars. This represents a double-digit percentage growth year-over-year. The ability to control post-acute spend is a primary driver of this revenue expansion. Post-acute care is one of the most expensive components of the Medicare Advantage cost structure. Reducing the average length of stay by even one day across millions of members generates hundreds of millions in retained earnings.

The "medical loss ratio" (MLR) is the percentage of premium dollars spent on clinical care. Insurers strive to keep this ratio low to maximize operating margin. The nH Predict tool is a mechanism for MLR suppression. Internal NaviHealth presentations described the tool as a way to "manage" medical costs. The documents explicitly linked the reduction of Skilled Nursing Facility days to financial targets. UnitedHealthcare executives utilized these "savings" to offset rising outpatient costs in other sectors. The data demonstrates that the algorithm was not deployed to improve patient outcomes. It was deployed to mathematically optimize the spread between premiums collected and claims paid. The 2024 Senate report concluded that the insurers used these tools to deny care "simply to make more money." This finding contradicts the company's public assertions that the tool is merely a guide for clinical decision-making.

4. Operational Directives and Staff Coercion

The investigative record reveals a culture of coercion within the NaviHealth division. Employees were not merely encouraged to use the software. They were compelled to obey it. Witness testimony and internal memos show that case managers were instructed to ignore provider notes that contradicted the nH Predict date. One specific directive instructed staff: "Do NOT guide providers." This command was intended to prevent case managers from helping physicians frame their authorization requests in a way that might trigger an approval. The goal was to minimize the "authorization rate" and maximize the "termination rate."

Performance reviews for NaviHealth employees were tied to their adherence to the algorithm's predicted length of stay. A deviation of more than 1 percent from the predicted discharge date was flagged as a performance failure. This created a direct conflict of interest. The case manager's employment security depended on denying care. The patient's health depended on extending care. The statistical weight of the employment incentive virtually ensured that the algorithm would prevail. Reports from 2023 and 2024 indicate that many experienced clinicians left the organization due to ethical concerns. They were replaced by staff who were more compliant with the data-driven directives. This personnel shift further solidified the dominance of the nH Predict model in the authorization workflow.

5. Legal Challenges and Regulatory Fallout (2023-2026)

The aggressive application of nH Predict triggered a wave of legal and regulatory actions between 2023 and 2026. The most prominent litigation is the class action lawsuit Estate of Gene B. Lokken et al. v. UnitedHealth Group. Filed in November 2023 in the District of Minnesota. This case alleges that the use of the algorithm constitutes a breach of contract and a breach of fiduciary duty. The plaintiffs argue that the 90 percent error rate proves the system is defectively designed. They contend that UnitedHealthcare knows the model is flawed but continues to use it because it is profitable. The court denied UnitedHealthcare's motion to dismiss the case in early 2025. This ruling allowed the discovery phase to proceed. The discovery process is expected to expose further internal datasets regarding the algorithm's programming and error rates.

Regulatory scrutiny intensified following the release of the Senate Permanent Subcommittee on Investigations report in October 2024. The report titled "Refusal of Recovery" provided the first public confirmation of the denial rate increases. It forced the Centers for Medicare and Medicaid Services (CMS) to accelerate the enforcement of new rules. CMS Final Rule 0057-F was finalized in January 2024. It mandates that by 2026 insurers must provide specific reasons for denials. It requires them to include the data used to make the determination. It also prohibits the use of algorithms to override medical necessity without a human review. However. The retroactive nature of the harm remains unaddressed. The patients denied care between 2020 and 2024 have no regulatory recourse outside of the class action litigation. The timeframe of 2025 to 2026 represents a period of frictional adjustment. UnitedHealthcare attempts to recalibrate its denial engine to comply with the letter of the new law while maintaining the cost-saving efficacy of the algorithm.

6. The Human Impact of Algorithmic Denial

The data points translate into tangible suffering for the elderly population. The Lokken estate represents a typical case. Gene Lokken was an elderly beneficiary who suffered a fall and required skilled nursing care. His physician determined he needed continued inpatient therapy to regain the ability to walk. The nH Predict algorithm determined he should be discharged. UnitedHealthcare issued a denial based on the algorithm. Lokken was forced to pay out-of-pocket to stay in the facility. His family drained their savings. He died shortly after the denial was issued. The "savings" generated by this denial contributed to the quarterly earnings of UnitedHealth Group. This pattern is repeated across tens of thousands of cases. Patients with strokes. Patients with hip fractures. Patients with congestive heart failure. All are subjected to the same mathematical probability curve. The algorithm does not see a patient. It sees a row of data in a regression model. It sees a cost center that must be minimized.

7. Integration with Other Optum Verticals

The NaviHealth acquisition must be viewed as part of a larger ecosystem of control. Optum does not just manage the authorization. It also owns the providers. Optum is the largest employer of physicians in the United States. It owns surgical centers. It owns urgent care clinics. It owns pharmacy benefit managers. The integration of NaviHealth allows Optum to control the entire patient journey. When a patient is discharged from a hospital. Optum’s nH Predict tool determines where they go. Optum’s physicians provide the care. Optum’s pharmacy supplies the drugs. This vertical integration creates a closed loop of revenue. The denial of post-acute care often pushes the patient back into the home. There. They may be serviced by Optum’s home health division. This shifts the revenue stream from a third-party nursing home to an internal Optum division. The denial is not just a cost-saving measure. It is a revenue-redirection strategy. The 2023 to 2026 period solidified this closed-loop model. It made UnitedHealth Group an inescapable force in the delivery and financing of American healthcare.

8. Future Outlook: The AI Governance Battle

The conflict over nH Predict sets the precedent for the use of artificial intelligence in healthcare administration. The industry is watching the Lokken case closely. A judgment against UnitedHealthcare would establish a legal liability for algorithmic error. It would force insurers to prove the clinical validity of their models before deploying them. Conversely. A victory for UnitedHealthcare would validate the "black box" model of utilization management. It would signal to the market that 90 percent error rates are an acceptable cost of doing business. The years 2025 and 2026 will be defined by this legal battle. The outcome will determine whether medical necessity remains a clinical determination or becomes a derived variable in a corporate financial model. The data currently suggests the latter is the operating reality. The metrics of denial. The volume of profits. The silence of the regulatory apparatus. All point to the continued ascendancy of the algorithm over the physician.

The digitization of medical necessity has produced a statistical anomaly that defines the current era of managed care. Between 2023 and 2026 UnitedHealthcare aggressively integrated an artificial intelligence model known as nH Predict into its claims adjudication process. This tool is not merely a passive calculator. It effectively functions as a decision-making proxy that overrides the clinical judgment of treating physicians. The data suggests a systematic strategy to curtail post-acute care for elderly Medicare Advantage beneficiaries. This investigation isolates the mechanics of the algorithm and the verified metrics of its failure rates alongside the legal and regulatory fallout that continues to unfold.

1. The Mechanism: How nH Predict Generates Denials

The core of this controversy lies in a proprietary algorithm developed by NaviHealth. UnitedHealth Group acquired this subsidiary in 2020 and folded it into its Optum division. The nH Predict tool does not examine a patient. It does not interview caregivers or review daily nursing notes for nuance. Instead it utilizes a database of six million past patient records to generate a "length of stay" prediction. The model uses regression analysis to compare a specific patient's diagnosis codes and age against this historical aggregate. The output is a rigid discharge date that UnitedHealthcare medical reviewers use as a target.

The fatal flaw in this methodology is the application of aggregate averages to individual pathology. A standard recovery curve for a hip fracture does not account for a specific patient’s comorbidities such as diabetes or cognitive decline. Yet nH Predict treats these variances as statistical noise rather than clinical realities. When the algorithm predicts a 14-day recovery the carrier often issues a denial for coverage on day 15 regardless of the patient's actual ability to walk or bathe. This creates a friction point where the financial imperative of the insurer directly contradicts the medical necessity determined by the facility's doctors.

UnitedHealthcare has consistently argued that nH Predict is merely a "guide" for its medical directors. However whistleblower testimony and internal documents cited in federal court filings indicate otherwise. Staffers were reportedly pressured to adhere to the algorithm’s predicted discharge dates within a variance of 1 percent. This rigorous adherence turns a predictive suggestion into a hard limit. The result is a system where a mathematical model grounded in historical billing data dictates the future of patient care.

2. The Metric: A 90 Percent Error Rate

The most damning statistic to emerge from the legal discovery process regarding nH Predict is its overturn rate. According to the class action complaint filed in the Estate of Gene B. Lokken et al. v. UnitedHealth Group lawsuit the algorithm’s denials are overturned more than 90 percent of the time when appealed. This figure suggests a tool that is functionally defective for the purpose of accurate medical necessity determination. A 90 percent failure rate in any other industry would trigger an immediate recall. In the context of Medicare Advantage it functions as a barrier to access.

This high error rate is profitable because of a second critical metric. Only approximately 0.2 percent of patients appeal their denials. The vast majority of beneficiaries are elderly and recovering from major medical events like strokes or falls. They lack the energy and legal knowledge to fight a massive insurance bureaucracy. When a denial letter arrives stating that care is no longer medically necessary most families assume the decision is final. They either pay out of pocket or remove their loved one from the facility prematurely. The discrepancy between the 90 percent error rate and the 0.2 percent appeal rate generates a massive financial windfall for the insurer. It effectively monetizes patient fatigue.

3. Case Study: The Estates of Gene B. Lokken and Dale Henry Tetzloff

The human cost of these algorithmic denials is documented in the lead plaintiffs of the primary class action lawsuit. Gene B. Lokken broke his leg and required skilled nursing care to recover his mobility. Despite his physician's assessment that he needed continued inpatient assistance the nH Predict algorithm determined his length of stay had exceeded the average for his condition. UnitedHealthcare issued a denial. His family was forced to pay out of pocket to keep him in the facility. He died while his estate was still fighting for coverage.

Dale Henry Tetzloff suffered a severe stroke. His recovery was slow and required intensive therapy. The algorithm set a discharge target that did not align with his neurological progress. When the insurer cut off payment his family faced a choice between unsafe discharge or financial ruin. They spent approximately $70,000 of their savings to maintain his care in the facility after the denial. Tetzloff also passed away before the resolution of his claims. These cases illustrate the "infinity loop" of denials described by legal counsel. Even when a family wins an appeal the algorithm often generates a new denial days later. This forces the appeals process to restart from zero.

The Lokken lawsuit alleges breach of contract and breach of the implied covenant of good faith. In February 2025 United States District Judge John Tunheim allowed these core claims to proceed. The court recognized that the exhaustion of administrative remedies could be deemed futile given the systemic nature of the denials. This ruling marked a significant piercing of the corporate veil. It established that the use of a tool with a known 90 percent error rate could constitute a violation of the insurer's contractual obligation to provide medically necessary care.

4. The Senate Investigation and the Doubling Denial Rate

The United States Senate Permanent Subcommittee on Investigations (PSI) launched a probe into these practices that culminated in a scathing report released in October 2024. The subcommittee subpoenaed over 280,000 pages of documents from the major Medicare Advantage insurers. The data revealed a sharp escalation in denial rates following the integration of AI tools. For UnitedHealthcare specifically the denial rate for post-acute care requests more than doubled between 2020 and 2022. It rose from 10.9 percent to 22.7 percent. This increase coincided directly with the expanded deployment of NaviHealth's nH Predict.

Senator Richard Blumenthal characterized these findings as evidence that insurers were automating the denial of care to boost profits. The report highlighted internal strategies where "length of stay" management was treated as a financial lever. By reducing the number of days a patient spent in a skilled nursing facility the insurer could retain a larger portion of the capitated payments received from the federal government. The Senate's findings corroborate the whistleblower allegations that the goal was not clinical accuracy but duration reduction.

The report also detailed how the "prior authorization" process had mutated. It was no longer a check against fraud. It had become a primary method of utilization management. The data showed that skilled nursing facilities were disproportionately targeted. This sector serves the most frail demographic of the Medicare population. The automated nature of the reviews allowed the insurer to scale up denials without a corresponding increase in clinical staff workload. This efficiency came at the direct expense of patient access.

5. The Enforcers: Medical Directors and Whistleblowers

The implementation of nH Predict required human compliance. The algorithm provided the date but a medical director had to sign the denial. Whistleblower accounts have exposed the pressure applied to these employees. Amber Lynch is a former NaviHealth case manager. She told investigative reporters that she realized she was "just a moneymaker" for the company. Her role shifted from patient advocacy to data enforcement. Employees were tracked on their adherence to the predicted discharge dates. Those who authorized stays longer than the algorithm’s suggestion were labeled as outliers.

This internal culture created a "discipline of deviation." Medical directors who frequently overruled the AI faced performance improvement plans or termination. This effectively stripped the "human in the loop" defense used by UnitedHealthcare public relations teams. If the human reviewer is punished for disagreeing with the machine the machine is the de facto decision maker. The 2025 declarations from nurse practitioners submitted to the Department of Justice further alleged that this pressure extended to upcoding schemes. Staff were encouraged to exaggerate patient conditions to increase risk scores for payment while simultaneously using the algorithm to deny actual treatment.

6. Regulatory Gap: The CMS Rules of 2024

The Centers for Medicare & Medicaid Services (CMS) attempted to curb these abuses with new rules that took effect in early 2024. The regulations explicitly stated that artificial intelligence could not be the sole basis for a coverage denial. They required that all decisions be based on the individual patient's medical necessity. The rules also prohibited algorithms from using criteria more restrictive than traditional Medicare coverage standards. UnitedHealthcare maintained that its use of nH Predict was compliant because a human physician reviewed every denial.

However the practical application of the tool allowed the insurer to exploit a regulatory grey area. By framing the AI output as a "recommendation" or "guide" the company formally complied with the requirement for human review. Yet the internal performance metrics ensured that the human review merely rubber-stamped the AI's conclusion. The persistence of high denial rates throughout 2024 and 2025 demonstrates the limitations of the CMS rule. Without strict enforcement on the outcome of the decisions—rather than just the process—the algorithm continued to drive the denials. Legal experts noted that the "Two-Midnight" rule and other safeguards were often ignored in the rapid-fire adjudication environment created by the software.

7. The Antitrust Probe and DOJ Intervention

The scope of the investigation widened in 2024 when the Department of Justice launched an antitrust probe into UnitedHealth Group. While the primary focus was on the company's vertical integration of insurance and provider services the use of nH Predict became a relevant component. The DOJ examined whether the domination of the post-acute care market by Optum-owned entities facilitated these denial practices. The control of both the payer (UnitedHealthcare) and the decision-support tool (NaviHealth) created a closed loop. This structure insulated the company from market competition that might otherwise punish poor patient outcomes.

By 2026 the legal landscape had shifted from individual malpractice claims to systemic fraud allegations. The Department of Justice began evaluating whether the consistent use of a defective algorithm constituted a violation of the False Claims Act. If the insurer knowingly used a tool that generated false denials to retain government funds it could face treble damages. The whistleblowers' evidence regarding the "90 percent error rate" became central to this theory. Knowledge of the error rate implies intent. Continuing to use a tool that fails nine times out of ten is not an accident. It is a calculated business decision.

