The Blue Cross Blue Shield Association released an analysis on September 24 estimating that rising inpatient coding intensity added approximately $942 million in costs to BCBS member plans between 2023 and 2025.
About $653 million — roughly 70 percent — was tied to secondary diagnoses that shifted cases into higher-paying diagnosis-related groups.
BCBSA argues that expanding use of AI-enabled coding tools appears to be accelerating the pattern. The American Hospital Association disputes that interpretation.
BCBSA’s findings
The analysis examined de-identified Blue Cross and Blue Shield claims and found that the proportion of inpatient cases classified as medically complex increased from roughly 37 percent at the beginning of 2023 to 40 percent by the end of 2025.
| Finding | BCBSA result |
| Medically complex cases, early 2023 | ~37% |
| Medically complex cases, late 2025 | ~40% |
| Estimated additional BCBS spending | $942 million |
| Amount linked to secondary diagnoses | ~$653 million |
| Additional complex cases from secondary diagnoses | 55,000+ |
| Approx. added reimbursement per excess complex case | ~$11,800 |
Source: BCBSA September 2026 analysis
The central finding was not simply that hospitals billed more.
BCBSA argues that diagnostic complexity increased much faster than observable treatment intensity — patients were increasingly assigned secondary diagnoses that moved their stays into higher-severity DRGs, even though measures like ICU utilization, transfusions, reoperations and length of stay did not rise correspondingly.
BCBSA calls that gap “clinical discordance.”
Secondary diagnoses change payment, but how?
Hospital inpatient reimbursement commonly relies on diagnosis-related groups.
Cases can fall into different severity levels depending partly on whether the record contains qualifying secondary diagnoses classified as a complication or comorbidity (CC) or major complication or comorbidity (MCC).
A documented secondary condition can change which severity tier a case falls into. That does not automatically make the coding improper — the question is whether the diagnosis is clinically supported and properly documented.
THE DISPUTE
Two readings of the same coding data
BCBSA position
AI finds more secondary diagnoses that raise DRG severity
Coded complexity is rising faster than treatment intensity
Hospitals in the highest coding quartile showed similar or lower resource use than peers
Result looks more like coding escalation than sicker patients
AHA position
Inpatient acuity has genuinely increased (5% case-mix rise, 2019-2024)
Lower-acuity care continues migrating to outpatient settings
AI helps capture legitimate complexity that historically went undocumented
Better documentation is not the same as upcoding
Sources: BCBSA September 2026 analysis; AHA fact sheet, August 2026
The caveat in the analysis (yes, there’s one)
BCBSA’s research demonstrates an association between rising coding intensity, greater AI adoption, and higher reimbursement.
It does not directly establish that AI software caused every additional diagnosis or that $942 million represents improper billing.
Several limitations are worth noting, including:
The analysis relied primarily on claims data rather than complete clinical documentation — reviewing medical charts would provide a stronger basis for determining whether individual diagnoses were clinically justified
Also, BCBSA did not characterize the full $942 million as fraud! Moreover, A secondary diagnosis that raises a DRG’s severity can be entirely legitimate when clinically supported
And the study is the second BCBSA AI-coding analysis in 2026 — a March report estimated $663 million in inpatient and $1.67 billion in outpatient spending linked to similar patterns
The hospital counterargument
The American Hospital Association released a detailed fact sheet in August specifically disputing claims that provider AI tools are improperly driving coding intensity. AHA’s counterargument has several parts.
Higher-Acuity Inpatient Population
Lower-acuity care continues shifting to outpatient settings, leaving hospitals with a more medically complex inpatient population.
Aging & Chronic Illness
The population is aging, while chronic illness is becoming more prevalent and contributing to greater clinical complexity.
Case-Mix Index Increased
AHA’s analysis with Vizient found that hospital case-mix index increased by roughly 5% between 2019 and 2024.
Better Documentation & Coding
Improved documentation and coding technology can capture conditions that previously went unrecorded.
AI Still Requires Human Validation
AI can support coding and documentation workflows, but human validation remains necessary to ensure the information is accurate and appropriately supported.
AHA’s position is that an increase in documented severity does not inherently equal upcoding — it may mean hospitals are documenting patients more accurately.
The broader pattern
The story fits a larger dynamic. Providers are deploying AI to improve documentation and capture legitimate reimbursement. Insurers are using AI and automated analytics to scrutinize claims and identify suspected overbilling. The financial incentives are substantial on both sides.
For revenue-cycle and clinical-documentation-integrity teams, the practical consequence is that AI-generated diagnoses need defensible clinical support.
AI Suggestion ≠Reportable Diagnosis
An AI-identified possible diagnosis and a diagnosis that the medical record clinically supports for reporting are not automatically the same thing.
Human Review Still Matters
As AI handles more of the initial documentation work, clinical validation, query quality, coder review, and post-bill auditing become increasingly important.
The Coding Debate Continues
The coding-intensity debate is continuing, with BCBSA building on earlier analyses to examine how AI-enabled coding may affect diagnosis patterns and healthcare costs.
As payer analytics grow more sophisticated at comparing hospitals, DRG distributions, and treatment patterns, unusually rapid increases in CC/MCC capture may attract scrutiny — even when the diagnoses themselves are technically codable.
