The Regulatory Reckoning
Retrospective chart review has been the backbone of Medicare Advantage risk adjustment for two decades. Health plans send coders into historical charts to identify diagnoses that were clinically present but never submitted to CMS. It’s legitimate work when done properly. The problem is how the industry has done it.
For years, the dominant model was add-only. Find codes. Submit codes. Increase RAF scores. Revenue goes up. The incentive structure rewarded volume over accuracy, and regulators noticed. The DOJ’s $556M settlement with Kaiser, the $117.7M Aetna settlement in March 2026, and accelerating OIG audits all point in the same direction: add-only retrospective programs are now a federal red flag.
The OIG’s February 2026 Medicare Advantage Industry-wide Compliance Program Guidance explicitly flagged add-only chart reviews and health risk assessments that generate diagnoses never considered in patient care as high-risk practices. CMS wants supplemental data used as a two-way street: adding missed diagnoses and removing unsupported ones. Plans operating one-way programs are on notice.
Why Add-Only Programs Fail
The structural problem with add-only retrospective review is that it creates an asymmetric risk profile. Codes go in. Codes never come out. Over time, a plan’s submitted diagnoses drift away from clinical reality. Members accumulate HCCs that may have been accurate at one point but no longer reflect their actual condition. At the population level, this looks exactly like the coding intensity pattern CMS is targeting.
CMS sees this pattern in the data. When a plan’s risk scores consistently rise while clinical outcomes stay flat, that’s a signal worth investigating. The Aetna DOJ settlement specifically targeted an add-only chart review program from payment year 2015 and false morbid obesity codes from 2018 through 2023. The whistleblower was a former coding auditor who saw the problem from inside the organization and brought it to the government.
The financial exposure is significant and growing. Kaiser’s $556M settlement covered diagnoses that the DOJ complaint alleged lacked clinical support across approximately 500,000 members. Aetna paid $117.7M. These aren’t theoretical risks or warning letters. They’re settled cases with public records, public dollar amounts, and real consequences for the organizations involved.
What Two-Way Retrospective Actually Looks Like
A defensible retrospective program operates the way CMS describes its supplemental data process: as a two-way street. Coders review charts to find missed diagnoses (adds), but they also flag unsupported codes for removal (deletes). Every submitted HCC gets validated against MEAT-based clinical evidence in the chart. Codes without adequate support get pulled before they become audit liabilities sitting on the plan’s balance sheet.
This requires AI that can scan a chart and identify both what’s present and what’s missing. Explainable AI tools link each coding decision to specific lines in the clinical note, showing which MEAT elements are satisfied and which aren’t. The coder still makes the final call, but they’re working with evidence and context instead of assumptions and time pressure.
It also requires a cultural shift that goes beyond technology. Coding teams have been measured and rewarded for years based on codes added. The number of adds was the productivity metric, the bonus driver, the performance indicator. Shifting to a model where removing an unsupported code counts as a compliance win, not a revenue loss, takes deliberate leadership and incentive redesign. Plans that can’t make that cultural change will keep running programs that look like revenue engines to regulators.
The Future of Retrospective
Retrospective review isn’t going away. Plans will always need to reconcile what was documented with what was submitted. But its purpose is changing. It’s no longer an offensive growth engine. It’s a defensive compliance layer designed to clean existing data and ensure what’s already been submitted can survive scrutiny.
The plans that understand this distinction are already restructuring their programs around Retrospective HCC Coding workflows that validate, clean, and defend every submitted diagnosis, treating accuracy and deletion as equal priorities to capture. That’s the model that survives audits, DOJ investigations, and the next five years of increasing regulatory pressure. Everything else is borrowed time.