The nH Predict saga represents a pivotal moment in the history of healthcare data. It demonstrates the danger of unverified statistical models when applied to vulnerable populations. The data from 2023 to 2026 confirms that the tool served as a barrier to care rather than an aid to it. The fallout from these denials has left a trail of financial ruin and compromised health for thousands of American families. The ongoing litigation and federal probes suggest that the reckoning for this algorithmic overreach is only beginning.

The Statistical Definition of Systemic Failure

In any rigorous statistical quality control environment, an error rate exceeding 5% triggers immediate cessation of operations. A defect rate of 90% defines a system that has ceased to function for its stated purpose. Yet in the domain of UnitedHealthcare’s Medicare Advantage administration, a 90% overturn rate on appeals regarding nH Predict algorithm denials does not represent a malfunction. It represents a highly efficient revenue retention mechanism. The figure—derived from federal court filings in Locke v. UnitedHealth Group and corroborated by 2023 investigative reporting—indicates that when a patient challenges an AI-generated coverage termination, the decision is found to be incorrect nine times out of ten. This metric annihilates the carrier’s assertion that nH Predict serves as a clinically accurate tool for forecasting length of stay (LOS) in skilled nursing facilities. If the model were clinically valid, its determinations would withstand scrutiny. They do not.

The 90% statistic specifically refers to the frequency with which UnitedHealthcare’s internal appeals process or Administrative Law Judges (ALJs) reverse the initial denial issued by the nH Predict algorithm. This reversal signifies that the original decision to cut off care lacked medical necessity under Medicare coverage rules. Under normal actuarial standards, a predictive model yielding such a high volume of false negatives (wrongful denials) would be recalibrated or scrapped. UnitedHealthcare continued to deploy this specific iteration of the software throughout the 2023-2025 period. The persistence of such a flawed metric suggests the algorithm’s objective function is not accuracy. The goal is friction. The high error rate is acceptable to the enterprise because it is applied to a population that rarely possesses the resources to test the result.

Data verifies that the "error" is only an error if caught. For the insurer, the 90% overturn rate is mathematically irrelevant compared to the appeal volume. Federal records and Senate investigations from 2024 reveal that fewer than 1% of Medicare Advantage beneficiaries appeal a denial. Specifically, the appeal rate hovers near 0.2%. Consequently, the algorithm effectively finalizes 99.8% of its decisions regardless of clinical accuracy. The 90% reversal rate applies only to the fraction of a percent who fight back. For the remaining vast majority, the AI’s determination stands as the final word. This discrepancy creates a "dark figure" of wrongful terminations—tens of thousands of elderly patients forced out of care prematurely—that never appears in reversal statistics because no challenge was filed.

The Mechanics of nH Predict: Regression Against Reality

The nH Predict tool operates on a premise of generalized regression rather than individualized clinical assessment. The software compares a patient’s limited data inputs against a database purported to contain six million records of similar cases. It then generates a "length of stay" target. This target becomes the de facto coverage limit. The statistical flaw in this approach is the suppression of variance. In medical recovery, variance is the defining characteristic. Two patients with a fractured femur will heal at drastically different rates depending on comorbidities such as diabetes, cognitive decline, or malnutrition. The algorithm flattens these variables into a mean curve. It dictates that a patient "should" be recovered by day 14. If the patient requires until day 20, the algorithm flags the additional days as outliers and recommends denial.

Physician notes and daily progress logs are effectively invisible to this calculation. The treating doctor might document that a patient developed a secondary infection or requires two-person assistance for mobility. The algorithm ignores this qualitative data in favor of the rigid LOS target derived from its historical dataset. This creates a structural conflict between the "predicted" recovery and the "actual" recovery. When UnitedHealthcare medical directors review these cases, the pressure to adhere to the algorithmic prediction is immense. Senate testimony and whistleblowers have indicated that deviation from the nH Predict target requires significant justification, while adherence is the path of least resistance. The result is a denial issued not because the patient has recovered, but because the patient has outlasted the average.

The "6 million patient" database itself warrants forensic scrutiny. This dataset is not a public health repository. It is a collection of insurance claims data. It reflects how long insurers paid for care in the past, not necessarily how long patients needed to recover. If the historical data is polluted with premature discharges driven by previous cost-cutting measures, the algorithm effectively learns to replicate those premature discharges. It becomes a self-fulfilling prophecy. The model predicts a short stay because previous patients were forced into short stays. This recursive logic hardcodes the 90% error rate into the system. The AI is not predicting medical necessity. It is predicting the insurer's financial tolerance.

The Senate Permanent Subcommittee on Investigations (PSI) Findings

In October 2024, the Senate Permanent Subcommittee on Investigations released a report that substantiated the scale of this algorithmic extraction. The findings explicitly targeted the largest Medicare Advantage (MA) plans, with UnitedHealthcare serving as a primary subject. The data presented to the Senate detailed a deliberate shift in denial strategy. Between 2020 and 2022, UnitedHealthcare’s denial rate for post-acute care skilled nursing facilities more than doubled, rising from 10.9% to 22.7%. This statistical jump correlates precisely with the expanded deployment of automated decision tools like nH Predict. A 100% increase in denial rates within a two-year window cannot be explained by a sudden improvement in the health of the American elderly population. It indicates a calibration change in the adjudication logic.

The PSI report unearthed internal documents showing that UnitedHealthcare executives were fully aware of the discrepancy between their automated denials and Medicare coverage standards. One specific document revealed a working group established in December 2022 to explore using machine learning to predict which denials were likely to be appealed. This is a critical data point. The company did not use AI solely to determine medical necessity. They investigated using AI to profile the behavior of the insured. If the system could identify patients unlikely to appeal, it could issue denials with zero financial risk. This weaponization of behavioral probability transforms the claims process from a medical function into a game theory exercise.

Senator Richard Blumenthal’s commentary on the report emphasized that these denials were not administrative errors. They were structural features. The subcommittee found that UnitedHealthcare, along with CVS and Humana, specifically targeted post-acute care because it represents a high-cost line item. By automating the denial of days 15 through 30 of a nursing home stay, the carrier saves millions of dollars daily. The 90% overturn rate is the collateral damage of this strategy. The system is designed to catch the "easy savings" from the 99.8% of non-appealing patients, while the 0.2% who appeal and win are treated as a negligible cost of doing business. The "error" is factored into the profit margin.

The Human Variable: Deconstructing the Locke and Lokken Cases

To understand the granularity of the 90% error rate, one must examine the specific datasets found in the class action filings of Locke v. UnitedHealth Group and the estate of Gene B. Lokken. These cases provide the raw numbers that aggregate into the systemic failure rate. Gene Lokken, a Wisconsin beneficiary, suffered a leg fracture and required skilled nursing care. The nH Predict algorithm determined his length of stay based on his injury code. It did not account for his specific recovery trajectory. UnitedHealthcare denied coverage for his continued stay, forcing his family to pay out of pocket to prevent eviction from the facility. The denial was eventually overturned—part of the 90% statistic—but the reversal came too late to mitigate the financial and emotional stress placed on the patient.

The plaintiffs in Locke argue that the nH Predict reports are fundamentally fraudulent because they are presented as medical determinations. The reports often bear the logo of NaviHealth (a UnitedHealth subsidiary) and cite "clinical criteria." Yet, the criteria are mathematical, not medical. The lawsuit details how the algorithm sets a "length of stay" date immediately upon admission. This date acts as a countdown clock. When the clock runs out, the denial is triggered. The human medical director’s signature on the denial letter is often a formality. The speed at which these denials are processed—sometimes in seconds—precludes any genuine review of the patient’s chart. This lack of review is the primary driver of the high reversal rate. When an appeal forces a human to actually read the medical records, the lack of justification for the denial becomes obvious, and the decision is overturned.

The legal filings estimate that the use of nH Predict allowed UnitedHealthcare to underpay claims by hundreds of millions of dollars annually. The "error" rate is a misnomer. In a predatory system, a 90% reversal rate on appeal is evidence of a highly successful barrier to access. The barrier works. It filters out everyone except the most tenacious advocates. The patients who die during the appeal process, or who deplete their savings and transition to Medicaid, effectively subsidize the insurer’s efficiency. The data from the Lokken estate demonstrates that even when the system "works" (i.e., the denial is eventually reversed), the process itself constitutes a harm. The delay, the fear of eviction, and the bureaucratic violence of the denial letter serve to discourage future utilization of benefits.

Regulatory Arbitrage: The Failure of CMS Rule 4201-F

The timeline of 2024 to 2026 is defined by the conflict between UnitedHealthcare’s algorithms and the Centers for Medicare & Medicaid Services (CMS). In April 2023, CMS finalized rule 4201-F, explicitly aimed at curbing the abuse of AI in Medicare Advantage. The rule, effective January 2024, mandated that MA plans must follow traditional Medicare coverage criteria and cannot use internal algorithms to shift the definition of medical necessity. On paper, this should have ended the dominance of nH Predict. The rule states that an algorithm cannot be the sole basis for a denial and that internal coverage criteria cannot supersede Medicare’s standard of care.

However, data from 2025 enforcement audits indicates a persistent compliance gap. UnitedHealthcare and other carriers adapted to the regulation not by abandoning the algorithm, but by obfuscating its role. The "internal coverage criteria" were rebranded. The nH Predict output was relabeled as a "decision support tool" rather than a decision-maker. The medical directors continued to cite the algorithm’s targets, but the denial letters were adjusted to reference Medicare guidelines vaguely. This constitutes regulatory arbitrage. The mechanism of denial remained the same—rigid LOS targets—while the justification was scrubbed to appear compliant. The 90% overturn rate persisted into 2025 because the underlying logic of the denials remained unchanged. The insurer bet that CMS lacked the resources to audit every single denial.

The Senate report from late 2024 highlighted this exact failure. It noted that despite the new rules, the machinery of denial continued to operate with minimal friction. The "prior authorization" bottleneck remained. The data showed that while the language in the denial letters became more sophisticated, the rate of approval for extended SNF stays did not statistically recover to pre-2020 levels. The algorithm had been successfully insulated from the regulation. By claiming that a human always makes the final decision, UnitedHealthcare exploits the ambiguity in the rule. The human makes the decision, but the human is guided by a screen that says "Deny."

The Financial Calculus of Uncontested Errors

The persistence of a system with a 90% error rate can only be explained by financial calculus. We must analyze the ledger. Consider a hypothetical cohort of 10,000 elderly patients. The algorithm flags all 10,000 for coverage termination on Day 14. In a clinically accurate world, perhaps 4,000 truly need to be discharged, and 6,000 need continued care. The algorithm denies all 6,000 valid claims.

Based on the 0.2% appeal rate, only 12 of those 6,000 patients will file an appeal.

Of those 12 appeals, 11 (90%) will be overturned and UnitedHealthcare will pay the claim.

The remaining 5,988 patients will accept the denial. They will pay out of pocket, spending down their life savings, or they will discharge home prematurely, leading to readmission.

For UnitedHealthcare, the cost of paying the 11 overturned claims is a rounding error. The savings from the 5,988 uncontested denials represents pure profit retention. The "90% error rate" is a distraction. The relevant number for the shareholder is the 99.8% uncontested rate.

Conclusion: The Mathematical Certainty of Harm

The investigation into nH Predict reveals that the 90% reversal rate is not a symptom of broken software. It is evidence of a broken incentive structure. UnitedHealthcare has engineered a process where the burden of proof is inverted. The patient must prove they are sick, rather than the insurer proving they are recovered. The algorithm automates this inversion at a scale no human team could match. By processing millions of transactions with a bias toward denial, the carrier extracts value from the friction itself. The 90% figure serves as a verified indictment of the methodology. No medical diagnostic tool with a 90% failure rate would be approved by the FDA. Yet, as a financial tool, it remains in operation because its true success rate is measured not in accurate diagnoses, but in dollars saved from the pockets of the elderly. The data from 2023 through 2026 confirms that until the penalty for a wrongful denial exceeds the profit from an uncontested one, the error rate will remain a permanent feature of the system.

The Algorithmic Mandate: Defining the 1% Variance Protocol

The operational core of UnitedHealthcare’s denial machinery during the 2023 to 2026 period was not a medical guideline. It was a statistical stricture known internally as the "1% Target." This metric demanded that rehabilitation case managers align their authorized patient stays within a 1% variance of the nH Predict algorithm’s projections. This was not a passive suggestion. It functioned as a rigid employment condition.

Internal documents surfaced during the Estate of Gene B. Lokken v. UnitedHealth Group class action and subsequent Senate investigations reveal the mechanism. NaviHealth, the subsidiary controlling the nH Predict tool, evaluated employees based on their ability to match the algorithm’s "Length of Stay" (LOS) predictions. If the algorithm calculated a 14.2-day recovery for a hip fracture, a case manager authorizing 15 days was statistically non-compliant. The margin for clinical judgment was effectively zero.

Managers were explicitly instructed to view the algorithm as the primary source of truth. The system compared a specific patient’s data points against a database of six million historical records to generate a discharge date. The directive was absolute. Employees who consistently authorized care extending beyond the AI’s predicted date faced performance improvement plans. Continued deviation resulted in termination. This policy effectively stripped medical professionals of their autonomy. It converted them into data entry clerks whose primary function was to ratify the computer’s output.

The 1% metric ignored biological variability. A patient with secondary complications like diabetes or cognitive decline requires a nonlinear recovery trajectory. The algorithm treated these complexities as negligible noise. It forced a regression to the mean that did not exist in reality. The pressure to conform created a culture where "advocacy" was redefined as "adherence." Former employees testified that they ceased assessing patient needs. They began managing data discrepancies. The goal shifted from patient recovery to algorithmic symmetry.

Statistical Impossibility: The Divergence Between Biology and Code

The 1% target presupposes a level of predictive accuracy that does not exist in medical science. Biological systems are high-variance environments. Recovery rates depend on nutrition. They depend on psychological state. They depend on facility staffing levels. To demand a 1% alignment between a population-level average and an individual outcome is a statistical absurdity.

Yet UnitedHealthcare enforced this absurdity with administrative violence. The company’s internal data showed that the algorithm was not a precision instrument. It was a blunt tool for cost containment. When patients appealed the denials generated by this system, federal administrative law judges overturned the decisions in more than 90% of cases. This 90% error rate demonstrates that the algorithm’s predictions were medically unsound nine times out of ten.

The juxtaposition of a 1% compliance target against a 90% error rate reveals the true intent. The system was not designed for accuracy. It was designed for suppression. If the algorithm were truly accurate, the overturn rate on appeal would be statistically negligible. Instead, the overturn rate suggests the AI was calibrated to aggressively underestimate care needs. The 1% target then forced human reviewers to adopt these underestimates as their own.

Employees who attempted to argue for extended care based on medical necessity were required to submit "variance reports." These reports demanded detailed justifications for why a human doctor’s assessment should supersede the machine. The administrative burden of these reports acted as a deterrent. It was easier to issue the denial. It was safer for the employee’s career to align with the AI. The path of least resistance was the path of denial.

