BCBS research shows hospitals' AI systems inflate health care costs

Dr. David Wennberg and colleagues at Blue Health Intelligence put the estimate in a new report they prepared for the Blue Cross Blue Shield Association.

Many hospitals are now using AI "ambient listening tools" to record physicians' conversations with patients and feed information from the conversations into the patients' electronic health records.

"While the hospital can achieve clear productivity gains by automating the documentation and coding processes, careful attention must be paid to ensure the newly billed diagnoses are reflective of the true acuity and level of treatment for that patient admission," Wennberg and his colleagues write in their report.

The analysts attributed the increases in coding intensity mainly to use of ambient listening tools and closely related types of AI tools.

What it means: Health insurers have been saying for years that hospitals are using AI systems to bad their bills. Now, the Blues have a formal analysis to support that argument.

The analysis could help employers and their health plan administrators counter hospitals' assertions that they haven't done anything wrong and the patients are just a lot sicker than they used to be.

Blue Health Intelligence: Blue Health Intelligence is a Chicago-based organization that's owned by the Blue Cross Blue Shield Association and some of the association's member companies.

The analysis methods: The analysts started with anonymized commercial inpatient claims data with service dates from April 1, 2022, through March 31, 2025.

All of the patients involved had primary coverage from a Blue Cross or Blue Shield plan that participated in the study.

Hospitals and other health care providers use diagnosis codes to show what health problems patients have and how sick the patients are. The providers use procedure codes to show what medical procedures have been performed.

The analysts divided the hospitals into two groups: hospitals that had a big increase in diagnosis severity codes between 2022 and 2025, and hospitals without big increases in diagnosis severity codes.

The analysts also classified the claims as complex or non-complex, based on whether patients appeared to have major complications or other major health problems.

The analysts then looked at whether big increases in coding intensity correlated with any changes in the intensity of care the patients received.

The analysis results: At the 10% of the hospitals with the biggest increases in the percentage of patients classified as complex, the percentage classified as complex increased to 59.8% in the first quarter of 2025, from 46.8% in the second quarter of 2022.

For the hospitals, the percentage of patients classified as complex increased to 55.4%, from 51.2%.

When the analysts looked only at women who were admitted to the hospital after giving birth, they found that, at hospitals with a big increase in case complexity, the percentage of maternity patients classified as having postpartum anemia increased to 12.3% in the first quarter of 2025, from 4% in the second quarter of 2022.

At the other hospitals, the percentage of maternity patients classified as having postpartum anemia held steady at about 8%.

At both the hospitals with a surge in postpartum anemia diagnoses and at the other hospitals, the percentage of maternity patients who had blood transfusions held steady at about 1%.

"The reported level of severity for maternity admissions increased, but the intensity of the treatment did not," the analysts write. "This suggests a discordance between the reported level of complexity of the patient and the treated condition."

The audit: One health insurer involved in the analysis audited the maternity claims at a hospital identified as having an unusual surge in postpartum anemia claims.

At that hospital, the percentage of maternity patients classified as having postpartum anemia increased to 2.9%, from 23.5%, and the percentage classified as having all kinds of post-hemorrhagic anemia increased to 10.9%, from 5.8%.

At the other hospitals, the percentage of patients classified as having post-hemorrhagic forms of anemia of all kinds increased to 5.1%, from 4.5%.

When the insurer audited the hospital with the spike in the number of maternity patients with postpartum anemia for postpartum anemia, "less than 20% of the cases met established clinical criteria for postpartum anemia."

The limitations: One possible limitation is that the data collected came during the post-COVID-19 period, when some hospitals may have had disruptions in coding efforts unrelated to AI use.

Another limitation is that the analysts did not try to adjust for variables other than surges in coding intensity when they divided hospitals into the surge and no-surge groups.

The backdrop: Some patients have expressed discomfort about doctors asking them to consent to the use of ambient listening systems, and lawmakers in states like California have started to introduce legislation regulating the use of ambient listening systems.

Related: Key California lawmaker calls for tight health privacy rules for AI patient data collection

But health care provider organizations and health coverage provider organizations of all kinds have been making heavy use of AI-powered and human-powered coding improvement programs, to create more detailed patient records, improve patient care and maximize case severity-related revenue.

A Senate committee recently investigated rising patient health problem intensity scores at UnitedHealth Medicare Advantage plans. The investigators found no evidence of errors or fraud, but they did find that the company's plan enrollee evaluators have been going into enrollees' homes, looking at the enrollees closely and identifying health problems that the enrollees' doctors had not noticed or had not recorded.

Use of ambient listening tools in health care could be about to grow rapidly: Amazon recently unveiled a package of AI support tools for medical offices. The package includes an ambient listening tool and a tool that can use information from patient encounter recordings to document the codes included in insurance claims.