A few months ago, I argued that clinical documentation integrity (CDI) needed to stop reviewing claims one case at a time because payers have already moved to population-level statistical modeling, powered by artificial intelligence (AI), to flag outliers and deny claims at scale. I did not expect to be revisiting that argument so soon, or from the other direction. The federal government is now doing what payers have been doing for years, and it has more money, more staff, and considerably less patience.
Healthcare fraud has always driven the largest share of False Claims Act (FCA) recoveries, but fiscal year (FY) 2025 set a new bar. Of the $6.8 billion the Department of Justice (DOJ) recovered in FCA settlements and judgments last year, $5.7 billion came from healthcare, roughly 84 cents of every dollar. The statute behind all of this, the False Claims Act (31 U.S.C. § 3729), imposes treble damages for any claim submitted with actual knowledge, deliberate ignorance, or reckless disregard of its falsity, a standard that population-level data analytics make far easier to meet at scale.
On April 7, 2026, the acting attorney general folded DOJ Health Care Fraud Unit into a new National Fraud Enforcement Division, giving it centralized leadership, expanded staff, and a broader mandate. Two months later, the 2026 National Health Care Fraud Takedown charged 455 defendants, including 90 physicians and other licensed clinicians, across 56 federal districts and 45 states, in connection with more than $6.5 billion in alleged false claims.
To date the focus has been on other healthcare sectors (e.g., labs, telehealth, hospice, home health, etc.) but it is only a matter of time before hospitals are in the crosshairs.
Taylor Chenery, at Bass, Berry & Sims the attorney representing providers in government investigations, told Chief Healthcare Executive in a recent article that federal agencies are now “extremely data focused,” and are no longer waiting for a whistleblower to hand them a case. Chenery’s advice to health systems is what CDI professionals should already be following: know your data.
This should sound familiar. I have written before about how a whistleblower complaint against Integra built its case on a simple statistical threshold. A facility’s rate of a targeted diagnosis more than double the national average, or three percentage points higher than its peers, was treated as evidence of a false claim in that complaint. It was based entirely on data. Calculations that at one time took months to complete can now be completed in a matter of hours by federal agencies looking for fraud.
Moreover, outlier-detection logic that flags claims for commercial payers has become standard equipment for federal prosecutors, but the consequences of a federal audit are far steeper than a payer denial.
What does this change for CDI and coding departments? Arguably nothing about the work itself, and everything about the stakes. If a federal analyst pulled your hospital’s coding data for the last eight quarters, would documentation patterns explain the trend? Would query practice hold up as clinically driven, or would it look to an algorithm with no clinical judgment, indistinguishable from a pattern of upcoding? Does your staff understand how their workflow tools make recommendations? Are CDI and coding staff allowed to override recommendations from technology, and if so, what is that process?
Are known issues with technology’s query or coding recommendations tracked? What efforts are made to remedy them? Hospitals must prove awareness and action; complacency is not an option.
Using artificial intelligence (AI) tools may be viewed as fighting fire with fire in the war against payer friction, but CDI and coding activities must still be grounded in compliant practices. The hospital is responsible for how they use these tools. Staff must avoid simply accepting AI recommendations. Perform due diligence by reviewing the complete health record. Is this truly an opportunity or simply the byproduct of natural language processing (NLP) which can take words out of context?
Is there corroborating clinical evidence beyond one abnormal finding? What is the quality of corroborating clinical evidence? As the federal government joins this battle, hospitals may soon be fighting the war on both the compliance and revenue fronts.
CDI departments were originally implemented to find incremental revenue. Although their scope has grown with maturity, current economic pressures are pushing many departments back to their original mission with the help of AI tools. There is nothing inappropriate about being paid accurately for the resources used to care for Medicare and Medicaid beneficiaries when done compliantly.
An anomaly within claims data flagged by an algorithm is not, on its own, evidence of noncompliance. It should be confirmed by chart review. Hospitals need to be aware that sudden shifts in Case Mix Index (CMI) or CC/MCC capture due to implementing a new documentation tool or education strategy can look like a coding anomaly to software with no context for why the number moved. Every process implemented for documentation integrity is now a strategy a federal investigator could misread.
Health systems that want to avoid learning about a problem from a federal audit or subpoena should start where the government will, with their own claims data. That means running the same outlier analysis a federal analyst would run. Examine trends by MS-DRG, physician, and service line, at least quarterly. Investigate anomalies before they become someone else’s discovery request.
This also means revisiting the Program for Evaluating Payment Patterns Electronic Report (PEPPER); the tool is not as sophisticated as AI analysis, but any area where a hospital sits as an outlier is still worth confirming for compliant, defensible coding.
Moreover, technology should not replace clinical or coding judgment. Staff cannot blindly follow AI recommendations. CDI and coding leaders need to know how recommendations are generated; staff need to be able to override technology when necessary; and known issues with query or coding logic must be tracked and corrected. A workflow tool that functions as a black box is a liability whether it ever produces an error or not.
The same discipline applies to documentation and query practice generally. Changes in process, whether from innovative technology, a new education initiative, or a consulting engagement, should be logged, defined in policy, and monitored for staff adherence. Be proactive so a shift in CMI or CC/MCC capture already has a documented, defensible explanation before a federal auditor ever asks for one.
The U.S. Department of Health and Human Services (HHS) Office of Inspector General (OIG) compliance program guidance has long framed this type of internal auditing as a baseline expectation, not an aspirational goal. That expectation is especially relevant to CDI, where the clinical narrative behind a coding trend is precisely what a data-only audit cannot capture. What story is your organization’s data telling?
None of this work is new for CDI professionals, who have spent careers translating clinical nuance into coded data. What is new is who sits at the other end of that data, and how quickly they can act on it. The DOJ no longer needs a whistleblower to notice when your hospital’s numbers look different from everyone else’s. It only needs your claims file, which they can easily access.
Organizations that actively monitor their own data, rather than managing it reactively, are the ones that will spend the rest of 2026 explaining their anomalies on their own terms, instead of the government’s.