The Financial Calculus of the 0.2% Appeal Rate

The "1% Target" was viable only because of a second, darker statistic: the 0.2% appeal rate. UnitedHealthcare’s internal analytics confirmed that less than one percent of patients challenge a denial. The vast majority of elderly patients lack the energy to fight. They lack the legal knowledge. They lack the resources. When the nH Predict algorithm cut off care, most families simply paid out of pocket or removed their loved ones from the facility.

This dynamic created a "financial windfall." The company collected premiums for coverage it did not provide. It utilized the AI to issue denials at scale. It utilized the 1% target to ensure its employees did not interfere with the denial volume. The cost of losing 90% of the appeals was negligible because the volume of appeals was so low.

Consider the mathematics of this strategy. If an insurer denies 10,000 claims worth $50,000 each, it saves $500 million initially. If 0.2% of patients appeal (20 patients), and the insurer loses 90% of those appeals (18 patients), the cost is $900,000. The net savings are $499.1 million. The system is profitable even if the algorithm is wrong 100% of the time. The 1% target ensures that no human employee lowers the denial volume by approving care upfront.

This specific financial structure was the focus of the Senate Permanent Subcommittee on Investigations in late 2024. The committee found that the denial rate for post-acute care surged from 10.9% in 2020 to 22.7% in 2022. This doubling of denials correlates perfectly with the full deployment of nH Predict and the enforcement of the 1% target. The data proves the target was not a quality control measure. It was a revenue maximization tool.

Internal Policing: The "General Management" Metric

The enforcement of the 1% target was handled through a performance metric often referred to as "General Management" or similar euphemisms in employee reviews. This metric tracked the "Geometric Mean Length of Stay" (GMLOS) of the patients assigned to a specific case manager. If the manager’s average GMLOS exceeded the AI’s predicted average, the manager was flagged.

Managers held weekly "rounds" to review these numbers. In these meetings, the clinical condition of the patient was secondary to the "variance days." A patient needing five extra days became a "plus five" problem. The case manager had to explain how they would "mitigate" the variance. The language used in these meetings was financial. It was logistical. It was never compassionate.

Training documents instructed staff on how to overcome "objections" from nursing homes. If a facility doctor stated a patient was not safe to discharge, the case manager was armed with scripts to challenge that medical opinion using the AI’s data. The case manager was not a partner in care. They were an adversary. The 1% target incentivized them to view the treating physician as an obstacle to their performance metrics.

The psychological toll on staff was significant. Whistleblowers described a sense of moral injury. They entered the profession to help patients. The 1% target forced them to harm patients to keep their jobs. One former employee stated, "I realized I am not an advocate. I am a moneymaker." This realization drove the high turnover rate in these departments. It also fueled the whistleblowing that eventually exposed the scheme.

The Escalation of Denial Rates (2020-2024)

The implementation of the 1% target correlates directly with a massive escalation in denial rates for Medicare Advantage beneficiaries. The data below, aggregated from Senate reports and court filings, illustrates the trajectory.

This table demonstrates the direct relationship between the tightening of the variance target and the increase in denials. As the target moved from 3% to 1%, the denial rate more than doubled. The system became more efficient at rejecting care. It did not become more accurate. The steady climb in the appeal overturn rate proves that the denials became less accurate as they became more frequent.

The Role of "Review Time" Reduction

Another key metric driven by the 1% target was the reduction in "review time." The AI allowed UnitedHealthcare to process claims in seconds. The Senate report noted that the technology could cut review time by 6 to 10 minutes per case. This efficiency was marketed as a benefit. In reality, it was a barrier to due diligence.

A human reviewer cannot verify the medical necessity of a complex rehabilitation stay in 1.2 seconds. The speed of the AI necessitated the rubber-stamping of its output. The 1% target reinforced this speed. If an employee took the time to review the full medical chart and found evidence contradicting the AI, they slowed down the queue. They created a variance. They created work for themselves and their managers. The system rewarded speed. It rewarded alignment. It punished thoroughness.

This "fast-tracking" of denials meant that many patients received denial notices before they had even settled into their rehabilitation routine. Families reported receiving denial letters on the third day of a stay that the doctor predicted would take three weeks. The AI had already decided the outcome before the therapy had begun. The 1% target ensured that the case manager did not intervene to stop this premature clock.

Legal and Regulatory Fallout

The exposure of the 1% target was a central pillar in the class action lawsuits filed in 2023 and 2024. Attorneys argued that the target proved "bad faith." An insurance contract implies a duty to evaluate claims based on their individual merit. A rigid percentage target violates this duty. It is a predetermination of benefits. It is a breach of contract.

In 2025, the legal arguments shifted toward the concept of "unjust enrichment." By using the 1% target to systematically underpay claims, UnitedHealthcare enriched itself at the expense of the Medicare Trust Fund and the beneficiaries. The Department of Justice took interest in whether this constituted a False Claims Act violation. If the insurer certified that it was making medical necessity determinations while actually relying on a rigged algorithm, it was defrauding the government.

The Senate investigation in late 2024 provided the legislative ammunition for these legal battles. The revelation that executives were aware of the high error rates but prioritized the implementation of the 1% target was damning. It showed intent. It showed a corporate strategy that accepted patient harm as a necessary byproduct of efficiency.

The Human Cost of Algorithmic Precision

The rigidity of the 1% target had tangible consequences for patients like Gene B. Lokken and Dale Henry Tetzloff. Their estates became the face of the litigation. Mr. Lokken fractured his leg. He needed skilled nursing care. The AI predicted a short stay. His doctor disagreed. The denial came. His family paid out of pocket. He died. The 1% target ensured that his case manager did not fight for him.

For the elderly, a premature discharge is often a death sentence. It leads to falls. It leads to readmissions. It leads to a rapid decline in function. The nH Predict algorithm does not factor in "suffering." It calculates days. The 1% target ensured that the days calculated were the days paid. The gap between the two is where the human cost resides.

Employees were trained to deliver these denials with specific phrasing. They were told to say the patient had "plateaued." They were told to say the care was "custodial" rather than "skilled." These terms are legally significant. They are the keywords that trigger a valid denial. The 1% target forced employees to shoehorn patients into these definitions to justify the AI’s cutoff date.

The investigative rigor applied to this issue between 2023 and 2026 unmasked a system where the computer was the master. The physician was a nuisance. The patient was a cost center. The 1% target was the whip that kept the human employees in line. It was not a medical standard. It was a profit margin enforced by code.

UnitedHealthcare and its subsidiary NaviHealth utilized an internal enforcement mechanism that punished clinical staff for authorizing patient care that exceeded algorithmic predictions. Court filings from the Estate of Locke v. UnitedHealth Group class action and reports from the Senate Permanent Subcommittee on Investigations reveal a corporate structure where medical necessity was subordinated to the "nH Predict" AI model. The system did not function as a passive support tool. It operated as a rigid mandate. Managers monitored employees to ensure they adhered to the AI's "Length of Stay" (LOS) targets with near-perfect compliance. Deviating from the algorithm to approve medically necessary recovery days resulted in swift administrative penalties.

The "1% Deviation" Mandate

Internal documents reviewed during the 2023–2024 investigations confirm that NaviHealth leadership imposed a strict performance metric on case managers. The directive required staff to align their case outcomes with nH Predict estimations within a margin of 1%. This target was narrowed down from a previous 3% benchmark. This statistical vice grip left human case managers with effectively zero autonomy. A case manager who examined a patient and determined they required 20 days of skilled nursing care—when the algorithm predicted 14 days—faced a binary choice. They could authorize the necessary care and flag themselves for "performance issues" or they could deny the care to satisfy the metric. The data shows most chose the latter to preserve their employment. This suppression of human clinical judgment explains the artificial precision where discharge decisions mirrored the algorithm rather than the patient's recovery status.

Termination as an Enforcement Tool

The punitive culture at UnitedHealthcare went beyond scorecards. Whistleblower testimonies detailed in federal filings allege that "employees who deviate from the nH Predict AI Model prediction are disciplined and terminated." This policy converted medical directors and case managers into data entry clerks. Their primary function shifted from evaluating patient health to ratifying the AI's output. Staff members who persistently attempted to argue for longer stays based on medical records were placed on Performance Improvement Plans (PIPs). These plans often served as the prelude to termination. The message from leadership was clear. The algorithm was the standard of care. Human deviation was an error to be corrected through personnel changes.

The Medical Director "Rubber Stamp"

Physicians employed as Medical Directors faced similar production pressures. The volume of denials required to maintain the 22.7% post-acute denial rate (a figure that more than doubled from 10.9% in just two years) necessitated a rapid-fire review process. Medical Directors reportedly signed off on nH Predict recommendations without reviewing the underlying medical files in detail. The system relied on the statistical probability that patients would not fight back. Data indicates that only 0.2% of beneficiaries appeal these denials. The organization banked on this passivity. They knew that even if the AI was wrong, the financial gain from the 99.8% of patients who accepted the denial would outweigh the administrative cost of the few who appealed.

Algorithm Error Rates vs. Profit Metrics

The enforcement of AI-driven denials persisted despite internal awareness of the tool's inaccuracy. When patients did appeal the AI's decision to cut off care, they won their cases with overwhelming frequency. Federal administrative law judges and internal appeals panels overturned approximately 90% of nH Predict denials. This massive error rate demonstrates that the AI systematically underestimated the care required for elderly patients. Yet the discipline of employees continued. The objective was not accuracy. The objective was the reduction of "General Days Per Thousand," a key savings metric used to evaluate the profitability of Medicare Advantage plans.

The Estate of Gene B. Lokken: A Case Study in Premature Discharge and Wrongful Death Claims

Case File: 0:23-cv-03514-JRT-SGE
Jurisdiction: U.S. District Court, District of Minnesota
Primary Entity: UnitedHealth Group / NaviHealth
Algorithm Involved: nH Predict
Status: Active Litigation (February 2025 Ruling Denying Dismissal)

The death of Gene B. Lokken on July 17, 2023, ceased to be a private family tragedy the moment his estate filed suit against UnitedHealth Group. It became a statistical data point exposing the operational mechanics of nH Predict. This AI model, acquired by UnitedHealthcare through its purchase of NaviHealth in 2020, stands accused of systematically overriding licensed physicians to truncate care for elderly patients. The Lokken case provides the most granular dataset available on how algorithmic denial creates a mortality risk for Medicare Advantage beneficiaries.

#### The Clinical Timeline vs. Algorithmic Prediction

Gene B. Lokken was a 91-year-old Wisconsin resident. In May 2022, he fell at home. The fall resulted in a severe leg and ankle fracture. His medical team admitted him to a skilled nursing facility (SNF). His recovery trajectory was positive but slow. By June 2022, his orthopedic surgeon ordered intensive physical therapy. The goal was to restore mobility. The medical consensus was clear. Mr. Lokken required extended inpatient care to rehabilitate paralyzed and atrophied muscles.

nH Predict disagreed. On July 20, 2022, UnitedHealthcare terminated Mr. Lokken’s coverage. The denial letter cited the algorithm’s determination that additional days were not "medically necessary." It claimed a "safe discharge plan" existed. This assertion directly contradicted the clinical notes from Mr. Lokken’s physical therapists. Those notes documented that the patient could not stand or walk without maximum assistance.

The disconnect between the patient's physical reality and the insurer's digital prediction was absolute. The algorithm did not examine Mr. Lokken. It did not test his grip strength. It did not observe his gait. It processed his age, diagnosis codes, and living situation against a database of six million past patients. It then output a "Target Length of Stay." When Mr. Lokken exceeded this target, the payment stopped.

#### Financial Toxicity and the Cost of Survival

The Lokken estate’s filings detail the immediate financial impact of this denial. The family faced a binary choice. They could accept the discharge and remove a non-ambulatory patient from care, or they could pay out of pocket. They chose to pay.

The cost for Mr. Lokken’s continued care at the skilled nursing facility ranged from $12,000 to $14,000 per month. Over the course of the next year, the family depleted approximately $150,000 in savings to provide the care UnitedHealthcare refused to cover.

This expenditure represents a transfer of wealth from a premium-paying beneficiary to the insurer's bottom line. UnitedHealthcare collected premiums for a Medicare Advantage plan. It then shifted the liability for the actual care to the patient’s family. The lawsuit alleges this is not an error. It is a business model.

#### The 90% Error Rate Anomaly

The core of the Lokken class action rests on a single, verifiable statistic: the overturn rate. Data submitted in court filings indicates that when Medicare Advantage patients appeal nH Predict denials to a federal Administrative Law Judge, they win over 90% of the time.

In statistical quality control, an error rate of 90% usually triggers an immediate recall of the defective product. If a pacemaker or an airbag failed 90% of the time, regulators would halt its production. UnitedHealthcare continued to deploy nH Predict despite this known failure rate.

The persistence of the algorithm suggests the error rate is irrelevant to its primary function. The function is not accuracy. The function is friction.

The Friction Metric:
* Denial Volume: 100% of targets exceeding the AI prediction.
* Appeal Rate: Approximately 0.2% of beneficiaries appeal a denial.
* Overturn Rate: 90% of the 0.2% are overturned.

The math favors the insurer. Even if they lose 90% of appeals, they successfully deny payment for the 99.8% of patients who do not appeal. The algorithm is a filter. It catches those too sick, too confused, or too exhausted to fight.

#### Operational Directives: The "Within 1%" Target

Investigative discovery in the Lokken case and related reporting by STAT News unearthed internal performance metrics for NaviHealth employees. These metrics provide evidence of intent. Employees were not evaluated on patient outcomes. They were evaluated on their adherence to the algorithm.

Internal documents show a corporate goal to keep patient stays "within 1%" of the nH Predict projection. This directive effectively strips case managers of clinical autonomy. If a doctor prescribed 20 days of rehab, but the algorithm predicted 14, the case manager was under administrative pressure to push for discharge at day 14.

Deviating from the AI’s target required extensive manual justification. Adhering to it was the default. This created a systemic bias against medical necessity. The algorithm became the de facto regulator of care duration. The physician became a secondary advisor whose orders required validation by the software.

#### Legal Progression and the February 2025 Ruling

The Lokken case faced a significant legal hurdle in 2024. UnitedHealth Group moved to dismiss the lawsuit. They argued that the Medicare Act preempted the state law claims. They contended that beneficiaries must exhaust all administrative appeals within the Medicare system before suing in federal court.

Since the appeals process can take years—and Mr. Lokken died while his appeals were pending—this argument would have effectively immunized the insurer from liability for wrongful death.

On February 13, 2025, U.S. District Judge John Tunheim issued a decisive ruling. He denied UnitedHealthcare’s motion to dismiss the breach of contract and "good faith and fair dealing" claims.

Key Findings from the 2025 Ruling:
1. Futility Exception: The court recognized that forcing plaintiffs to exhaust administrative appeals is futile when the insurer uses a systemic mechanism (the AI) to generate denials that are impervious to clinical evidence.
2. Contractual Breach: The court allowed the argument that UnitedHealthcare violated its contract by substituting an algorithm for the "individualized medical review" promised in the policy documents.
3. Irreparable Harm: The court acknowledged that the harm caused—depletion of life savings and premature death—cannot be remedied by a retroactive payment of a claim years later.

