The US insurance association Blue Cross Blue Shield Association (BCBSA) has identified additional diagnoses worth 942 million dollars in an analysis of its own billing data over two years, since clinics have increasingly used AI software for coding. According to the data, the actual treatment of the affected patients did not change measurably. The hospital association AHA disputes the interpretation of AI as a cost driver.
Analysis finds striking coding patterns in bowel surgeries
BCBSA evaluated billing data from its member insurers from early 2023 to the end of 2025, covering about one-third of the US population. The focus was on diagnosis groups for major bowel surgeries: The association recorded an increase in additional coded secondary diagnoses such as partial bowel obstructions or elevated acid levels. Of the total 942 million dollars in additional costs, around 653 million dollars – about 69 percent – can be specifically attributed to these additionally billed secondary diagnoses; the analysis does not provide a detailed explanation for the remainder.
Particularly striking: In the quarter of hospitals with the largest jumps in coding severity, intensive care unit usage was 11.5 percent, lower than in the other hospitals (13.2 percent), as was the rate of blood transfusions (3.6 versus 3.9 percent). More documented severity did not correspond with more actual treatment in these cases. Luke Chalker, responsible for product and data science at BCBSA, puts it succinctly: if patients were truly sicker, it would also show up in more treatment. BCBSA does not name specific software manufacturers in the study and says it has no data on which hospitals actually use AI coding tools – the report therefore speaks of a correlation, not a proven causal relationship.
Hospital association counters with data on case severity
The hospital association American Hospital Association (AHA) had already rejected criticism of growing coding intensity as “not supported by evidence” in a paper published at the end of July 2026, as reported by XenoSpectrum. The AHA attributes the increase instead to an aging, chronically sicker population and to lighter cases increasingly being treated on an outpatient rather than inpatient basis – leaving, on average, more severe cases among the remaining hospital patients. A joint study by AHA and the consulting firm Vizient puts the rise in the so-called case-mix index between 2019 and 2024 at around five percent.
From the hospitals’ perspective, AI software mainly helps capture existing but previously undocumented diagnoses more accurately – not create new diagnoses out of thin air. Both sides rely on the same underlying billing data but draw opposite conclusions from it: for BCBSA, the absence of additional treatment is a warning sign; for the AHA, it simply reflects better documentation of conditions that already existed.
It remains open whether a body independent of the insurers will ever review the raw data: so far, the only quantified estimate comes from the party that would directly benefit from lower billing as the one paying the bills. For healthcare systems outside the US with different reimbursement logic – such as Germany’s DRG system with state oversight – no comparable study exists yet, even though AI-supported documentation tools are increasingly in use there too.
