Blue Cross says hospital AI coding tools cost its plans $942 million from early 2023 to late 2025, with no matching change in patient acuity. Medicare’s public claims can’t test that claim directly. They can show the baseline the fight starts from. Here are nine years of Medicare inpatient stays in 16 condition families, and the share coded at the top “with MCC” (major complication or comorbidity) tier.
Each line is one condition family (a DRG split into “with MCC” and lower tiers). y = share of that family’s Medicare discharges coded at the MCC tier. The heavy red line is the 16-family total at a fixed 2022 family mix, so a shift between conditions can’t move it. The shaded band marks 2023–24, the stretch that overlaps BCBSA’s window. Hover a line for values.
All Medicare fee-for-service inpatient discharges in the public file, all DRGs. The denominator shrank by a third.
The climb came first
In Medicare, the coded-complexity climb had mostly happened before anyone was selling AI coding. The median family gained … points from 2016 to 2022, then … from 2022 to 2024. The fixed-mix total followed the same shape: …, then …. Whatever AI coding did to Blue Cross’s commercial claims, it hadn’t started a new climb in Medicare fee-for-service by the end of 2024. Medicare’s own audit contractors may be one reason, but that is a guess this chart can’t test.
The denominator moved too
Medicare FFS inpatient discharges fell from … to …. Some of that is Medicare Advantage growth, and some is short stays moving to observation status. When the simpler patients leave the inpatient FFS pool, the share coded “with MCC” rises even if nobody codes differently. That is the problem at the heart of BCBSA’s claim. A rising complexity share proves nothing by itself. BCBSA’s argument depends on the part this file doesn’t have: patient acuity didn’t change, and neither did treatment.
“If it was worth coding, there should have been something done.”— Razia Hashmi, MD, BCBSA, via Becker’s Hospital Review
Two lines to clickMajor bowel procedures is BCBSA’s example: in commercial claims, the share at the highest complexity level went from 10.2% to 22.7%. In Medicare the MCC share went from … (2016) to … (2024), and … of that came after 2022. Red blood cell disorders is the closest this file gets to the anemia fight. Here anemia is the principal diagnosis, not the secondary one BCBSA flagged. Its MCC share nearly doubled, from … to …. A claims file can’t say whether anyone ordered the transfusion.
Where this is thin — read before quoting
Different payer, different window. BCBSA analyzed Blue plans’ commercial claims from early 2023 to late 2025. This is Medicare fee-for-service through 2024. Flat Medicare lines don’t refute BCBSA. They show that the Medicare trend and the commercial trend could be different stories.
“With MCC” is not the same as “complex.” BCBSA’s 37%→40% figure counts claims coded “medically complex,” and its definition may be broader than one tier (for example, it may include CC). This chart tracks only the top MCC tier.
Suppression. CMS drops hospital-DRG cells with 10 or fewer discharges. The small, lower-tier cells are the ones most likely to disappear, which nudges every share up by a similar amount each year.
Excluded on purpose: heart failure (its MCC share goes from 56% to 98% between 2016 and 2024, a jump too large to read as a coding or acuity trend, so we left it out) and respiratory infections (COVID reshuffled which DRG pneumonia patients landed in from 2020 to 2022). COPD’s 2016→2017 jump also looks like a definitional shift.
No acuity or treatment data. This file has discharges and payments. It has no labs, transfusions, orders or severity scores, so it can’t separate “sicker” from “documented as sicker.” That’s the whole dispute.