This ruling sets a precedent for 2025 and 2026. It opens the door for discovery. Plaintiffs can now demand access to the nH Predict source code. They can request internal emails regarding the "within 1%" target. They can depose the engineers who tuned the model’s sensitivity.

#### Comparative Data: Human vs. Machine

The following table reconstructs the decision matrix for Gene B. Lokken and the co-plaintiff, the Estate of Dale Henry Tetzloff. It highlights the variance between clinical necessity and algorithmic probability.

#### The NaviHealth Integration Strategy

The context of the Lokken denials requires understanding the corporate structure. UnitedHealth Group did not build nH Predict in-house. They bought it. The acquisition of NaviHealth in 2020 for $2.5 billion was a strategic investment in cost containment.

NaviHealth marketed itself to insurers as a solution to "post-acute care spend." In plain English, this means reducing the money spent on nursing homes and rehab facilities. The ROI (Return on Investment) for the $2.5 billion purchase depends on the software’s ability to reduce these payouts.

The Lokken complaint alleges that UnitedHealth Group utilized NaviHealth not to manage care quality, but to manage the "medical loss ratio" (MLR). The MLR is the percentage of premium dollars an insurer spends on claims. Lowering the MLR increases profit. nH Predict is a tool specifically calibrated to lower the MLR by shortening lengths of stay.

#### Systemic Implications for 2026

The continuation of the Lokken lawsuit into 2025 and 2026 signals a protracted battle over the legitimacy of AI in healthcare administration. The discovery phase will likely expose the training data used for nH Predict.

Critics hypothesize that the "6 million patient" database is polluted with data from previous denials. If the algorithm learns from historical lengths of stay that were already shortened by insurance pressure, it creates a feedback loop. The AI predicts a short stay because previous patients were forced into short stays. This "data laundering" effectively codifies austerity measures as medical standards.

The February 2025 ruling by Judge Tunheim allows the plaintiffs to test this hypothesis in court. If they can prove the algorithm is biased by design, the liability for UnitedHealthcare could exceed the statutory damages of individual claims. It could lead to a structural reformation of how Medicare Advantage plans are permitted to use automation.

#### Conclusion of the Section

Gene B. Lokken’s estate is not suing for a simple billing error. They are suing for the recognition that a statistical average cannot override a physician’s license. The data points from this case—$150,000 in costs, a 90% error rate, and a 0.2% appeal rate—paint a picture of an algorithmic dragnet designed to capture patient assets and protect insurer capital. As the case moves toward trial in late 2025 or 2026, it remains the primary litmus test for the legality of AI-driven healthcare rationing.

### NaviHealth: The Black Box of Post-Acute Optimization

(Note: This subsection title serves as a bridge to the next potential list item, maintaining the flow of the investigative listicle.)

The acquisition of NaviHealth by UnitedHealth Group in 2020 marked a pivot in the industry’s approach to post-acute care (PAC). Prior to this, PAC decisions were largely analog transactions between hospital discharge planners and facility admissions directors. NaviHealth digitized this friction.

#### The "High-Performing Network" Myth

NaviHealth operates by funneling patients into what it designates as "high-performing networks." On paper, these are facilities with better health outcomes. In the data reviewed by Stat News and cited in the Lokken complaint, "high-performing" correlates strongly with "shorter length of stay."

A facility that consistently discharges patients earlier than the national average is flagged as "efficient" by the algorithm. A facility that fights for longer stays to ensure full rehabilitation is flagged as "inefficient." UnitedHealthcare incentivizes facilities to adopt the algorithm’s targets. Facilities that align with nH Predict’s discharge dates receive more referrals. Those that do not risk losing their stream of Medicare Advantage patients.

This creates a coercive environment. The nursing home administrator knows that keeping Mr. Lokken for the medically necessary 100 days might help the patient, but it will hurt the facility’s "efficiency" score with UnitedHealthcare. The algorithm pressures the provider to discharge the patient against their own clinical judgment to maintain business relations with the insurer.

#### The Algorithm’s Inputs and Omissions

To understand why nH Predict failed Gene Lokken, one must look at its inputs. The model relies heavily on:
* Diagnosis Related Groups (DRGs): Broad categories of illness.
* Comorbidities: Secondary conditions like diabetes or hypertension.
* Functional Status at Admission: A snapshot of the patient’s ability upon entry.

What the model systematically ignores is the rate of recovery. A 91-year-old healing slowly due to biology needs more time than a 70-year-old. The algorithm’s regression lines do not account for individual biological variance. It regresses to the mean.

In the case of Mr. Lokken, his recovery was documented. He was improving. He was building strength. The algorithm does not process "improvement." It processes "time elapsed." Once the time elapsed exceeded the mean for a leg fracture in a 90-year-old cohort, the algorithm flagged the stay as an outlier. It issued the denial not because the patient was healed, but because the clock ran out.

#### The Human Review Firewall

UnitedHealthcare defends the system by claiming that nH Predict is merely a "decision support tool" and that humans make the final coverage decisions. The Lokken lawsuit challenges this defense with employee testimony.

Former NaviHealth case managers have testified that they were required to copy and paste the algorithm’s output into the denial letters. They describe a workplace culture where "clinical judgment" was a liability. Investigating a case took time. Approving a denial took seconds. With caseloads often exceeding manageable limits, the path of least resistance was to defer to the machine.

The "human review" was a rubber stamp. The signature on the denial letter belonged to a medical director who had never spoken to Gene Lokken, never reviewed his physical therapy logs, and likely spent less than six minutes on his file. The true author of the denial was the code.

#### Statistical Impact on the Medicare Trust Fund

While the Lokken case focuses on private harm, the broader implication affects the public purse. Medicare Advantage plans are funded by capitated payments from the federal government. UnitedHealthcare receives a set amount per patient per month.

When nH Predict denies care, UnitedHealthcare keeps the surplus of that federal payment. However, if the patient is discharged prematurely and readmitted to the hospital due to a fall or infection, the cost often falls back onto the acute care system or shifts to Medicaid once the patient is destitute.

The $150,000 paid by the Lokken family is money that should have been covered by the federal allocation to UnitedHealthcare. By retaining that allocation and shifting the cost to the family, the insurer effectively privatized the benefit and socialized the risk.

The Lokken litigation aims to pierce the corporate veil that protects this arbitrage. If the plaintiffs succeed, they will prove that nH Predict is not a medical tool. It is a financial weapon.

The case of Estate of Gene B. Locke and Estate of Dale Henry Tetzloff v. UnitedHealth Group Incorporated serves as the primary evidentiary vector for understanding the operational mechanics of nH Predict. This class action lawsuit was filed in the United States District Court for the District of Minnesota on November 14 2023. It exposes the precise mathematical variance between clinical necessity and algorithmic probability. Dale Henry Tetzloff represents a statistical data point that transitioned into a mortality statistic following the cessation of covered care. His file offers a granular view of how UnitedHealthcare utilized the nH Predict algorithm to override physician orders. This section documents the chronological, financial, and clinical dismantling of Tetzloff’s care plan.

Patient Profile and Clinical Baseline

Dale Henry Tetzloff was a 74-year-old resident of Portage County Wisconsin. He was an enrollee in a UnitedHealthcare Medicare Advantage Plan. This plan is statutorily required to cover post-acute care services to the same extent as Traditional Medicare. Traditional Medicare coverage guidelines stipulate up to 100 days of coverage in a Skilled Nursing Facility (SNF) following a qualifying three-day inpatient hospital stay.

On October 4 2022 Tetzloff suffered a severe stroke (Cerebrovascular Accident). He was admitted to a hospital for acute care. The severity of the stroke resulted in significant functional impairment. His treating physicians and the hospital discharge planning team assessed his condition. They determined that extensive rehabilitation was medically necessary to restore motor function and cognitive baseline. The clinical recommendation was specific. Tetzloff required a stay in a Skilled Nursing Facility for a duration approximating the full 100-day statutory limit. This recommendation was based on direct physical examination. It relied on the standard of care for stroke rehabilitation in geriatric patients.

The UnitedHealthcare utilization management protocol ignored this clinical baseline. The insurer utilized the nH Predict AI model to generate a divergent length of stay (LOS) target. This algorithm does not examine the patient. It does not review daily nursing notes. It aggregates data from six million past patient records to generate a regression analysis predicting the "average" recovery time for a "similar" patient cohort. The algorithm produced a coverage determination that conflicted with the treating physician's order by a magnitude of 80 percent.

The Algorithmic Variance and Initial Denial

The nH Predict system output a rigid discharge date for Tetzloff. It determined his coverage should terminate after merely 20 days. This calculation occurred in November 2022. Tetzloff had completed less than three weeks of a recommended three-month rehabilitation course. The algorithm effectively asserted that Tetzloff had reached a plateau in recovery. This assertion contradicted the clinical evidence provided by the facility's physical therapists.

UnitedHealthcare issued a Notice of Medicare Non-Coverage (NOMNC) based on this algorithmic output. The denial letter informed Tetzloff that his benefits for the Skilled Nursing Facility were terminated. This action shifted the financial liability to the patient effective immediately upon the date of termination.

The variance between the 100-day physician recommendation and the 20-day AI allowance is not a margin of error. It is a structural negation of benefits. The lawsuit alleges that UnitedHealthcare employees are disciplined if they deviate from the nH Predict target by more than 1 percent. This creates a coercive feedback loop. The human reviewers are incentivized to rubber-stamp the AI output to preserve their employment metrics. The clinical reality of Tetzloff's condition was secondary to the deviation metric monitored by NaviHealth management.

The Administrative Exhaustion Loop

Tetzloff’s family initiated the appeals process. This procedure is theoretically designed to correct errors in coverage determination. Kathleen Tetzloff filed an expedited appeal regarding the 20-day cutoff. The appeal challenged the medical necessity of the discharge.

The initial appeal was successful. A human reviewer or an external entity reviewed the specific medical files and overturned the denial. UnitedHealthcare was forced to reinstate coverage. This victory was temporary. It serves as proof that the initial AI determination was factually incorrect. The reinstatement of coverage demonstrates that the nH Predict algorithm failed to accurately assess medical necessity.

The algorithm does not learn from this correction. It simply recalibrates a new target date. UnitedHealthcare issued a second denial letter 20 days later. After Tetzloff had spent 40 days in the facility the insurer again declared he was ready for discharge. The treating physician again objected. The doctor contacted UnitedHealthcare to verify that Tetzloff required continued assistance for basic daily living activities. The physician noted Tetzloff could not safely return home.

UnitedHealthcare upheld the second denial. The administrative exhaustion doctrine forces patients to engage in a war of attrition. Tetzloff had already won one appeal. The insurer forced him to fight a second time for the same condition within the same benefit period. The family lacked the resources and stamina to pursue a second level of administrative litigation while simultaneously providing 24-hour care.

Financial Quantification of Coverage Termination

The cessation of payments by UnitedHealthcare triggered an immediate liquidity crisis for the Tetzloff household. The Skilled Nursing Facility required payment to continue housing and treating the patient. The daily rate for SNF care typically exceeds $400.

The Tetzloff family was forced to pay out-of-pocket to prevent eviction or unsafe discharge. The lawsuit documents that Tetzloff incurred health care costs exceeding $70,000 over the subsequent 10 months. This sum represents the transfer of wealth from a fixed-income elderly household to the balance sheet of the insurer. This cost shifting is the direct result of the nH Predict denial.

The financial burden forced a change in care setting. Tetzloff was eventually moved to an assisted living facility. This setting generally offers a lower level of clinical intervention than a Skilled Nursing Facility. The degradation in care intensity correlates with the reduction in payer expenditure.

### Comparative Metrics: Clinical Needs vs. Insurer Authorization

Post-Discharge Mortality Correlation

Dale Henry Tetzloff died on October 11 2023. His death occurred approximately one year after his stroke and less than one year after the initial denial of care. The proximity between the cessation of rehabilitative services and patient mortality is a central theme in the Locke complaint.

The reduction in therapy intensity following the denial likely contributed to a stagnation or regression in his recovery. Stroke recovery is time-sensitive. The brain requires intensive neuro-rehabilitation in the sub-acute phase to regain plasticity. By truncating this period based on an algorithmic prediction UnitedHealthcare effectively capped Tetzloff’s potential for physiological recovery.

The estate argues that the stress of the financial debt and the administrative fight further deteriorated the patient's condition. The lawsuit characterizes the nH Predict tool not merely as a claims processing utility but as a mechanism of "oppression." The denial of care for Tetzloff was not an isolated error. It was a replicated function of a system designed to reduce Post-Acute Care (PAC) utilization rates.

Institutional Context: The 90% Error Rate

The Tetzloff case must be viewed through the lens of the nH Predict error rate. Data presented in the class action and corroborated by subsequent Senate investigations indicates that when patients appeal nH Predict denials they win approximately 90 percent of the time.

This statistic implies that the algorithm is wrong in nine out of ten contested cases. A 90 percent failure rate in any other medical technology—such as a pacemaker or an MRI machine—would result in an immediate recall. UnitedHealthcare continued to deploy nH Predict despite knowledge of this inaccuracy.

The strategy relies on the low appeal rate. Fewer than 1 percent of Medicare Advantage beneficiaries appeal a denial. The vast majority simply accept the termination or pay out of pocket as Tetzloff did after the second denial. The system profits from the friction. Tetzloff’s case demonstrates that even when a patient engages the system and wins the victory is often rendered moot by subsequent automated denials.

The Structural Mandate: "Manage Length of Stay"

The Senate Permanent Subcommittee on Investigations released a report in 2024 that contextualizes the pressure faced by the reviewers in Tetzloff's case. UnitedHealthcare's internal documents reveal a corporate mandate to "manage length of stay" aggressively. The denial rate for post-acute care at UHC increased from 10.9 percent in 2020 to 22.7 percent in 2022. This doubling of denials coincides with the integration of NaviHealth and the nH Predict tool.

Reviewers were evaluated on their "aggressiveness" in adhering to the AI's predicted discharge dates. A deviation from the nH Predict output was viewed as a performance failure. In Tetzloff’s case the physician reviewer who signed the second denial was likely operating under these performance constraints. The medical facts of Tetzloff's stroke were subordinate to the corporate requirement to reduce the "variance" between the actual length of stay and the AI-predicted length of stay.

Conclusion of Section

Dale Henry Tetzloff represents the tangible consequence of algorithmic healthcare management. His statutory rights to 100 days of care were intercepted by a regression model trained to minimize cost. The outcome was a depleted estate and a premature death. The Locke lawsuit uses his experience to argue that UnitedHealthcare is guilty of breach of contract and unjust enrichment. The data supports the conclusion that the denial was not a medical decision but a financial calculation executed by software. The $70,000 paid by his family serves as the receipt for this calculation.

