UnitedHealth nH Predict AI Claim Denial System (2023)

Incident Summary

UnitedHealth subsidiary naviHealth deployed an AI algorithm called nH Predict to predict patient length of stay and automatically determine Medicare Advantage claim approvals for post-acute care, including nursing home and rehabilitation services.^[1] The system overrode physician judgment in consequential healthcare decisions, operating as an automated decision loop that processed claims faster than meaningful human review could occur.

Key Facts

• nH Predict generated predictions for patient care duration and was used to deny continued coverage for post-acute services including nursing home stays and rehabilitation.^[1]
• A documented 90% error rate was identified — nine out of ten appealed denials were ultimately reversed on review, indicating the AI system was systematically incorrect in its assessments.^[1]
• Denial rates for post-acute services increased from 8.7% to 22.7% after the system’s deployment, and skilled nursing home denial rates increased approximately ninefold.^[3]
• Employees reported being pressured to keep patient stays within 1% of the AI-predicted length, regardless of clinical judgment.^[1]
• A class-action lawsuit was filed in November 2023 and was allowed to proceed on breach of contract and good faith claims.^[2]
• A subsequent Senate investigation found similar AI-driven denial patterns across multiple Medicare Advantage insurers.^[3]

Threat Patterns Involved

The primary pattern is decision loop automation — an AI system executing consequential healthcare decisions in a rapid feedback loop, operating faster than human oversight could meaningfully intervene. Physicians’ clinical assessments were systematically overridden by algorithmic predictions. The secondary pattern is overreliance and automation bias, where organizational processes deferred to the AI’s output despite a demonstrated 90% error rate on appeal, indicating that human reviewers either could not or did not override the system’s determinations.

Significance

This incident represents one of the most documented cases of AI-driven automated decision-making causing measurable harm at scale in healthcare. The 90% reversal rate on appeal demonstrates that the system was not functioning as intended, yet the organizational structure continued to defer to its outputs. The case has regulatory implications for the use of AI in insurance claim adjudication and has prompted legislative attention to algorithmic decision-making in healthcare. The Senate investigation revealed that the pattern extends beyond a single insurer, suggesting a sector-wide structural problem with AI-automated coverage determinations.