How AI scribes are shifting coding intensity, reimbursement

What the research says about coding intensity

Two recent analyses have identified two areas in which coding intensity has shifted alongside the adoption of AI scribes and ambient listening technology.

First, the Blue Cross Blue Shield Association (BCBSA) and its data analytics partner, Blue Health Intelligence, recently found significant changes in care for maternity admissions. Researchers analyzed deidentified claims data from tens of thousands of maternity cases nationwide, revealing a significant increase in cases coded for acute posthemorrhagic anemia, a serious condition that usually requires blood transfusions.

However, many patients coded with the diagnosis never received a transfusion or other appropriate treatment, highlighting a disconnect between coding and clinical practice. That disconnect contributed to about $22 million in additional spending from 2023-2024.

But it's not just happening in maternity care. A second analysis released by healthcare market intelligence company, Trilliant Health, found increased outpatient coding intensity following hospital adoption of AI scribes.

The Trilliant Health analysis studied national all-payer claims data for evaluation and management (E/M) billing patterns at six large hospitals and health systems from 2018-2024. All of the organizations had publicly announced their adoption of AI scribes and ambient listening technologies.

There was "consistent upward redistribution of both new and established outpatient E/M visits toward CPTs 99204-99205 and 99214-99215" across all organizations, according to the analysis. Notably, coding for high-intensity new patient E/M visits ranged from 12-20 percentage points to as high as 80% at one health system.

Concern or coincidence?

The research clearly showed an upward shift in coding intensity and medical billing patterns in certain service lines. However, the underlying causes of these changes were not so clear. Luke Chalker, chief product officer at Blue Health Intelligence, explained how the analysis made a connection to AI scribes and ambient listening technologies.

"We started harvesting data points that were very clearly trending upwards, but not just trending upwards, having variability in how they were trending," he explained to RevCycle Management. "So, facility-to-facility differentials that couldn't be really explained in any normal context like site of care or through a public or population health issue."

"Posthemorrhagic anemia was one that came through in our assessment, which is less around DRG migration and more around the complexity of codes being added for anemia," Chalker continued. "What we found, I think pretty clearly, was a massive step-change growth at facilities."

That step change in growth is what made it clear there were coding changes at play, he explained. Primarily, there was no significant change in how care was delivered across facilities during the analysis. However, there was a coinciding trend in technology adoption: deployment of AI scribes, many of which are also described as revenue cycle optimization technologies.

That's why it is unsurprising to Allison Oakes, chief research officer at Trilliant Health, to see more intense coding for outpatient cases.

"In general, it seems to be that these AI-enabled scribing tools are allowing clinical documentation to be captured more thoroughly and accurately," she said. "And this is a little bit of a double-edged sword."

On one hand, AI has been able to automate burdensome processes, making them not only more efficient but also less prone to human error. On the other hand, as more information from the AI scribes is entered into the medical record, more revenue capture can occur.

"Because we see this systematic increase in coding intensity, it probably suggests that, historically, providers have actually been under-coding their visits," Oakes explained. "That could be out of incomplete clinical documentation, suboptimal coding practices or fear of violating the False Claims Act, or some combination of those things."

"The fact that this ambient AI scribing technology is able to document so much more comprehensively for providers, potentially catching higher acuity clinical diagnoses, also provides extremely reliable timestamps as it relates to the length of a visit. Those things converge, so when an AI-based technology is following the rules of those billing codes, we're seeing this upward shift in intensity," she added.

What's the rub?

In light of emerging data, the use of AI scribes is creating friction. After all, these tools have become invaluable to providers.

"It allows me to be present and engage with my patient while I'm collecting a history," explained Luann Racher, M.D., associate professor in the Department of Obstetrics & Gynecology at the University of Arkansas for Medical Sciences (UAMS).

Patients are also responding positively to the technology. Patients feel like they are getting specific attention from their providers, who no longer have to stare at a computer or take written notes to remember what happened for documentation.

Physicians like Racher are also getting their own time back. A study of AI scribe use at The Permanent Medical Group showed the technology saved 15,000 hours after 2.5 million uses in a year. AI scribes have the potential to reduce the clinical documentation burden and clinician burnout.

The rub, though, is healthcare affordability.

"When you code for higher reimbursement, that doesn't just affect a payer. It affects a member on a high-deductible plan. It affects a member in their premiums over time," Chandler stated.

Blue Health Intelligence estimated approximately $2.3 billion more in healthcare spending, with about $663 million in inpatient spending and at least $1.67 billion in outpatient spending that can be tied to more aggressive practices enabled by AI.

Patients aren't necessarily seeing more or higher quality care as coding intensity changes, Oakes added. But they will see higher bills from more intense billing codes associated with their visits.

"At the end of the day, it's a set of rules that providers have to follow when it comes to figuring out what code to bill, and these AI technologies are just that much better at following those rules and potentially pushing the envelope a little bit," she said. "It all just depends on how they're tuned, more or less."

Most AI scribe technology has safeguards in place to prevent true upcoding. For example, Racher and her colleagues still have to evaluate the clinical notes taken by their ambient listening technology before sending them off to the billing team.

"The goal of using this generative AI tool is never to upcode, never to overcharge. It's always looking at doing accurate and precise documentation that then is transferred over into how the billing is coded," she stated.

Despite existing guardrails, the shift in coding intensity following the deployment of AI scribes is giving payers pause.

"Where it becomes an issue for payers is when it's clear that the technology is doing something to drive reimbursement up, but there's no clear path that the technology is doing anything to drive care to change as well," Chandler stated. "So, from a guardrails perspective, I think that's what payers are looking for; they're trying to understand."

More research is in the works to fully understand the implications of AI scribes on healthcare affordability. After all, the application of AI, particularly ambient listening and generative AI, during clinical encounters is still in its infancy. Many organizations like UAMS have only been using the tools for months.

However, these initial analyses highlight the vulnerability of the provider reimbursement system.

"As AI is changing our documenting practices," Oakes asked, "does that mean we potentially need to be reconsidering what the rules are that we use for determining the billing codes?"

Jacqueline LaPointe is a graduate of Brandeis University and King's College London. She has been writing about healthcare finance and revenue cycle management since 2016.