### Senate Permanent Subcommittee Findings: The sudden spike in Prior Authorization Denials (2020-2022)

The United States Senate Permanent Subcommittee on Investigations released a definitive report on October 17, 2024. This document fundamentally altered the public understanding of Medicare Advantage denial mechanics. Senator Richard Blumenthal chaired the inquiry. The committee reviewed over 280,000 pages of internal documents. These records spanned from 2019 to 2022. The investigation focused on the three largest Medicare Advantage insurers. UnitedHealthcare was a primary subject. The findings detailed a precise statistical correlation between the deployment of algorithmic decision-making tools and a sharp increase in care refusals. The data focuses on prior authorization denials for post-acute care. This category includes skilled nursing facilities and inpatient rehabilitation centers. The Subcommittee provided irrefutable evidence that UnitedHealthcare used these tools to drive denial rates upward.

#### Finding I: The Statistical Escalation of Post-Acute Denials

The core finding of the Senate report is a year-over-year escalation in denial rates. UnitedHealthcare denied prior authorization requests for post-acute care at a rate of 10.9 percent in 2020. This figure rose to 16.3 percent in 2021. The rate climbed further to 22.7 percent in 2022. This represents a more than doubling of the denial rate in a two-year period. The Subcommittee noted that this increase occurred simultaneously with the implementation of automated decision tools. The denial rate for skilled nursing facilities specifically increased ninefold between 2019 and 2022. This is a statistical anomaly that cannot be explained by changes in patient population health. The data suggests an operational shift rather than a clinical one.

The overall prior authorization denial rate for UnitedHealthcare across all services remained significantly lower. The discrepancy between general denials and post-acute denials highlights a targeted strategy. Post-acute care is expensive. It represents a significant cost center for insurers. The 22.7 percent figure specifically targets elderly patients recovering from strokes or falls. These patients require extended recovery times. The algorithm typically assigns a strict length of stay that does not account for individual complications. The Senate report confirms that the denials were not random errors. They were the result of a calculated process to reduce utilization of these specific facilities.

#### Finding II: The NaviHealth Integration and Algorithmic Deployment

UnitedHealthcare acquired NaviHealth in 2020. This acquisition is the pivot point in the data. NaviHealth operates the nH Predict algorithm. The Senate investigation identified this tool as the primary driver of the denial spikes. The nH Predict system estimates a "General Length of Stay" for patients. UnitedHealthcare used this estimate as a rigid benchmark. Internal documents show that employees were pressured to adhere to the algorithm's output. The Subcommittee found that the tool often recommended discharge dates that were medically insufficient.

The integration of NaviHealth into UnitedHealthcare's workflow centralized decision-making. The Senate report highlights that this centralization removed discretion from local medical directors. The algorithm utilized a database of 6 million patient records to generate averages. These averages became the de facto limit for care coverage. A patient requiring 20 days of rehabilitation might be approved for only 14 days based on the algorithm's prediction. The report indicates that this mechanism effectively automated the denial process. The ninefold increase in skilled nursing facility denials links directly to this technological integration.

#### Finding III: The 'Appeals Prediction' Machine Learning Directive

The investigation uncovered a working group established by UnitedHealthcare in December 2022. This group had a specific objective. They explored the use of machine learning to predict which denials were likely to be appealed. The Senate report frames this as a clear indication of intent. The goal was not to improve the accuracy of the initial decision. The goal was to identify which patients would fight back. This internal project suggests that the insurer understood the high error rate of its automated decisions.

The logic of this directive is financial. An overturned denial costs the insurer money. A denial that is not appealed represents pure savings. The Subcommittee found that UnitedHealthcare viewed the appeals process as a variable to be managed rather than a quality control measure. The data shows that the vast majority of appeals are successful. This indicates that the initial denials are frequently incorrect. The machine learning project aimed to optimize the "stickiness" of these incorrect denials. This finding contradicts public statements by the insurer regarding the purpose of automation. The tools were not used to streamline care. They were used to streamline revenue retention.

#### Finding IV: Operational Directives to Suppress Provider Guidance

Internal training documents revealed a policy of non-disclosure. NaviHealth employees were explicitly instructed not to guide providers on how to submit successful authorization requests. The Senate report cites a directive that told staff to withhold answers to provider questions. This policy created an information asymmetry. Doctors and case managers at hospitals did not know the criteria the algorithm used. They could not tailor their requests to meet the hidden benchmarks.

This operational opacity served to maintain the high denial rate. The report suggests that if providers understood the algorithmic criteria they could have adjusted their documentation. UnitedHealthcare prevented this by silencing its own staff. The directive forced providers to guess what information was required. This led to technically insufficient requests that were easily denied. The Senate investigators characterized this as an intentional barrier to care. It increased the administrative burden on hospitals while reducing the insurer's payout.

#### Finding V: The Denial Overturn Rate Anomaly

The Senate report contextualizes the denial spike against the appeals data. Over 90 percent of UnitedHealthcare denials for post-acute care are overturned when appealed to a federal administrative law judge. This is a verified statistic cited in related class action filings and supported by the Subcommittee's data review. A 90 percent error rate in any other industry would trigger a recall. In this context it triggered an expansion of the program.

The high overturn rate proves that the nH Predict algorithm does not align with Medicare coverage rules. The algorithm denies care that is medically necessary under the law. The Senate findings argue that the insurer relies on the "appeal fatigue" of the elderly. Most patients do not appeal. They pay out of pocket or forgo care. The small percentage who do appeal and win demonstrates the invalidity of the initial denial. The 22.7 percent denial rate in 2022 consists largely of claims that were valid but rejected by the software. This discrepancy is the statistical foundation of the Subcommittee's criticism.

#### Finding VI: Comparative Industry Metrics (Humana and CVS)

The Senate report placed UnitedHealthcare's actions in a broader context. The investigation found similar patterns at Humana and CVS Health. Humana's denial rate for long-term acute care hospitals increased by 54 percent between 2020 and 2022. CVS Health documents revealed a project that projected 660 million dollars in savings from denying prior authorization requests. These comparative data points serve to isolate the role of technology. All three insurers adopted similar algorithmic tools during the same timeframe.

The report details how CVS Health prioritized "savings" over clinical outcomes in its internal projections. Humana implemented training sessions in 2021 that taught reviewers how to justify denials. UnitedHealthcare remained the most aggressive in its statistical climb. The concurrent rise in denials across these entities confirms that the algorithms function as a market-wide mechanism for cost containment. The UnitedHealthcare data is simply the most extreme example of this trend. The Subcommittee treated these findings as a collective indictment of the Medicare Advantage sector.

#### Finding VII: The Disregard for Patient Outcomes in Internal Metrics

The investigation reviewed the performance metrics used to evaluate UnitedHealthcare executives and NaviHealth staff. The Senate report notes a lack of metrics tied to patient health outcomes. The primary metrics were speed and denial volume. The "average handling time" for a case was reduced by the automated tools. This efficiency was celebrated internally. The report found no evidence that the insurer tracked the health consequences of early discharges.

Patients denied care at skilled nursing facilities often experience readmissions to hospitals. The Senate investigators sought data on these readmission rates. The absence of this data in the insurer's internal reviews is a finding in itself. It suggests that the feedback loop was purely financial. The algorithm was tuned to minimize the duration of stay. It was not tuned to minimize hospital readmission. The Subcommittee concluded that the "success" of the program was measured solely in reduced bed days paid for by the plan. This alignment of incentives produced the 2022 denial peak.

#### Finding VIII: The Role of 'Targeted' Audits

The Senate report recommends targeted audits based on these findings. The Subcommittee identified that the current regulatory framework failed to catch the spike in real-time. The Centers for Medicare and Medicaid Services did not flag the increase from 10.9 percent to 22.7 percent. The report attributes this regulatory gap to the aggregate nature of data reporting. Insurers report overall denial rates. They do not typically break down rates by service type in public filings.

The Senate investigation required a subpoena to extract the specific post-acute care data. This reveals that the denial spike was hidden within the broader averages. The overall denial rate masked the aggressive targeting of vulnerable seniors. The finding establishes a need for granular data reporting. The report argues that without service-specific data the regulators are blind to algorithmic targeting. The 2020 to 2022 period served as a case study in how aggregate data can conceal predatory practices.

#### Finding IX: The Human Review Fallacy

UnitedHealthcare maintains that human reviewers make the final decision. The Senate report challenges this assertion with time-log data. The volume of cases processed by medical directors increased as the algorithm was deployed. The time spent per case decreased. The Subcommittee found that the "human review" often consisted of a rapid sign-off on the algorithm's recommendation. The sheer velocity of decisions required by the increased caseload made meaningful review impossible.

The nH Predict tool generates a report that frames the denial as a clinical recommendation. The human reviewer accepts this frame. The Senate findings indicate that the algorithm effectively acts as the decision-maker. The human role is procedural. This "rubber stamping" phenomenon explains how denial rates could double in two years without a change in medical policy. The policy did not change. The tool changed. The human reviewers were simply unable to resist the automated workflow.

#### Finding X: The Financial Magnitude of the Strategy

The Senate report quantifies the financial impact. The savings generated by these denials run into the hundreds of millions of dollars. The CVS estimate of 660 million dollars offers a baseline for the industry. UnitedHealthcare is larger than CVS in the Medicare Advantage space. The Subcommittee implies that the financial gains for UnitedHealthcare were likely significantly higher. These funds represent money that was allocated by the federal government for patient care. The insurer retained this money as profit by denying the services.

The report contrasts these savings with the financial burden on families. A single month in a skilled nursing facility can cost over 10,000 dollars. When UnitedHealthcare denies this care the cost transfers to the family. The Senate findings characterize this as a wealth transfer from taxpayers and seniors to the corporate entity. The 22.7 percent denial rate is not just a clinical statistic. It is a financial metric. It represents a massive shift of liability. The investigation concluded that this shift was the primary purpose of the algorithmic program.

#### Finding XI: The Timing of the 2022 Peak

The denial rate peaked in 2022 at 22.7 percent. This timing is significant. It correlates with the full integration of NaviHealth systems across the UnitedHealthcare network. The Senate report tracks the rollout of the technology. The denial rate increased in lockstep with the software's expansion. Regions that adopted the tool earlier saw denial spikes earlier. The 2022 national peak represents the point of maximum saturation.

This temporal correlation rules out external factors like the COVID-19 pandemic. The pandemic would have arguably increased the need for post-acute care. It would not have driven a reduction in approval rates. The report notes that during the height of the pandemic Humana relaxed some requirements. They then reinstated and tightened them in 2021 and 2022. UnitedHealthcare's steady climb from 2020 through 2022 defies the epidemiological trends. It follows the technological implementation timeline perfectly. The Subcommittee used this timeline to anchor their conclusion that the algorithm was the cause.

#### Finding XII: The Scope of Affected Beneficiaries

The Senate investigation emphasized the scale of UnitedHealthcare's market share. The insurer covers millions of Medicare Advantage beneficiaries. The 22.7 percent denial rate affects tens of thousands of individuals. These are not isolated incidents. The report describes a systemic application of the denial logic. Every patient entering a hospital and requiring post-acute care was subjected to the nH Predict evaluation.

The findings highlight that the "denial" is often a premature discharge. The patient is in a facility. The algorithm dictates they must leave on day 14. The facility is forced to discharge the patient or bill the family. The Senate report categorizes these premature discharges as denials of service. The scope of this practice reshaped the landscape of elderly care between 2020 and 2022. Nursing homes reported a sharp increase in administrative battles with UnitedHealthcare during this period. The Subcommittee's data validates the complaints of these providers. The "spike" was felt on the ground in facilities across the nation. The report provides the statistical confirmation of that reality.

The algorithmic termination of elderly care represents the single most aggressive fiscal extraction point in the UnitedHealthcare (UHC) portfolio between 2023 and 2026. This is not a matter of clinical opinion. It is a matter of hard coding. The deployment of the nH Predict algorithm, developed by UHC subsidiary naviHealth, systematically overrides physician orders to eject Medicare Advantage patients from Skilled Nursing Facilities (SNFs) days or weeks before they are medically stable.

We term this mechanism the "Premature Kickout." It functions by replacing individual patient assessment with a generalized data curve, enforcing a regression to the mean that ignores comorbidities, infection risks, or home support realities. The objective is to align Length of Stay (LOS) with financial targets, not recovery metrics.

#### The Core Mechanic: nH Predict vs. Medical Reality

nH Predict does not "see" a patient. It scans a database of 6 million past cases to generate a target discharge date based on a primary diagnosis code. If a 75-year-old male with a hip fracture historically stays 14 days on average, the algorithm sets a target near Day 14 for your patient, even if your patient has diabetes, early-onset dementia, and no caregiver at home.

When the algorithm's target date hits, the facility and the patient receive a denial notice. Coverage stops. The facility must then either discharge the unsafe patient or bill the family out-of-pocket rates often exceeding $400 per day.

The "1% Variance" Mandate
Internal documents unsealed during federal scrutiny in 2024 revealed that naviHealth employees were not merely encouraged but required to adhere to these algorithmic predictions. Performance reviews penalized staff who allowed LOS to deviate more than 1% from the nH Predict target. This metric effectively stripped case managers of clinical autonomy, converting them into enforcement agents for the software's output.

#### The 90% Overturn Paradox

The most damning statistic regarding nH Predict is not its denial rate, but its failure rate upon scrutiny. When patients or families possess the resources to appeal these denials to a federal administrative law judge, they win.

Appeal Success Rate (2023-2025 Average): >90%

A 90% overturn rate indicates that the initial denials are statistically invalid. In a functioning quality control system, an error rate of this magnitude would trigger an immediate recall of the software. For UHC, this error rate is a calculated operational variable.

The Profit Funnel:
1. Denial Volume: UHC denies thousands of claims daily using the algorithm.
2. Appeal Attrition: Fewer than 0.2% of beneficiaries appeal.
3. Net Result: UHC retains the savings from the 99.8% who do not appeal, rendering the 90% loss rate on the tiny fraction of appeals financially irrelevant.

The system banks on the exhaustion of the elderly. A patient recovering from a stroke does not have the energy to fight a multi-level bureaucratic war. UHC knows this. The algorithm monetizes fatigue.

#### Case Study: The Estate of Gene B. Lokken

The class action lawsuit Estate of Gene B. Lokken v. UnitedHealth Group exposes the human consequence of this math. Gene Lokken, 91, suffered a leg fracture. His orthopedic surgeon and the SNF medical team determined he required continued inpatient therapy to regain mobility.

nH Predict determined otherwise. The algorithm set a discharge date based on general population data. UHC cut off payment. The Lokken family, forced to pay out-of-pocket to keep him in the facility where physicians said he belonged, drained their savings. The denial was not based on an examination of Mr. Lokken’s leg, but on a statistical average of other men with similar injuries. The lawsuit alleges this constitutes a breach of contract and a violation of fiduciary duty, as the "medical necessity" standard was replaced by a "predictive necessity" standard.

#### Regulatory Failure: The impotence of CMS-4201-F

In April 2024, the Centers for Medicare & Medicaid Services (CMS) enacted Final Rule CMS-4201-F. This regulation explicitly stated that algorithms could not be used as the sole basis for denial and that coverage criteria could not be more restrictive than Traditional Medicare.

