Blue Cross analysis links AI coding to $942M in added costs

AI-assisted documentation and coding contributed $942 million in added costs for Blue Cross insurers over two years, measured against a 2023 baseline. Secondary diagnoses drove $653 million of that figure during 2024 and 2025.

Luke Chalker, senior vice president of product and data science at the Blue Cross Blue Shield Association, delivered the finding Thursday at a briefing for reporters. The analysis had gone through inpatient claims billed to Blue Cross plans and looked for what the codes implied about the patients behind them. When stays were documented as more complex, the care recorded on those claims should have moved too. It had not. Documentation had shifted. Treatment had not followed.

“Critically, what we found is underneath all of that data [was] no change in corresponding care for a more complex patient,” Chalker said. “If patients are truly sicker, we’d expect to see more treatment.”

The association represents 31 independent Blue companies covering more than 100 million people. Its review of the inpatient claims those plans paid found that hospital stays were being billed as medically complex more often. Providers had documented secondary conditions at higher rates, and those conditions were the ones that could reclassify a stay into a higher-paying severity tier. The share of inpatient stays classified as medically complex rose from 37 percent to 40 percent from 2023 through the end of 2025. About 70 percent of the coding-intensity increase came from more than 55,000 additional cases counted from the 2023 baseline. In each of those cases, secondary diagnoses pushed the claim into a higher-severity diagnosis-related group. Higher-severity groups carry higher reimbursements, so attaching the secondary codes produced a larger bill for the same primary stay. Each excess complex case ran an estimated $11,000.

A hospital’s payment for an inpatient stay runs through a diagnosis-related group. The classification originated at Yale in the 1970s and has structured U.S. hospital payment since the early 1980s. A grouper program assigns each admission from ICD diagnoses, procedures, age, sex, and discharge status. Secondary diagnoses and complications move a stay into higher-paying severity tiers.

Human coders once searched charts, labs, and clinical notes by hand for every secondary condition that could change the group. Ambient listening tools now convert clinician–patient conversations into structured notes inside the electronic health record. Autonomous coding then scans those notes, lab results, and problem lists for billable secondaries that can reclassify the DRG and raise the payment. BCBSA traced the coding-intensity changes since 2023 to systemic adoption of AI revenue-cycle management tools. Hospitals and health systems have said the tools help them code claims more accurately and efficiently amid increased scrutiny from payers. A June survey cited by the association found more than 63 percent of healthcare organizations reporting AI use in their revenue-cycle workflows.

Diagnoses were only half of what the claims showed. The other half was the care delivered.

Among people undergoing major bowel surgery, diagnoses of partial intestinal blockage rose 55 percent from the first quarter of 2023 through the fourth quarter of 2025. Diagnoses involving excess acid in the body rose 33 percent. Within those same procedures, claims at the highest complexity level climbed from 10.2 percent to 22.7 percent while non-complex cases fell from 36.6 percent to 32.8 percent. The shifts together accounted for nearly $61 million in incremental costs.

Anemia patients showed no corresponding rise in blood transfusions. Hospitals in the top quartile for complex diagnosis-related group cases had similar or lower treatment intensity than their peers—on ICU utilization, transfusion, reoperation, and median length of stay—across 65 percent of completed DRG cases. Those hospitals coded posthemorrhagic anemia at 13.7 percent, against 9.9 percent elsewhere. Among patients carrying the diagnosis, the transfusion rate was 16.9 percent at the top-quartile hospitals and 19.3 percent at the others.

“The disconnect between diagnoses and treatment suggests that AI is identifying more billable conditions, not sicker patients,” Chalker said. “We now see that coding has materially changed. We see that. Non-Blues see that—they talk about it sometimes in earnings reports and things like that. But we find no evidence of a corresponding change in care, and that’s because the reimbursement mechanisms that exist allow this.”

Dr. Razia Hashmi, vice president of clinical affairs at BCBSA, put the split between hospitals in a single question. “The question that is worth asking is [with] two similarly situated hospitals, treating similar patients, why would one hospital diverge?” she said. “There may be an element of correct coding there, but the likelihood that this is technology-enabled upcoding is higher, in my view.”

Robert Boos, former vice president and chief revenue officer at Centra Health in Lynchburg, Va., rejected that framing in a Becker’s interview. “The idea that hospitals are coding aggressively to drive up costs is misleading,” Boos said. “Coding is governed by strict federal and industry standards, and health systems invest heavily in compliance, training and auditing to ensure accuracy.”

Chalker separated the claims the association would still honor from the ones it flagged. When a hospital documented that it had delivered additional care, payment followed. “That’s the stuff … that’s a no-brainer from my perspective. That’s the stuff that hospitals should bill for, and that’s the stuff we should pay for,” he said.

The white paper released Thursday focused on hospital inpatient services. BCBSA plans to continue releasing analyses on outpatient care and other diagnosis-related groups. The association acknowledged a built-in limit: the findings relied on claims rather than clinical documentation, which would be a more direct measure of whether patients were substantially sicker than peers without the increased coding complexity. Chalker said he believed a similar analysis with clinical data on hand would make a stronger case. Speaking to Becker’s as chief product officer at Blue Health Intelligence, BCBSA’s analytics arm, he described a data set effectively complete for more than 60 million Americans across every geography. Leveraging that reach, he said, had given the Blues the ability to tap into the coding fingerprints faster than most.

“And at least in the work that we've done with Blue Plans and who do have more access [to charts] because of provider relationships with that clinical data, they've been able to kind of re-emphasize and demonstrate this effect,” Chalker said.

The study examined hospitals that had publicly announced adoption of AI coding tools. At one such hospital, overall case complexity climbed 6.7 percent following that disclosure. Other hospitals in the same state saw complexity grow less than 1 percent over the same stretch.