Can AI accurately code secondary diagnoses? It can find possible conditions in a medical record, but finding a condition is only the first step. A qualified coder must determine whether the provider’s documentation and inpatient coding guidelines support reporting it.
That distinction matters. The Blue Cross Blue Shield Association estimates that increased coding of medically complex hospital stays added $942 million in spending from 2023 through 2025, with most of the increase tied to secondary diagnoses. Its findings raise a question for hospitals using AI assisted coding: Who reviews each suggested diagnosis before it reaches the claim?
A new analysis from the Blue Cross Blue Shield Association (BCBSA) puts a big number on a question coding teams deal with every day: Just because a condition appears somewhere in the record, does it belong on the claim?
BCBSA’s analysis looked at Blue Cross and Blue Shield claims from 2023 through 2025. Three findings stand out:
Those numbers should get the attention of anyone responsible for coding accuracy. They also need some context. This is an analysis of claims, not a chart-by-chart determination that every additional diagnosis was wrong, and a patient does not have to receive a particular treatment for every secondary diagnosis to be valid. Clinical evaluation, diagnostic work, a longer stay, or increased monitoring may also make a condition reportable under the official inpatient coding guidelines.
That’s exactly why reporting decisions need a qualified human coder.
AI can scan a chart and flag a lab result, a phrase in a note, or a condition mentioned earlier in the stay. That may be useful as a prompt to look closer. It is not the end of the coding process.
A coder has to ask: Is the diagnosis supported by the provider’s documentation? Did it affect this admission? Does it meet the reporting criteria? Is there conflicting information that needs a query? What do the coding guidelines say about this particular case?
Take an abnormal lab value. An AI tool may spot it instantly and suggest a diagnosis. But the value alone does not establish that diagnosis or make it reportable. The coder needs to review the record, understand what the provider concluded, and determine whether the condition meets the rules for that setting. Sometimes the right next step is a compliant query. Sometimes the right answer is to leave the code off.
The reverse matters, too. A legitimate secondary diagnosis can be missed when someone looks only for an obvious treatment. A patient may have received extra monitoring, a diagnostic evaluation, or a change in the care plan. A skilled coder can connect those details to the documentation and the applicable guidelines.
Hospitals have good reason to make sure every reportable condition is captured. Missing a supported diagnosis can misrepresent the patient’s complexity and affect reimbursement and quality data. Adding one that does not meet the criteria creates a different problem: a claim that is harder to defend, a greater risk of denials, and an inaccurate picture of the care provided.
That is the point BCBSA’s findings bring into focus. More codes do not automatically mean better coding. The goal is a claim that accurately reflects the documented care, with a clear reason for every diagnosis reported.
At HIA, this is the work our coders and auditors do: review the full record, apply the guidelines, and use their judgment when the answer is not obvious. Technology can help surface a possible condition. A human needs to decide whether it belongs on the claim—and be able to explain why.
If your team is using AI to suggest diagnoses, now is a good time to look at what happens after the suggestion. Who validates it? How are questionable secondary diagnoses reviewed? Can your team show the documentation and coding rationale behind the final code? Those answers matter more than how many conditions a tool can find.
For more than 30 years, HIA has been the leading provider of compliance audits, coding support services and clinical documentation audit services for hospitals, ambulatory surgery centers, physician groups and other healthcare entities. HIA offers PRN support as well as total outsource support.
The information contained in this coding advice is valid at the time of posting. Viewers are encouraged to research subsequent official guidance in the areas associated with the topic as they can change rapidly.