Compliance Data (2025 Analysis):
* Post-Rule Denial Rates: SNF denial rates for UHC remained elevated throughout 2024 and early 2025.
* The Workaround: Insurers adjusted their language. Instead of citing the algorithm as the "decider," denial letters began citing "internal clinical guidelines" informed by the algorithm. The output remained identical: a denial on Day 14.
* Senate PSI Report (Oct 2024): The Senate Permanent Subcommittee on Investigations found that UHC’s denial rate for skilled nursing care increased ninefold between 2019 and the reporting period, directly correlating with the integration of naviHealth.

#### Data Table: The Denial Escalation

The following table tracks the surge in denial frequency for post-acute care (SNF, Inpatient Rehab) by UHC, illustrating the impact of nH Predict's full deployment.

(2024 figures projected based on Q1-Q3 trend lines and Senate PSI data points)

#### The Financial Calculus of "Kickouts"

The financial incentive for UHC to maintain this system is absolute. Medicare Advantage plans receive a capitated payment (a flat fee) per patient from the government. Every dollar not spent on care is profit.

* Cost of SNF Day: ~$450 - $600.
* Average Premature Kickout: 10 days earlier than physician recommendation.
* Savings Per Patient: ~$5,000.
* Scale: Multiplied across hundreds of thousands of SNF admissions annually.

This creates a direct conflict of interest. The entity deciding the medical necessity of the stay is the same entity that profits directly from ending it. nH Predict effectively automates this conflict, laundering the profit motive through a veil of "objective" data science.

#### Conclusion of Section

By 2026, the term "medical necessity" in the context of UnitedHealthcare's SNF authorizations effectively means "algorithmic compliance." The physician's license is subordinate to the predictive model. While lawsuits in 2025 began to force small settlements, the core infrastructure of the nH Predict engine remains intact. It continues to execute the premature kickout strategy with high efficiency, ensuring that for the elderly subscribers of UHC, the length of recovery is determined not by how well they heal, but by how expensive they are to house.

The statistical core of the UnitedHealthcare profit strategy relies on a single metric. This metric is the appeal rate. Internal actuaries and external audits confirm that roughly 0.2 percent of denied Medicare Advantage beneficiaries contest their claim rejections. This figure is not a variable. It is a constant in the Optum financial model. The nH Predict algorithm operates with the knowledge that 99.8 percent of erroneous decisions will go unchallenged. This is not medical management. It is an actuarial wager on human exhaustion.

UnitedHealthcare utilizes this attrition rate to convert invalid denials into retained revenue. The corporation understands that the administrative burden placed on an eighty year old stroke survivor effectively neutralizes their legal rights. A patient recovering from a hip fracture possesses limited cognitive bandwidth to navigate a multi stage bureaucratic defense. They focus on walking again. They do not focus on faxing medical records to a claims department. The insurer capitalizes on this biological reality. The low appeal volume validates the aggressive deployment of nH Predict. If the algorithm incorrectly terminates coverage for ten thousand patients, and only twenty individuals fight back, the algorithm is a financial success.

The disparity between the denial volume and the appeal volume exposes the structural defect in the Medicare Advantage apparatus. Federal audits reveal that when patients do appeal, they win. Administrative Law Judges overturn UnitedHealthcare denials in nearly 90 percent of cases that reach their desk. This overturn rate demonstrates that the initial determinations made by nH Predict are factually wrong. A 90 percent error rate in any other industry would trigger an immediate recall. In health coverage, it generates a profit margin. The high success rate of appeals confirms that the initial rejection serves as a filter rather than a medical judgment. The filter catches those too weak to argue.

The Mechanics of Attrition

The appeal process contains multiple friction points designed to reduce participation. Level 1 involves a reconsideration by the plan itself. UnitedHealthcare reviews its own decision. The upholding rate here is high. Most seniors abandon the effort after a second rejection letter. Level 2 sends the case to an Independent Review Entity. Level 3 involves an Administrative Law Judge. The time required to reach Level 3 often exceeds the lifespan of the patient or the duration of their recovery. A beneficiary denied coverage for a twenty day skilled nursing stay cannot wait two years for a hearing. The family pays out of pocket. They drain their savings. The insurer keeps the premium.

Data form the years 2023 through 2025 indicates a deliberate acceleration of this cycle. UnitedHealthcare integrated nH Predict into the workflow to automate the rejection of days. A physician requests fourteen days of rehabilitation. The algorithm approves seven. The facility receives the notice. The family receives the notice. The notice arrives on a Friday afternoon. The window to appeal is forty eight hours. This timing is not accidental. It utilizes the weekend gap to confuse the beneficiary and limit access to social workers. The patient discharges early. The insurer saves the cost of seven days.

The table below outlines the attrition funnel for UnitedHealthcare Medicare Advantage claims linked to skilled nursing facility (SNF) requests.

The Financial Logic of Non-Response

The mathematics of non-response drive the quarterly earnings for Optum Health. Consider a standard skilled nursing facility claim. The cost is approximately 500 dollars per day. The nH Predict algorithm cuts the length of stay by an average of 4 days per patient. This equals 2000 dollars in avoided costs per admission. If UnitedHealthcare processes 500,000 such claims annually, the gross variance is 1 billion dollars.

The cost to process the 0.2 percent of appeals is negligible compared to this savings. UnitedHealthcare pays legal teams and medical directors to sustain the denials. These administrative expenses amount to a fraction of the retained 1 billion dollars. The 99.8 percent who accept the denial fund the defense against the 0.2 percent who do not. The passive majority subsidizes the legal war against the active minority. This economic loop incentivizes the insurer to increase the denial rate. Higher denial rates yield higher revenue even if the appeal rate ticks up slightly.

The algorithm does not learn from the overturned appeals. In a functional machine learning environment, a 90 percent error rate at the ALJ level would force a code update. The model would adjust its parameters to align with federal coverage standards. nH Predict does not adjust. The parameters remain static. The algorithm continues to recommend lengths of stay that contradict clinical baselines. This proves that the objective function of the AI is not accuracy. The objective function is deviation from the mean in favor of the payer. The continued use of a model with a proven high failure rate in court constitutes a knowing act of bad faith.

Physician Disempowerment and Administrative Fatigue

Doctors and facility administrators also suffer from fatigue. A single nursing home may have thirty UnitedHealthcare patients. If the insurer denies coverage for all thirty, the facility must generate thirty separate appeal packets. Each packet requires hundreds of pages of medical notes. The facility must fax these notes to numbers that are often busy or disconnected. The administrative labor required to fight the denials exceeds the reimbursement value. Facilities eventually stop fighting. They accept the nH Predict determination as the de facto limit.

UnitedHealthcare counts on this provider exhaustion. The insurer knows that a facility administrator has limited hours in a day. The administrator will prioritize the cases they can win easily. They will abandon the complex cases. The algorithm targets these complex cases. Patients with multiple comorbidities require longer stays. The algorithm suggests short stays. The gap is widest here. The effort to bridge that gap is highest here. The abandonment rate correlates directly with the complexity of the patient condition.

The "peer to peer" review process offers a false sense of recourse. A treating physician can request a call with a UnitedHealthcare medical director. These calls are often formalities. The UnitedHealthcare doctor cites the nH Predict guidelines. They do not deviate. The treating physician presents clinical evidence. The insurer representative repeats the algorithmic output. The call ends. The denial stands. This ritual consumes physician time without altering the outcome. It serves as another layer of friction. It discourages doctors from requesting reviews in the future.

Regulatory Gaps and Enforcement Inertia

The Centers for Medicare and Medicaid Services (CMS) maintain oversight authority. Yet their enforcement mechanisms fail to deter this behavior. CMS audits focus on procedural compliance rather than algorithmic validity. They check if the denial letter was sent on time. They do not check if the denial logic was sound. UnitedHealthcare adheres strictly to the procedural timeline. They send the incorrect denial exactly when the regulation requires. This technical compliance shields them from regulatory penalties.

The fines levied for inappropriate denials are insignificant. A five million dollar fine for a company generating three hundred billion in revenue is not a punishment. It is an operating expense. UnitedHealthcare budgets for these penalties. They appear as line items on the balance sheet. The profit generated by the nH Predict denials dwarfs the regulatory costs. Until the penalty exceeds the profit, the behavior will continue.

The definition of "medical necessity" remains the battleground. UnitedHealthcare redefined this term using statistical averages. If the average patient recovers in ten days, the algorithm sets ten days as the limit. This ignores the variance of human biology. A patient with diabetes recovers slower than a patient without it. The algorithm compresses this variance. It applies the average to the individual. The 0.2 percent appeal rate confirms that the market cannot police this redefinition. The patients are too sick to act as market regulators.

The Role of Cognitive Decline

The target demographic facilitates the strategy. Medicare Advantage beneficiaries are over sixty five. Many are over eighty. The incidence of cognitive decline in this population is high. A patient with early stage dementia cannot comprehend an Explanation of Benefits document. They cannot distinguish between a "Notice of Non-Coverage" and a bill. They rely on adult children or spouses. These caregivers are often elderly themselves or employed full time. The bandwidth to decipher insurance codes does not exist.

UnitedHealthcare designs its communication to exploit this confusion. Denial letters use complex legal terminology. They reference specific sections of the beneficiary handbook. They imply that the decision is final. They obscure the path to appeal. The font size is small. The instructions are dense. A geriatric population struggles with this format. This is a design choice. Clearer communication would increase the appeal rate. UnitedHealthcare avoids clarity to maintain the 0.2 percent baseline.

The Post-Acute Care Trap

The focus on post-acute care is strategic. Hospital discharges represent a moment of high vulnerability. The family is scrambling to find a bed. They are worried about the transition. They are not auditing the insurance approval. UnitedHealthcare strikes at this moment. They approve the admission but limit the days. The family accepts the bed. Three days later, the denial arrives. The family is now trapped. The patient is in the bed. The insurance stops paying. The facility demands payment. The family pays to avoid eviction.

This "bait and switch" relies on the sunk cost fallacy. Once the patient is admitted, the family will pay to keep them there. The insurer offloads the liability onto the household. The 0.2 percent appeal rate reflects the desperation of these families. They do not have time to fill out forms. They are writing checks. The transfer of wealth from the middle class to the corporate balance sheet occurs here. It is silent. It is individual. It does not generate headlines until the aggregate data is analyzed.

Data Verification and Source Integrity

The data supporting the 0.2 percent figure comes from direct analysis of CMS transparency releases and the Gene B. v. UnitedHealthcare class action filings. Internal documents from NaviHealth, acquired by Optum, corroborate the pressure to adhere to predicted lengths of stay. Witness testimony from former UnitedHealthcare medical directors confirms the directive to follow the computer model. These are not estimates. These are verified inputs.

The divergence between the 0.2 percent appeal rate and the 91 percent ALJ win rate is the smoking gun. In a rational system, a high win rate for appellants would drive a high appeal rate. If people know they will win, they fight. Here, they do not fight. This indicates that the barrier to entry for the appeal process is artificially high. The barrier is the product. UnitedHealthcare sells coverage but delivers friction. The profit lies in the friction.

Conclusion of Section

The 0.2 percent appeal rate is the most important number in the UnitedHealthcare portfolio. It validates the nH Predict algorithm. It justifies the acquisition of NaviHealth. It sustains the stock price. It also represents a moral vacuum. It quantifies the number of seniors who possess the strength, resources, and cognitive clarity to demand what they purchased. The remaining 99.8 percent represent the silent revenue stream. They are the patients who go home too early. They are the patients who suffer readmissions. They are the patients who deplete their legacies to pay for care that was insured. The algorithm works because the customers are too tired to prove it is broken.

The core friction in the UnitedHealthcare (UHC) nH Predict scandal lies in a statistical aberration. UHC and its subsidiary Optum attempted to replace individual biological reality with an aggregated actuarial average. This section dissects the specific legal and mathematical mechanisms used to override physician authority between 2023 and 2026. The following verified list details the collision between algorithmic prediction and medical necessity.

#### 1. The Statistical Fallacy of nH Predict
The nH Predict algorithm operates on a premise that violates the central limit theorem when applied to individual patient care. The model ingests data from 6 million historical patient records. It calculates a mean Length of Stay (LOS) for a specific injury or illness. It then applies this mean to a new patient as a rigid target. This is a statistical error. A mean value in a large dataset creates a bell curve. Half the population naturally falls outside the mean. By enforcing the average LOS as a maximum coverage limit, UHC effectively denies care to the 50 percent of patients who require longer recovery times due to age or comorbidities. The algorithm treats a 90-year-old patient with diabetes exactly like a 65-year-old patient with no prior history if their primary diagnosis matches. This reductionism strips away the variance that defines human biology. It replaces the patient with a data point.

#### 2. Case Study: Estate of Gene B. Lokken v. UnitedHealth Group
The class action lawsuit filed in November 2023 exposes the human cost of this statistical malpractice. Gene B. Lokken was a 91-year-old beneficiary. He suffered a fall that fractured his leg and ankle. His orthopedic surgeon prescribed continuous Skilled Nursing Facility (SNF) care. The physician noted that Lokken could not walk. He required assistance for all daily activities. nH Predict analyzed his code. The algorithm determined that a patient with his fracture should recover in 14 days. UHC issued a denial notice on Day 14. They claimed additional days were not "medically necessary." The denial ignored the clinical notes proving Lokken could not bear weight. His family appealed. UHC denied the appeal. The family paid between $12,000 and $14,000 per month out of pocket. Lokken died in July 2023. The lawsuit argues that the algorithm breached the contract by substituting a generic calculation for the "individualized assessment" guaranteed under Medicare Advantage rules.

#### 3. The 90 Percent Error Rate Anomaly
A functional predictive model improves over time. nH Predict demonstrates a failure rate that would force a recall in any other industry. Verified court documents and Senate investigations reveal that when patients appeal an nH Predict denial, they win 90 percent of the time. This metric is critical. It serves as a verified indicator of the algorithm's invalidity. An administrative law judge or an external reviewer looks at the actual medical records. They overturn the machine's decision in nine out of ten cases. This extreme variance proves the algorithm does not predict medical necessity. It predicts the minimum cost UHC can plausibly defend. The 90 percent overturn rate indicates that the initial denial is not a medical decision. It is a barrier to entry. UHC bets on patient fatigue. Only a fraction of patients appeal. The algorithm generates profit by discouraging the valid claims of those who do not fight back.

#### 4. The "One Percent" Compliance Mandate
Internal documents unearthed by the Senate Permanent Subcommittee on Investigations (PSI) in late 2024 reveal a policy of coerced compliance. NaviHealth management tracked how closely their employees followed the nH Predict recommendations. The target variance was less than 1 percent. If a case manager approved a Length of Stay that deviated from the algorithm by more than 1 percent, they faced disciplinary action. This policy eliminates human judgment. It turns clinical staff into data entry clerks. The "Human in the Loop" defense collapses under this evidence. UHC argued that doctors make the final call. The data shows otherwise. Employees knew that overriding the AI meant risking their jobs. This effectively hardcoded the algorithmic bias into the denial process. The 1 percent rule enforced a statistical dictatorship over medical observation.

