AI hospital coding added $942m to Blue Cross costs, insurer study says

Artificial intelligence tools used to scan medical records and document patient encounters may be increasing U.S. healthcare spending by identifying additional billable diagnoses without a corresponding increase in treatment, according to new research from the Blue Cross Blue Shield Association. Reuters reported that the association estimates more intensive medical coding contributed approximately $942 million of additional costs for Blue Cross companies during 2024 and 2025 compared with 2023 patterns.

Around $653 million of the increase came from more frequent billing of secondary conditions, according to the study. Providers are increasingly using AI systems to scan patient records or ambient-scribe technology that listens to clinical conversations and drafts documentation, creating opportunities to identify conditions that may previously have gone uncoded. More complete documentation can legitimately improve billing accuracy, but Blue Cross says the rapid rise in higher-severity coding has not consistently been matched by evidence that patients received additional treatment.

How can AI coding software increase hospital bills without changing treatment?

U.S. hospital payments frequently depend on how complex a patient’s case is classified. Secondary diagnoses and complications can move an admission into a higher reimbursement category even when the principal procedure remains unchanged, meaning the information written into a medical record can materially affect what an insurer ultimately pays.

AI tools can scan enormous amounts of clinical documentation far faster than human coders and identify mentions of conditions that might support higher-severity classifications. Ambient systems can also create more detailed notes by listening to doctor-patient conversations and turning speech into structured documentation.

That capability can be valuable when it captures genuine medical complexity that clinicians previously failed to document. The concern raised by the Blue Cross study is that coding intensity may be rising faster than underlying treatment intensity, suggesting software may be finding reimbursement opportunities that do not necessarily represent meaningful changes in patients’ clinical condition.

Blue Cross researchers examined inpatient claims and compared changes in diagnoses with the treatments normally associated with those conditions. Their argument is not that every AI-assisted diagnosis is invalid, but that the aggregate pattern warrants scrutiny because reimbursements rose substantially without corresponding increases in certain expected interventions.

What did Blue Cross find in patients undergoing major bowel surgery?

The association highlighted patients undergoing major bowel procedures as an example. Between the first quarter of 2023 and the fourth quarter of 2025, coding for secondary conditions including partial intestinal obstruction rose by approximately 55%, while coding for excessive acid buildup in the body increased by roughly 33%.

Researchers then examined whether treatment patterns rose in parallel. Blue Cross Vice President of Clinical Affairs Razia Hashmi told Reuters that greater medical complexity did not correspond with a comparable increase in treatments for people undergoing bowel-disease surgery.

Anemia provides another illustration. If significantly more patients genuinely develop serious anemia, researchers would ordinarily expect a rise in interventions such as blood transfusions. Blue Cross said transfusion rates did not increase in line with the higher number of anemia diagnoses, contributing to its concern that AI may be identifying more billable conditions rather than simply documenting a genuinely sicker patient population.

Those findings are association-level claims based on insurance data and should not be interpreted as proof that every hospital using AI coding software is billing improperly. They do, however, provide a quantitative basis for insurers to scrutinize how rapidly automated documentation is changing reimbursement patterns.

Why are insurers increasingly concerned about AI-driven medical upcoding?

Medical-cost inflation directly affects insurer margins because health plans generally set premiums before knowing exactly how expensive the following year’s claims will become. Unexpected increases in coding intensity can therefore raise medical expenses even when utilization, procedures and underlying disease prevalence remain relatively stable.

Centene Corporation and other insurers have previously warned that artificial-intelligence tools used by healthcare systems may contribute to aggressive or inappropriate reimbursement claims. Blue Cross’s new analysis expands that discussion by attaching an estimated dollar amount to the effect across its member organizations.

The Blue Cross Blue Shield Association represents 31 independently operated health insurers collectively covering more than 100 million people. That scale means relatively small changes in payment per admission can generate hundreds of millions of dollars of additional annual spending when applied across a large claims base.

The economic incentive exists on both sides. Hospitals face rising labor, drug and supply costs and benefit financially from documenting every legitimate condition that raises reimbursement, while insurers have incentives to challenge coding they believe exaggerates the actual complexity of care.

Could AI documentation also solve real problems inside American healthcare?

Yes, because healthcare documentation has long been one of the industry’s major administrative burdens. Physicians can spend substantial amounts of time entering notes into electronic health records, while coding teams manually review complex documentation to ensure hospitals are reimbursed correctly.

Ambient AI scribes can reduce that workload by listening during consultations and generating draft medical notes for clinician review. Record-scanning systems can also identify overlooked diagnoses that genuinely should have been documented, potentially improving accuracy and ensuring that hospitals are compensated for care they actually provide.

The policy challenge is separating legitimate documentation improvement from reimbursement optimization that adds cost without improving care. A system that catches a clinically relevant diagnosis can help patients and providers; a system optimized primarily to identify any language capable of supporting a higher billing category creates very different incentives.

That distinction will become more important as hospitals deploy AI widely. Once automated systems are embedded in documentation workflows, even small biases toward greater diagnostic severity can generate enormous aggregate financial effects.

What could the $942 million finding mean for CVS, Cigna and Centene?

The study does not directly attribute the entire $942 million to publicly traded insurers such as CVS Health Corporation, The Cigna Group or Centene Corporation. Reuters identified those companies because reimbursement intensity is relevant across the U.S. managed-care industry, while Centene has specifically discussed concerns around AI-assisted provider coding.

For insurers, stronger medical coding can push medical-loss ratios higher, particularly in commercial insurance and government-backed programmes where payment methodologies depend heavily on risk and diagnostic information. Companies can respond through audits, payment-integrity technology, contract negotiations and ultimately higher premiums, but those countermeasures introduce their own administrative costs.

The result could become an AI-versus-AI arms race. Providers may use software to find every reimbursable diagnosis, while insurers deploy separate algorithms to flag improbable coding patterns or compare diagnostic severity with actual treatment.

That would be a peculiar outcome for technology originally promoted partly as a way to lower healthcare administration costs. Artificial intelligence could reduce manual paperwork while simultaneously intensifying the financial contest over how that paperwork is translated into payments.

Could AI medical billing ultimately increase insurance premiums for consumers?

Higher claims costs eventually tend to affect premiums, employer health-benefit expenses or public-programme spending if the increase persists and cannot be offset elsewhere. A $942 million increase across Blue Cross organizations is meaningful, but it remains small relative to total U.S. healthcare spending and does not by itself establish that AI coding will materially raise premiums nationwide.

The larger concern is scalability. AI documentation tools are still spreading through hospital systems, meaning billing effects observed during 2024 and 2025 could become larger if the technology reaches substantially more patient encounters.

Insurers are therefore likely to push for clearer coding standards, audit processes and evidence linking higher-severity diagnoses with actual clinical treatment. Hospitals, meanwhile, will argue that historical under-documentation should not be preserved merely because automated tools now make it easier to capture legitimate medical complexity.

The Blue Cross findings expose an important contradiction in healthcare AI. Better information should theoretically make medicine more efficient, but in a reimbursement system where more documented complexity frequently means more money, software that finds additional information can increase spending even before it changes a single medical treatment.