Chetan Parikh, Founder and CEO, RAAPID
Since 2010, I’ve been working at the intersection of healthcare technology, building NLP platforms, collaborating with programmers, and digging deep into how risk adjustment works. The ground underlying retrospective chart review has fundamentally changed. This is not speculation, but direct insight from risk adjustment leaders, and is certainly reflected in recent OIG updates.
Retrospective programs were once a staple of risk adjustment, with plans relying on them to recover undocumented diagnoses and fill critical gaps. Those days are over. Federal regulators are drawing the line. Programs that simply add diagnostics without removing unsupported ones are no longer considered compliance tools. These are now clear signals of intent to inflate payments, and regulators are responding accordingly.
OIG has made its position undeniable
In the 2024 MA Industry Compliance Program Guidance, the OIG explicitly identifies conduct that is currently considered potentially fraudulent by federal investigation. One pattern is unmistakable: CMS fails to remove previously submitted diagnostic codes even after chart review reveals them to be unsupported or invalid.
A review process that finds unsupported code and then leaves it alone is not a minor oversight, but a clear compliance violation. When this failure becomes systemic, it forms the basis of regulatory enforcement.
OIG also called for the use of AI to encourage providers to add diagnoses that patients have not received. These are not theoretical concerns. These are the exact patterns that are subject to regulatory scrutiny.
The Department of Justice showed what law enforcement looks like
In January 2026, Kaiser Permanente affiliates paid a historic $556 million to resolve False Claims Act claims. This is the largest Medicare Advantage settlement ever. The Department of Justice made its case clear. Kaiser systematically collected historical records of unsubmitted diagnoses and pressured health care providers to add them through appendices, sometimes long after the patient was seen. This approach resulted in nearly $1 billion in unsupported payments related to approximately 500,000 diagnoses.
Cigna’s $172 million settlement in September 2023 followed a similar strategy, using chart reviews to secure additional payments while intentionally ignoring overpayments revealed in the same reviews.
What was common in both cases was the program’s architecture, not a lack of documentation. These one-way systems acted as a ratchet, always increasing the risk score and never adjusting it downward. In Kaiser, the Justice Department argued that the company was well aware that its addenda practices were widely illegal and chose to ignore repeated warning signs. If a system is designed to only work in one direction, regulators will target the design itself as the root problem.
$9.2 billion alarm bell
Long before the Kaiser settlement, the OIG issued a clear warning. A September 2021 assessment (OEI-03-17-00474) found that 162 MA organizations generated $9.2 billion in risk-adjusted payments from diagnoses that had no supporting service records and only appeared on medical record reviews and health risk assessments. 20 companies held a disproportionate share. The OIG’s message was clear. Some organizations used medical record reviews and HRAs to maximize payments rather than to ensure enrollees received the care they needed.
On January 26, 2026, CMS indicated its intention to close this loophole. The CY 2027 Advance Notice proposes to completely exclude diagnoses from unlinked chart review records from risk score calculations, and CMS estimates that this policy will reduce MA payments by $7.2 billion. Not only is the value of add-only retroactivity decreasing, but CMS is preparing to eliminate it completely.
Department of Justice sets new standards for enforcement
If your retroactive program only adds diagnoses but does not remove them, you are operating a system that is directly targeted by federal regulators. This is not speculation, but the OIG’s official position, reinforced by hundreds of millions of dollars worth of settlements with the Department of Justice.
The real question is not, “Have we found more HCC?” It is as follows: “Can you support all the diagnoses we have presented? protect What is our code? ” The retrospective process should be a two-way street. This means adding supported code, removing unsupported code, and documenting every decision with enough evidence to withstand scrutiny from all directions.
What steps should we take going forward?
I speak weekly with risk adjustment leaders who know that the program needs to change.
The first step is to look at your retrospective workflow and ask one question: Is the code correct? defensible?Delete? If all chart reviews only add diagnoses but cannot remove them, there is a structural problem that cannot be solved even with high accuracy rates. OIG elaborated on that. The Justice Department put a dollar figure on it.
Next, look at your technology. If an AI suggests HCC and cannot show clinical evidence to support it, it is not decision support. It’s a responsibility. For years I’ve seen programmers struggle with tools that have answers but no reasoning. Every recommendation the system makes should have a transparent path back to the clinical record that auditors can follow without questions asked.
Finally, shift investments to the point of care. The safest diagnosis is confirmed by a healthcare provider during the visit and documented in real time. Retrospectives still have a role, but they are more supportive, such as validation, cleaning, and audit preparation. The path to growth is pierced through encounters.
The federal government will no longer ask if your coding is accurate. I’m asking if that’s the case. defensible. Append-only programs, by design, do not pass this test.