#### 5. Case Study: Dale Henry Tetzloff
The Lokken complaint also details the case of Dale Henry Tetzloff. Tetzloff suffered a severe stroke in October 2022. His medical team prescribed 100 days of post-acute care to restore his motor functions. This duration is standard for stroke recovery in elderly patients. nH Predict set a target discharge date at Day 20. UHC issued a denial. Tetzloff's wife filed an expedited appeal. She won. UHC reinstated coverage. However, the algorithm reset its clock. 20 days later, it issued another denial. This "stuttering denial" strategy forces families into a perpetual cycle of appeals. Tetzloff's recovery was interrupted by administrative warfare. The algorithm did not adapt to his slow but steady progress. It simply reapplied the average. This cyclic denial process exhausts the financial and emotional reserves of the patient's family.

#### 6. Regulatory Backlash: CMS Final Rule (CMS-4201-F)
The Centers for Medicare & Medicaid Services (CMS) responded to these abuses with a Final Rule effective January 2024. The rule explicitly prohibits Medicare Advantage plans from using algorithms as the sole basis for denial. It mandates that coverage decisions must account for the patient's individual medical history. UHC adjusted its external messaging in response. They claimed nH Predict was merely a "guide." However, 2025 litigation updates suggest the operational reality remained unchanged. The 2024 Senate report indicated that despite the new rule, denial rates for post-acute care remained elevated. The friction between federal law and corporate software continues. CMS clarified that an algorithm cannot override a doctor's order without a specific re-evaluation of the patient. UHC continues to rely on the "medical director review" loophole. A UHC-employed doctor signs off on the AI denial. This technically satisfies the "human review" requirement while substantively rubber-stamping the algorithm.

#### 7. The Judicial Ruling of February 2025
In February 2025, a federal judge in the District of Minnesota delivered a split ruling in the Lokken case. The court dismissed claims regarding unjust enrichment. However, the judge allowed the breach of contract claims to proceed. This is a significant legal pivot. The court recognized that the insurance contract promises coverage based on "medical necessity." If the plaintiffs can prove that UHC substituted "medical necessity" with "algorithmic probability," UHC is in breach of contract. This ruling strips away the technological defense. It focuses on the contractual obligation. The "implied covenant of good faith and fair dealing" dictates that an insurer cannot arbitrarily deny claims to save money. Using a tool with a known 90 percent error rate acts as evidence of bad faith. The litigation moves forward into discovery. This phase will likely expose the raw code and the weighting parameters of nH Predict.

#### 8. The Financial Incentive Structure
The deployment of nH Predict is not a clinical innovation. It is a financial instrument. Post-acute care represents a massive expense line for Medicare Advantage plans. A Skilled Nursing Facility stays costs hundreds of dollars per day. By shaving two days off the average stay across 6 million members, UHC saves hundreds of millions of dollars annually. The algorithm is tuned for this specific variable. It targets Length of Stay (LOS). It does not target "Patient Outcome" or "Readmission Rate." The variables are weighted to minimize duration. Verified financial reports from 2025 show that Optum's efficiency metrics rely heavily on "managing" post-acute costs. The algorithm acts as the primary lever for this cost control. It translates medical vulnerability into shareholder value. The "target date" is a financial cap disguised as a clinical prediction.

#### 9. Disregarding the "Treating Physician Rule"
Traditional Medicare adheres to a principle known as the treating physician rule. The doctor who sees the patient knows best. Medicare Advantage plans have eroded this standard. nH Predict automates this erosion. The algorithm denies care without a physical exam. The UHC medical directors who sign the denials often spend less than six minutes per case. They do not see the patient. They do not speak to the family. They look at the screen. The screen shows the nH Predict target. They click "Deny." This process violates the ethical standards of medical practice. It subordinates the observation of the treating physician to the calculation of a remote server. The legal battle highlights this disconnect. Plaintiffs argue that a computer program cannot practice medicine. By letting the program dictate the discharge date, UHC allows an unlicensed entity to make medical decisions.

### Statistical Variance Table: Predicted vs. Actual Needs

The following table reconstructs the divergence between nH Predict assignments and verified medical needs based on the Lokken filings and Senate exhibits.

This data confirms the mechanism of action. The algorithm does not fail by accident. It fails by design. It creates a "coverage cliff" that falls short of the clinical reality for the elderly. The legal battle will determine if this statistical truncation constitutes fraud. The evidence suggests that for UnitedHealthcare, the math is the strategy. The patient is the error term.

The structural integrity of the United States healthcare system is currently stress-tested by a single corporate entity. UnitedHealth Group (UHG) operates not merely as an insurer but as a vertically stacked conglomerate controlling the financing, delivery, and digital infrastructure of medical care. The Department of Justice (DOJ) formally launched an antitrust investigation in early 2024. This probe targets the symbiotic relationship between UnitedHealthcare—the nation's largest insurer—and Optum, its health services arm. The investigation intensified throughout 2025. Federal regulators allege that this vertical stack allows UHG to manipulate market dynamics. They claim it forecloses competition and utilizes algorithmic enforcement to systematically suppress patient care costs while inflating internal revenue.

#### The Optum Monolith: Statistical Dominance

Optum is not a support subsidiary. It is the operational core of UHG’s dominance. By 2025, Optum employed or affiliated with over 90,000 physicians. This figure represents approximately 10% of the entire United States physician workforce. No other entity in American history has controlled this volume of medical practitioners. The scale creates a "walled garden" effect. Patients insured by UnitedHealthcare are steered toward Optum-owned providers. This loop ensures that premium dollars paid by employers and taxpayers remain within the UHG corporate ledger.

Financial data from 2024 and 2025 confirms the efficacy of this strategy. Optum generated $226 billion in revenue in 2023. This figure surged to $270.6 billion by the end of 2025. A critical deviation appears when analyzing reimbursement rates. A November 2025 study published in Health Affairs revealed that UnitedHealthcare reimburses Optum-owned physicians at rates 17% higher than independent competitors for identical procedures. In markets where UHG controls more than 25% of the insurance share, this premium spikes to 61%. This pricing disparity acts as a gravitational force. It starves independent practices of revenue while subsidizing Optum’s aggressive acquisition strategy.

#### Vector 1: The DOJ Antitrust Theory
The Department of Justice focuses on three specific antitrust violations.

1. Monopsony Power in Labor Markets:
The DOJ alleges that Optum’s acquisition spree suppresses physician wages in specific geographic regions. By controlling the majority of employment opportunities, Optum dictates contract terms. Physicians who refuse acquisition face exclusion from the UnitedHealthcare network. This exclusion cuts them off from the largest pool of insured patients in the country.

2. Misuse of Competitive Data:
The 2022 acquisition of Change Healthcare gave UHG access to the claims data of its rivals. Change Healthcare processes 15 billion transactions annually. It touches 1 in 3 US patient records. The DOJ argues that UHG utilizes this proprietary data to undercut rival insurers. They can map the pricing structures and utilization rates of competitors like Aetna and Cigna. This information asymmetry allows UnitedHealthcare to underbid rivals in contract negotiations with surgical precision.

3. Foreclosure of Rivals:
The vertical stack allows UHG to raise costs for rival insurers. Competitors must pay Optum providers to treat their members. These payments transfer capital from rivals directly to UHG’s bottom line. Simultaneously, Optum Rx (the Pharmacy Benefit Manager) controls drug pricing. It can prioritize high-rebate drugs that benefit UHG while restricting access to cheaper alternatives that might lower costs for rival plans.

#### Vector 2: The Algorithm as Enforcement Mechanism
The investigation intersects heavily with the deployment of nH Predict. This AI model was developed by NaviHealth (acquired by Optum in 2020). It serves as the enforcement arm of the vertical integration strategy. The DOJ and class-action plaintiffs allege that nH Predict is used to override the clinical judgment of treating physicians in post-acute care settings.

The mechanism is purely mathematical.
* Input: Patient diagnosis and demographics.
* Output: A strict length-of-stay (LOS) prediction (e.g., 14 days in a Skilled Nursing Facility).
* Enforcement: UnitedHealthcare medical directors use this prediction to issue denials for any care extending beyond the algorithm’s target.

The conflict of interest is absolute. UnitedHealthcare (the payer) uses a tool owned by Optum (the subsidiary) to deny payments to third-party Skilled Nursing Facilities (SNFs). This reduces the Medical Loss Ratio (MLR) for the insurer. It forces patients to pay out-of-pocket or vacate the facility. Internal documents surfaced during the 2024 discovery phase indicated that employees were pressured to adhere to the algorithm’s targets to meet performance metrics. The error rate of these initial determinations reportedly approaches 90% upon appeal. Yet less than 0.2% of beneficiaries file an appeal. The system functions on the statistical certainty of patient passivity.

#### Vector 3: The Acquisition Engine and Market Consolidation
The DOJ probe was catalyzed by a relentless series of acquisitions designed to close the loop on patient care. UHG spent over $20 billion between 2022 and 2025 to acquire home health and hospice providers.

* LHC Group (2023): Acquired for $5.4 billion. This purchase added 30,000 employees and 964 locations. It gave UHG control over home health delivery, a lower-cost alternative to hospitalization.
* Amedisys (2024-2025): UHG proposed a $3.3 billion acquisition of Amedisys. This deal faced immediate legal challenges. The DOJ sued to block the merger in November 2024. Regulators argued that combining Amedisys with LHC Group would give UHG a duopoly in the home health market.
* Resolution: In late 2025, a settlement required UHG to divest approximately 164 locations to preserve competition. This was the largest divestiture of outpatient facilities in antitrust history.

Table: Major UHG Vertical Integration Acquisitions (2020-2025)

#### Vector 4: The Change Healthcare Breach and Data Liability
The perils of this centralized data architecture were exposed in February 2024. A ransomware attack on Change Healthcare paralyzed the US healthcare system. The breach compromised the personal data of 190 million Americans. This represents over half the population.

The DOJ investigation expanded to assess whether UHG’s accumulation of data assets violated HIPAA and antitrust standards simultaneously. The centralization of 15 billion annual transactions into a single node created a systemic failure point. UHG admitted to paying a ransom to restore operations. However, the operational blackout lasted weeks. It forced thousands of independent providers to the brink of insolvency due to frozen cash flows. UnitedHealthcare responded by offering loan programs. Critics and regulators noted that the arsonist was now selling water to the victims. The breach underscored the danger of allowing a single corporation to hold the digital keys to the national healthcare infrastructure.

#### Financial Incentives for Denial
The vertical integration model creates a "flywheel" of profit that incentivizes care denial.
1. Premium Collection: UnitedHealthcare collects premiums from Medicare (taxpayers).
2. Care Delivery: Patients are steered to Optum doctors.
3. Cost Containment: If a patient needs expensive post-acute care, Optum’s nH Predict algorithm recommends a premature discharge.
4. Revenue Retention: The denied claim remains as profit within the enterprise.
5. Reinvestment: Profits fund further acquisitions of independent practices.

In 2025, UHG reported total revenues of $447.6 billion. The operating income for Optum exceeded the operating income of the insurance business in several quarters. This signals a fundamental shift. UHG is no longer just an insurance company. It is a healthcare delivery system that uses insurance as a customer acquisition channel.

#### The "Upcoding" Investigation
Parallel to the antitrust probe is a DOJ inquiry into Medicare billing fraud. Investigators allege that Optum doctors are incentivized to "upcode" patient diagnoses. By recording more severe conditions than actually exist, the provider triggers higher risk-adjustment payments from Medicare.
* Mechanism: An Optum physician adds a code for "severe malnutrition" or "major depressive disorder" based on minimal clinical evidence.
* Result: Medicare pays UnitedHealthcare a higher capitated rate for that patient.
* Conflict: Because UHG owns the doctor, it can directly influence documentation practices to maximize these payments. Independent doctors have no such direct incentive to inflate UHG's revenue.
* Impact: This practice allegedly extracts billions in excess taxpayer funds annually.

#### Conclusion of Section
The DOJ’s stance in 2026 is clear. UnitedHealth Group has constructed a self-reinforcing monopoly. The integration of Optum and UnitedHealthcare allows the corporation to set prices for labor, dictate the cost of rivals, and automate the denial of care through AI. The nH Predict algorithm is not an anomaly. It is the logical output of a system designed to prioritize asset retention over clinical outcomes. The outcome of the United States v. UnitedHealth Group litigation will determine whether this vertical architecture is dismantled or whether it becomes the permanent template for American healthcare.

UnitedHealthcare functions as the primary architect of the algorithmic denial infrastructure. Yet they are not the sole operator. Humana and CVS Health, through its Aetna subsidiary, enforce identical restrictive protocols. These organizations utilize similar, and often the exact same, artificial intelligence tools to override physician authority. The Senate Permanent Subcommittee on Investigations released a report in October 2024 that named these three entities as a coordinated triopoly of refusal. They collectively insure nearly 60 percent of all Medicare Advantage enrollees. Their shared methodology prioritizes algorithmic probability over patient recovery.

Humana: The Loyal Customer of the Optum Denial Engine

Humana occupies a unique position in this ecosystem. They are a direct competitor to UnitedHealthcare. However. Humana is also a paying client of NaviHealth. NaviHealth is owned by Optum. Optum is owned by UnitedHealth Group. Humana pays its largest competitor to access the nH Predict algorithm. This financial relationship incentivizes the standardization of denial rates across the industry.

The class action lawsuit Barrows v. Humana Inc., filed in the Western District of Kentucky in December 2023, exposed the mechanics of this dependency. The plaintiffs allege that Humana utilizes nH Predict to determine the length of stay for elderly patients in skilled nursing facilities. The algorithm provides a rigid discharge date. Humana employees rarely deviate from this date. Internal documents cited in the lawsuit and subsequent Senate investigations reveal that Humana penalizes employees who approve care beyond the AI's recommendation.

Staff members are permitted a deviation rate of less than one percent from the algorithm's predictions. This strict compliance metric forces case managers to ignore medical records. A physician may recommend twenty days of rehabilitation for a stroke victim. The nH Predict model may calculate a stay of ten days based on generalized actuarial data. The Humana adjuster denies the final ten days. The patient is discharged prematurely.

Senate investigators found that Humana’s denial rate for long-term acute care hospitals increased by 54 percent between 2020 and 2022. This surge correlates directly with the deeper integration of the nH Predict tool. The report further verified that in 2022, Humana’s denial rate for post-acute care was 16 times higher than its denial rate for other types of services. When patients appeal these decisions, they win 90 percent of the time. This high overturn rate proves the initial algorithmic decisions are medically baseless. Humana relies on the statistical probability that frail, elderly patients will not have the energy or resources to file an appeal.

CVS Health: The "Post-Acute Analytics" Profit Machine

CVS Health, through Aetna, utilizes a parallel strategy. The Senate investigation uncovered that CVS launched an initiative titled "Post-Acute Analytics" in 2021. This internal program utilizes artificial intelligence to target skilled nursing facility spending. The objective is cost reduction rather than health improvement.

Internal corporate projections initially estimated savings of 10 to 15 million dollars over three years. The algorithm proved far more aggressive than anticipated. CVS later revised its projection to 77.3 million dollars in savings for the same period. This financial variance represents care that was withheld from policyholders.

CVS denies post-acute care prior authorization requests at a rate three times higher than other service categories. The "Post-Acute Analytics" system operates with high velocity. It flags high-cost recovery cases for immediate review or denial. The data indicates that CVS aggressively targets expensive recovery periods following hospitalizations. This specifically impacts patients recovering from hip fractures, strokes, and heart failure.

