The implementation of AI tools in U.S. hospitals has resulted in an additional $942 million in costs for insurers. These systems are identifying more diagnoses, leading to higher service bills, although the level of care provided remains unchanged. This information was detailed in a report by the Blue Cross Blue Shield Association (BCBSA).

AI is being utilized for medical coding, where it reviews physician notes and lab results to identify comorbidities. When these are detected, hospitalizations can be categorized as more complex, affecting the reimbursement amount paid to the hospital.

BCBSA analysts examined insurance claims from early 2023 to late 2025, finding that the proportion of hospitalizations classified as medically complex rose from approximately 37% to 40% during this period.

Compared to 2023 levels, this shift cost insurers $942 million over the following two years.

About 70% of this total was attributed to secondary diagnoses. In over 55,000 cases, these diagnoses elevated the complexity of the insurance claims, resulting in an additional $653 million in payments to hospitals.

“It is critically important that we did not find any changes in the care that would correspond to a more complex patient,” said Luca Choker, senior vice president at BCBSA, one of the report's authors.

Choker noted that the $942 million estimate only considers instances where the volume of care did not change. If a hospital documented additional treatments, those expenses were excluded from the calculations.

The association's main argument centers on the disparity between diagnoses and treatment. For instance, in facilities that frequently recorded acute post-hemorrhagic anemia, blood transfusions were performed less often.

According to the association, this is the most compelling evidence that the increase in diagnoses reflects a change in documentation practices rather than a deterioration in patient conditions.

“One should ask: why is it that one hospital, treating similar patients under similar conditions, starts to deviate? It may partly relate to correct coding, but I believe the likelihood of this being an overstatement due to technology is higher,” remarked Razia Hashmi, BCBSA's vice president for clinical affairs.

However, the analysis did not establish a direct causal link between specific AI systems and the final cost outcomes.

The organization relied on insurance claims rather than comprehensive medical records and acknowledged that clinical data would provide a more accurate picture of patients' true conditions.

Hospitals Cite Other Factors

In July, the American Hospital Association (AHA) rejected claims that AI alone leads to unjustified increases in coding complexity.

The AHA believes that hospitals are seeing more patients in severe conditions. They attributed this trend to an aging population, the rise of chronic diseases, and the transfer of less complex cases from inpatient to outpatient settings.

Supporting this, they cited data from Vizient, which indicates that the complexity index of patients in U.S. hospitals has increased by roughly 5% from 2019 to 2024.

The AHA views the role of AI differently than insurers, stating that technology assists in accurately documenting existing conditions, while the final coding decisions rest with healthcare professionals.

Both sides utilize automation in this debate: hospitals for billing purposes, and insurers for claim verification and cost reduction.

Authorities also acknowledge the short-term rise in expenses. On September 23, Mehmet Oz, head of the Centers for Medicare and Medicaid Services, stated that initially, AI will increase healthcare costs as it enables existing billing systems to operate more efficiently.

In April, researchers at the Mayo Clinic trained a neural network to detect signs of pancreatic cancer an average of 475 days before diagnosis.

Previously, British hospitals began implementing the AI blood test PinPoint for diagnosing uterine cancer, achieving a 99% accuracy rate in a study involving 3,313 women.

Follow ForkLog on social media

Telegram (main channel) Facebook X