BCBSA says AI-coding is leading to increased costs

Everyone is talking about the BlueCross BlueShield Association report that came out last week, which put numbers to a problem that payers have been talking about for the last couple years: AI is leading to higher prices at hospitals.

Specifically, it reported that elevated billing at hospitals led to nearly a billion dollars in added cost for Blues plans over the last two years.

The main question here is: Are hospitals using AI to correctly bill for services or is there fraud at play here?

Hospitals are billing patients as more medically complex without additional treatment, the report says. “Compared with 2023 baseline rates, hospitals classified 55,158 additional cases as complex. Those cases generated $653 million in incremental reimbursements at an average of $11,800 per excess complex case.”

BCBSA focused on growing secondary diagnoses, or diagnoses that co-exists with another main diagnosis, which it says is leading to incremental new costs. For example, among patients with major bowel procedures, BCBSA found an increase in diagnosis for acute post hemorrhagic anemia, iron deficiency that occurs after significant blood loss; acidosis, a condition where the body makes too much acid; and moderate protein calorie malnutrition.

And yet, they’re not seeing commensurate coding that shows doctors are treating these secondary health problems. In the case of increased rates of anemia, for example, they’re not seeing more blood transfusions. However, the report also acknowledges the limit of coding data.

Hospitals that had the highest growth in patient complexity, diagnosed anemia 38% more often than their peers. However, there was not a huge difference in transfusions performed as a percentage of patients diagnosed at these institutions. Hospitals that are coding for higher complexity transfused 16.9% of patients with anemia versus 19.3% at hospitals with lower complexity diagnoses.

Notably, the existence of anemia does not necessarily lead to transfusion. It may lead to more monitoring or labs, but because of payment bundling isn’t showing up in discrete codes:

“Inpatient laboratory services are typically bundled into DRG payments, so claims data doesn't reliably reflect whether more labs were performed. Because of this bundling of routine hospital services covered in the DRG (and not separately coded), we defined "discordance" based markers reliably found on claims. In this example, we highlighted transfusion since this is reliably coded and we'd expect increases in acute post-hemorrhagic anemia to often require transfusion ("post-hemorrhagic" is an important qualifier over just anemia),” a spokesperson for BCBSA wrote via email.

You can read the report here.

There are two long-standing narratives at play here. Health providers say they’re routinely under reimbursed for their services and payers say that hospitals charge too much. Now, these narratives are being turbo charged by AI.

I’ll be publishing a deep dive on this issue later this week that will get into the push and pull between AI-enabled coding and AI-enabled payment integrity, and what it means for both operators and patients. Stay tuned.

Now, onto the news…

—Ruth