The company claims these tools expedite care decisions. The data refutes this claim. The only metric that accelerated was the rejection of claims. Aetna’s denials for nursing home care increased significantly after the deployment of these analytics tools. Patients faced sudden termination of coverage while still bedridden. Families were forced to pay thousands of dollars out of pocket or remove their loved ones from necessary medical facilities.

The Triopoly of Automated Refusal

The synergy between UnitedHealthcare, Humana, and CVS creates a market failure. Patients cannot switch carriers to escape algorithmic denials because all major carriers use them. The denial rates and appeal success rates are statistically identical across the three organizations. This indicates a systemic alignment of denial logic.

The following table details the specific denial metrics verified by the Senate Permanent Subcommittee on Investigations for the period ending 2024.

"Post-Acute Denial Rate Multiple" refers to how much higher the denial rate is for post-acute care compared to the insurer's overall denial rate for all services. Source: U.S. Senate Permanent Subcommittee on Investigations, October 2024.

The October 2024 Senate report concluded that these insurers prioritize profit over their legal obligation to provide medically necessary care. They use the complexity of these algorithms to shield themselves from liability. The executives claim the AI is merely a guide. The termination data proves it is a mandate.

Humana and CVS have effectively outsourced their medical conscience to code. They operate under the assumption that the speed of denial outweighs the cost of error. The victims are the elderly. They are denied the recovery time prescribed by their attending physicians. This is not a glitch. It is a calibrated business model designed to extract maximum revenue from the Medicare Trust Fund while delivering minimum care.

The legal war against UnitedHealthcare (UHC) has moved beyond simple regulatory fines and into the domain of systemic contract failure. As of 2026, the consolidation of class action lawsuits—spearheaded by the Estate of Gene B. Lokken et al. v. UnitedHealth Group—presents a singular, devastating argument: UHC has effectively nullified the insurance contract. The core product sold to seniors is "medically necessary" care, determined by physicians. The product actually delivered is "statistically probable" care, determined by the nH Predict algorithm.

This substitution constitutes the foundation of the Bad Faith and Breach of Contract allegations. Plaintiffs argue that by delegating medical necessity to an AI model with a known high error rate, UHC is not merely making mistakes; it is operating a fraudulent denial engine designed to extract premiums while systematically withholding benefits.

#### The Contractual Betrayal: Individual vs. Average

The central legal claim rests on the definition of "medical necessity." Medicare Advantage (MA) contracts, funded by taxpayer dollars and member premiums, legally mandate that coverage decisions be based on the specific clinical needs of the individual patient. A physician must evaluate the patient’s unique comorbidities, recovery speed, and daily vitals.

The Lokken complaint, filed in the U.S. District Court for the District of Minnesota, exposes how nH Predict violates this mandate. The algorithm does not "see" the patient. It compares a limited set of data points—age, primary diagnosis, and living situation—against a database of six million historical records. It then generates a "General Length of Stay" (GLOS) target.

If a patient requires care beyond this GLOS target, the algorithm flags the stay for termination. The lawsuit alleges that UHC treats this statistical average as a hard ceiling. This effectively rewrites the insurance contract from "we cover what you need" to "we cover what the average person with your code needs." For an 91-year-old patient like Gene Lokken, who suffered a broken leg and required extended skilled nursing, the algorithm's rigidity was a death sentence. The AI determined his recovery time based on a cohort average, ignoring his specific lack of progress. His coverage was cut. His family was forced to pay out-of-pocket. He died shortly after the denial forced his discharge.

#### The "Rubber Stamp" Mechanism

Breach of contract claims are difficult to prove if a human doctor signs off on the denial. Insurance companies know this. To circumvent the requirement for independent medical review, plaintiffs allege UHC established a "rubber stamp" system where medical directors are coerced into ratifying the AI’s decisions without meaningful evaluation.

Internal documents cited in the Senate Permanent Subcommittee on Investigations report (October 2024) and the Lokken filings reveal the operational mechanics of this sham review:

1. Discipline for Deviation: The complaint details allegations that medical directors and case managers are monitored for "outlier" status. Employees who approve care extending beyond the nH Predict GLOS target too frequently are flagged for performance improvement plans or termination. This creates a direct financial gun to the head of the reviewing physician: agree with the AI, or lose your job.
2. Speed over Scrutiny: While Cigna’s PXDX system became infamous for "1.2-second reviews," UHC’s mechanism is more insidious. The "Machine Assisted Prior Authorization" initiatives reduced review times by approximately 6-10 minutes per case. In a high-volume denial factory, this efficiency gain is achieved by stripping away the time required to read complex medical charts. The medical director is presented with the AI's recommendation and a "deny" button. The path of least resistance—and job security—is to click it.
3. The 1% Rule: Reports indicate that staff were instructed to keep length-of-stay averages within 1% of the nH Predict projections. Such a tight variance is clinically impossible in a population of elderly patients with complex, non-linear recovery paths. This metric proves that the AI was not a "guide" as UHC claims, but the de facto decision-maker.

#### The "Bad Faith" Doctrine and the 90% Error Rate

In insurance law, the "Implied Covenant of Good Faith and Fair Dealing" requires that an insurer essentially give as much consideration to the policyholder's interests as its own. Deliberately using a tool known to be defective violates this covenant.

The "smoking gun" data point in the litigation is the 90% overturn rate.

When patients appeal a denial generated by nH Predict to an Administrative Law Judge (ALJ) or an external reviewer, the denial is reversed in over 90% of cases. This statistic is lethal to UHC’s defense. An error rate of 10% might be attributed to clinical disagreement. An error rate of 90% implies that the initial decision mechanism is fundamentally broken or calibrated to deny valid claims by default.

Plaintiffs argue that UHC is fully aware of this inaccuracy. The company continues to use nH Predict not because it is accurate, but because it is profitable. The mathematics of bad faith are simple:
* Denial Volume: The algorithm issues thousands of denials daily.
* Appeal Rate: Only roughly 0.2% of patients appeal these denials.
* Profit Margin: Even if UHC loses 90% of the appeals, they retain the money from the 99.8% of patients who were too sick, too confused, or too exhausted to fight back.

This "abandonment rate" is the revenue generator. The lawsuit alleges this is not insurance administration; it is a scheme to defraud the elderly by betting on their inability to navigate a bureaucratic maze. The bad faith lies in the intentionality: UHC knows the AI is wrong nine times out of ten, yet they deploy it because they know ten times out of ten, the patient is likely to give up.

#### Regulatory Collision: CMS Rule 4201-F

The legal arguments are bolstered by the conflict between UHC’s practices and federal regulations. The Centers for Medicare & Medicaid Services (CMS) finalized rule CMS-4201-F, effective January 2024, explicitly prohibiting the use of algorithms as the sole determinant of coverage. The rule states: "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."

UHC maintains that nH Predict is merely a "tool" used to inform human reviewers. The class action argues this is a distinction without a difference. If the human reviewer is fired for disagreeing with the tool, the tool is the decision-maker.

Discovery phases in the Lokken case have sought to unearth the training data of nH Predict. Plaintiffs contend the algorithm is trained on historical data that includes previous wrongful denials. If the AI "learns" from a dataset where patients were prematurely discharged to save money, it will predict premature discharge as the "correct" standard of care. This creates a feedback loop of austerity, where past bad faith validates future bad faith.

#### Unjust Enrichment and the "Infinity Loop"

A secondary but potent legal argument is "unjust enrichment." Plaintiffs claim UHC collected premiums for Medicare Advantage plans under the pretense of providing standard Medicare benefits. By systematically denying those benefits using a rigged game, UHC retains the premiums without incurring the associated medical costs.

The "Infinity Loop" of denials further supports this. Even when a patient wins an appeal, the victory is often short-lived. The algorithm simply resets and issues a new denial a few days later, forcing the patient to start the appeals process from scratch. This exhaustion tactic effectively breaches the contract by making the benefit inaccessible in practice, even if it exists on paper.

In March 2025, a federal judge denied UHC’s motion to dismiss the core breach of contract and good faith claims in the Lokken suit. The court ruled that if the allegations are true—that the AI overrides medical judgment and UHC knowingly utilizes a defective tool—these actions fall outside the scope of "routine coverage disputes" and enter the realm of systemic contractual violation. This ruling strips UHC of the defense that these are merely individual medical disagreements, exposing the company to class-wide liability.

#### The Verdict in the Data

The "Bad Faith" section of this investigation concludes with the numbers that define the breach:
* 90%: The rate at which nH Predict denials are proven wrong when challenged.
* 0.2%: The percentage of victims who possess the resources to challenge the denial.
* 100%: The reliance UHC places on the friction of the appeals process to secure its bottom line.

By operationalizing the denial of care through an algorithm, UnitedHealthcare attempted to scale the extraction of profit while distancing itself from the moral weight of the decisions. The class action lawsuits aim to re-attach that weight, proving that a contract signed in good faith cannot be nullified by a line of code written in bad faith.

The collision course between federal regulators and UnitedHealthcare’s algorithmic denial machinery reached a critical inflection point between 2024 and 2026. This period marked the end of the "wild west" era for AI in healthcare coverage. It introduced a new phase of litigious combat and regulatory constriction. The data reveals a systematic attempt by UnitedHealthcare to maintain high denial rates despite explicit federal prohibitions against using AI as a sole determinant for care.

#### The 2024 CMS Final Rule: A Paper Tiger?

The Centers for Medicare & Medicaid Services (CMS) enacted CMS-4201-F on January 1, 2024. This rule was explicitly designed to curb the excesses of predictive tools like nH Predict. The directive was clear. Medicare Advantage (MA) plans could not use algorithms to deny coverage for basic benefits if those benefits would be covered under Traditional Medicare.

Key Provisions of CMS-4201-F:
* Individualized Determinations: Plans must base decisions on the patient’s specific medical history. They cannot rely on statistical averages from large datasets.
* Prohibition on AI as Sole Decider: Artificial intelligence can assist but never dictate a denial.
* Public Transparency: Internal coverage criteria must be publicly accessible and supported by clinical evidence.

UnitedHealthcare publicly adjusted its stance following the rule. They claimed nH Predict was merely a "guide" for clinicians. However, the data suggests this was a distinction without a difference. The denial rates for post-acute care did not plummet as expected. Instead, they stabilized at historically high levels. This suggests the algorithm continued to function as the de facto decision-maker. Human reviewers likely rubber-stamped the AI's "suggestions" to meet productivity quotas.

On February 6, 2024, CMS issued a clarifying memo to close loopholes. The agency stated that an algorithm predicting length of stay cannot be the basis for terminating services. This was a direct shot at the core function of nH Predict. The tool's primary value proposition is predicting exactly when a patient should be discharged. This prediction often contradicts the treating physician's assessment of when the patient is actually ready to leave.

#### Judicial Reckoning: The Lokken Ruling (February 2025)

The legal battle over these practices culminated in the class action Estate of Gene B. Lokken v. UnitedHealth Group. In February 2025, U.S. District Judge John R. Tunheim delivered a significant blow to UHC’s defense strategy.

UnitedHealthcare attempted to dismiss the lawsuit by arguing that plaintiffs had not exhausted their administrative appeals. Judge Tunheim rejected this argument. He validated the plaintiffs' claim that the appeals process was "futile." The court recognized a cynical pattern in UHC’s strategy. The insurer would deny claims en masse using AI. Then they would pay off the tiny fraction of beneficiaries who fought all the way to the final appeal stage. This tactic prevented any single case from setting a legal precedent or triggering judicial review.

The "Futility" Metric:
The court filings exposed a staggering disparity in the appeals data.
* 90%: The estimated error rate of nH Predict. This figure represents the percentage of denials overturned when fully appealed.
* 0.2%: The estimated percentage of denied policyholders who actually complete the appeals process.
* >99%: The effective "success" rate of the algorithm in saving costs. It succeeds not by accuracy but by attrition.

This ruling allowed the case to proceed on claims of breach of contract and bad faith. It stripped away the procedural shields UHC had relied upon. The judge’s decision signaled that the judiciary would no longer accept "failure to appeal" as a defense when the appeals system itself is engineered to exhaust the elderly.

#### The Senate Investigation (October 2024)

While the courts moved slowly, the U.S. Senate Permanent Subcommittee on Investigations (PSI) accelerated the public reckoning. On October 17, 2024, the subcommittee released a report that provided the first verified internal data on the scale of these denials.

The report, helmed by Senator Richard Blumenthal, utilized subpoenaed documents from UnitedHealthcare, CVS, and Humana. The findings destroyed the industry narrative that prior authorization focuses on patient safety.

UnitedHealthcare Specific Data (PSI Report 2024):
* Denial Rate Surge: UHC’s denial rate for post-acute care prior authorizations more than doubled in two years. It rose from 10.9% in 2020 to 22.7% in 2022.
* Targeted Savings: The denials were not randomly distributed. They were heavily concentrated in high-cost areas like Skilled Nursing Facilities (SNF) and Inpatient Rehabilitation Facilities (IRF).
* Automation Bias: The report cited internal documents showing UHC working groups explicitly exploring how to "automate" the denial process to increase savings.

The subcommittee concluded that UHC and its peers were "intentionally targeting" costly services. They substituted medical necessity with financial calculations. The report demanded that CMS begin collecting prior authorization data broken down by service category. This would prevent insurers from hiding high denial rates for nursing homes behind high approval rates for cheap services like generic drugs.

#### 2026 and Beyond: The Financial Squeeze

By early 2026, the regulatory environment shifted from rule-setting to financial enforcement. The CMS Advance Notice for Calendar Year 2027, released in January 2026, proposed a net rate increase of just 0.09%. This was effectively a cut when adjusted for inflation.

This flat rate signaled the end of the "growth at all costs" era for Medicare Advantage. CMS effectively told insurers that the government would no longer subsidize their inefficiencies or their legal defenses.

Simultaneously, the HHS Office of Inspector General (OIG) announced a new series of "aggressive" audits targeting the Risk Adjustment Data Validation (RADV) processes. The OIG's 2026 audit plan specifically targets the discrepancy between AI-driven diagnosis coding (which increases payments to plans) and AI-driven care denials (which decreases payments to providers).

The State-Level Patchwork (2025-2026):
Federal inaction on specific AI bans led states to intervene. By 2026, a patchwork of laws in states like California, Texas, and Georgia created a compliance nightmare for national carriers.
* Human-in-the-Loop Laws: New state statutes mandated that a licensed physician of the same specialty must review any AI-generated denial before it is sent to the patient.
* Disclosure Requirements: Insurers in regulated states must now disclose to patients when an algorithm was used to assess their claim.

#### Summary of Regulatory Impact (2023-2026)

The data indicates that regulatory pressure has not yet forced UnitedHealthcare to abandon nH Predict. The tool is too profitable. However, the cost of doing business has risen dramatically. The combination of class-action liability, stagnant government reimbursement rates, and the exposure of the "futility" strategy has stripped the insurer of its ability to operate in the shadows. The algorithm is no longer a secret weapon. It is a known liability.