Blue Cross says hospital AI coding tools added $942 million in costs
A Blue Cross Blue Shield analysis found hospitals coded more patients as medically complex without more treatment to match, adding nearly a billion dollars in claims over two years.
What happened: The Blue Cross Blue Shield Association says an analysis of its claims data found hospitals classified a growing share of inpatient stays as medically complex between 2023 and 2025, rising from about 37% to 40% of cases, without a matching rise in actual treatment intensity. BCBSA estimates the shift added roughly $942 million in costs to Blue plans, with about $650 million tied to secondary diagnoses that pushed stays into higher-paying billing categories. It says AI-enabled documentation tools, now used by more than 60% of hospital systems, are a likely factor, though it cannot prove AI caused any single case.
Why it matters: Hospital bills already feel opaque and expensive to patients, and this dispute shows software is now shaping the numbers on both sides of the transaction. If insurers pass added claims costs into premiums, ordinary policyholders could feel it even though the fight is happening entirely in back-office coding systems, far from any exam room.
How it works, plainly: AI tools scan clinical notes, labs and records to spot diagnoses a human coder might miss, such as anemia or complications, which can shift a hospital stay into a costlier billing category. BCBSA says one telltale sign is a diagnosis appearing more often without matching treatment. For example, more anemia diagnoses appeared but fewer blood transfusions followed at hospitals with the heaviest coding. Bowel-procedure claims coded at the highest complexity level nearly doubled to 22.7% of cases in the study window.
The rollout: The American Hospital Association rejects the upcoding framing, saying patients are genuinely older and sicker, less-complex procedures have moved to outpatient care, and its own analysis found real complexity rose about 5% from 2019 to 2024. It also notes insurers use AI to downcode or deny claims, citing a 2025 MedPAC estimate that Medicare Advantage coding practices added roughly $40 billion in federal payments. BCBSA's own study relies on claims, not full patient charts, so neither side has definitive proof.