EDITOR’S NOTE: Senior healthcare analyst Frank Cohen begins a three-part series about auditing in the algorithmic environment.
Something has changed in healthcare compliance, and if you work in coding, clinical documentation integrity (CDI), or revenue cycle, it is already affecting your daily work, whether you have noticed it or not.
Here is what happened. Providers started using artificial intelligence (AI) to help write clinical documentation. At the same time, the contractors who audit Medicare claims began using AI to identify problems in billing data. Both sides are now running the same kinds of tools – pattern recognition, anomaly detection, natural language processing (NLP) – just pointed in opposite directions. One side is trying to get the documentation right; the other side is trying to find where the documentation or the coding went wrong.
Think of it like a chess match: two players sitting at the same board, using the same kinds of engines, but striving for opposite objectives. The provider side is optimizing for accurate reimbursement; the enforcement side is optimizing to detect inflation in reimbursements. Same board. Same technology. Opposite goals.
If that sounds abstract, let me make it concrete.
Every Chart Is Now Being Read Twice
Every chart you touch is being evaluated twice. The first time is by whatever tool helped create it: the ambient documentation system that drafted the note, the coding software that suggested the CPT codes, the CDI query that surfaced an additional diagnosis. That first reading is on your side. It is trying to help.
The second reading happens later, and you will never see it. Somewhere in a Medicare contractor’s data warehouse, a model is looking at the claims your organization submits – not one chart at a time, but in bulk. It compares your billing patterns to those of every other practice in your specialty and region. It is looking for anything that sticks out. An evaluation and management (E&M) distribution that skews higher than your peers. A modifier utilization rate that jumped after you deployed a new documentation tool. A sudden increase in the capture of certain chronic conditions.
This is not speculation. Over the last two years, every major program integrity contractor – Uniform Program Integrity Contractors (UPICs), Recovery Audit Contractors (RACs), Medicare Administrative Contractors (MACs), the Medicare Advantage (MA) RADV Risk Adjustment Data Validation (RADV) apparatus – has incorporated some form of machine-driven analytics into its case-selection process. The names of the tools vary. The approach does not. They are looking at the shape of your data and working backward to the charts.
Why This Matters to You Personally
If you are a coder, you are the person closest to the data. You see the codes going out. You see the documentation coming in. You are in a better position than almost anyone else in your organization to notice when something shifts: when the AI starts suggesting higher-level codes more often, when a new modifier pattern emerges, when the documentation starts reading differently than it used to.
If you are a CDI specialist, the chronic conditions your AI tools are helping you surface – diabetic complications, vascular disease, CKD staging, heart failure subtypes – are the exact conditions that contractor models are being trained to validate. You are not just documenting better; you are documenting into a system that is actively checking your work.
That is not a reason to stop using these tools. It is a reason to understand that the environment around you has changed. The audit that used to start with a single flagged chart now starts with a statistical model that has noticed something about your organization’s data. By the time a human auditor opens a chart, the suspicion has already been generated by an algorithm.
The Structural Mismatch
Here is the takeaway I want every reader to take away. Most compliance programs are still built around chart-level review. Pull a sample, audit the records, and educate the coder. That process still has value. But it was designed to catch a different kind of problem than the one the modern audit is looking for.
The modern audit examines population-level patterns. It is comparing your distributions to your peers’ distributions. It is looking at trends over time. A chart-level compliance program cannot see those things, just as a microscope cannot see a satellite photo. The instrument is fine. The scale is wrong.
In Part Two of this series, I will show you exactly how a practice that is doing everything right at the chart level can still face a significant overpayment demand, and what that means for the people on the front lines of coding and documentation.
