Ask five risk adjustment leaders what “OnePass” means, and you will get five different answers. Some hear a single coder making a single judgment with no safety net. Others hear a vendor pitch to cut QA. Neither is what OnePass means, and the confusion is worth clearing up before any talk of cost or compliance, because the value of a single pass only makes sense once the terms do.
So this article starts where the industry conversation should: with definitions.
The Terms: OnePass, MultiPass, Multi-Level
OnePass is one AI processing cycle over the chart. The chart is analyzed once, end-to-end, and produces a complete set of coding results. That is the pass.
MultiPass review means running the same charts through a second, independent system or vendor. A different NLP engine, a different coding operation, a separate project. The chart is reprocessed from scratch, and the two sets of results are then reconciled.
Multi-level review is the human validation that happens inside a single pass. A coder validates the AI’s output. One or more auditors review the coder’s work. The health plan’s own QA team reviews the vendor’s deliverable. None of these steps reprocess the chart. Each one verifies work that already exists.
The distinction matters because “OnePass” is often misheard as “one person checks the chart.” It means nothing of the sort. OnePass means the chart is processed by AI once. Within that single pass, a RAAPID OnePass workflow supports the validating coder plus a configurable number of auditor levels, up to four. As many as five people can review a chart before it reaches the plan, and the plan’s own QA still sits on top of that. What OnePass removes is the second processing cycle, not the review layers.
Why Health Plans Adopted MultiPass Review
Health plans did not build multipass workflows out of habit or excess caution. For most of the industry’s history, multiple passes were the only rational response to the tools available.
Risk adjustment review began as 100% manual work. Then legacy NLP arrived, and its out-of-the-box accuracy was limited. No single vendor’s NLP could reliably surface every supportable add and every warranted delete, and the humans validating its output could not catch everything the engine missed either. Confidence in any one review cycle was low, so plans bought insurance the only way they could: independence. Run the same charts through a second vendor, a second engine, a second team, and let the second net catch what fell through the first.
Every additional pass existed to compensate for the limited confidence in the pass before it. Given the technology of the time, that was sound program design.
It is worth being precise about what multipass looks like in practice, because the popular image of “coding everything twice” is not how most programs run. Plans typically send only a slice of the population to the second vendor. In our experience, the share varies widely; some programs route as little as 5% of charts, others closer to half, set by batch quality, budget, and how much confidence the first pass earned. The full charts are not the full cost. The cost is that a second pass is operationally a second project: a second round of data sharing and BAAs, a second vendor relationship to manage, a second timeline, and a reconciliation effort at the end to merge two result sets produced under two different sets of guidelines. That overhead applies whether the second vendor sees 5% of charts or 50%.
Want the compliance case first?
What Changed: The Technology Caught Up
The case for a second pass rests on one assumption: that the first pass missed things a second system would find. Neuro-symbolic AI removes the basis for that assumption, and the mechanism matters more than the label.
The neural layer reads the chart the way an experienced reviewer would, surfacing candidate diagnoses along with the clinical evidence that supports them. The symbolic layer then tests every candidate against coding logic, clinical rules, and CMS-recognized source requirements before a human ever sees it. A candidate that lacks support does not survive that test.
Three properties of this design are what eliminate the need for a second processing cycle:
Two-way coding in the same pass. The workflow both adds diagnoses the record supports and flags diagnoses the record does not support for removal. Historically, catching deletes was a major reason for a second review, because add-focused first passes left unsupported codes sitting in the data. When adds and deletes are evaluated together in one cycle, that job is already done. The stakes on this are no longer theoretical. In March 2026, a major Medicare Advantage insurer paid $117.7 million to settle False Claims Act allegations over a chart review program that added diagnoses and never removed unsupported ones. OIG’s February 2026 compliance guidance flagged the same pattern: failing to remove unsupported codes is a compliance failure in its own right. Regulators are no longer only judging coding results. They are judging program design, and a one-directional program is a design that cannot explain itself.
Evidence linked at the point of coding. Every surfaced code carries its own justification: the MEAT evidence showing the condition was Monitored, Evaluated, Assessed, or Treated; where that evidence sits in the record; and why the code qualifies. The reviewer sees the reasoning, not just the recommendation. A chart questioned later already carries its defense.
Validation before the human, not after. Because the symbolic layer screens every candidate against coding and clinical rules first, the coder is validating pre-tested output rather than hunting for errors in raw suggestions.
The conclusion this builds toward is simple: after one evidence-linked, rule-validated, two-way cycle, a second engine has little left to surface that the first cycle did not already test and document. Whatever residual judgment remains is human judgment, and that is precisely what the review levels inside the pass are for. Reprocessing the chart through a second system does not address it. It adds cost, coordination, and a second set of results to reconcile.
What OnePass Does Not Replace
This is the part most often misunderstood, so it deserves its own section.
OnePass removes redundant coding passes. It does not remove QA. The coding vendor’s QA review of the coders’ work stays. The health plan’s QA review of the vendor’s deliverable stays. Multi-level review is not a casualty of OnePass; it lives inside it.
Here is what that looks like in a running program. The AI processes the chart once. An expert coder validates the output and makes the call, because final authority stays with the human. Behind the coder sit up to four configurable auditor levels. The completed, evidence-backed deliverable then goes to the health plan, where the plan’s own QA and compliance teams apply their existing multi-level oversight. Every one of those checkpoints operates on the same single pass. Nothing is re-coded from scratch, and nothing needs to be.
The accuracy numbers reflect this layered design. On its own, RAAPID’s AI reached 92% coding accuracy in an independent evaluation. With coder validation and multi-level audit operating inside the workflow, accuracy moves past 98% on internal benchmarks, with an evidence trail behind every submitted code. The point of those levels is not to re-find diagnoses. It is to certify, at each checkpoint, work that already carries its proof.
OnePass, Configured to Your Workflow
A single pass does not mean a single, rigid process. OnePass is designed so each health plan implements the review structure it wants inside the one coding cycle.
Plans configure how many audit levels sit behind the coder, from one to four. They configure the audit sampling depth, and in practice it varies with coder proficiency: a highly experienced coder’s charts might route 20% to second-level audit, a newer coder’s 50% or more, and programs with strict requirements run 100% audits before anything leaves the vendor. Those thresholds are set by coding leadership and the plan, not dictated by the tool. Plans also decide where their internal QA sits, reviewing the deliverable through their existing governance without triggering a new coding cycle.
RAAPID’s policy management layer applies each plan’s coding guidelines, custom rules, and ontologies within the workflow itself, so the pass reflects the plan’s standards rather than a generic default the plan has to audit around.
The result is the structure of a mature review program: coders, auditors, vendor QA, and plan oversight without the cost structure of a second vendor and a second project.
Now the Cost and Compliance Case
With the terms clear, the economics are easier to weigh honestly.
MultiPass review carries costs that never touch a diagnosis: duplicate data sharing agreements, two vendor relationships, two timelines, drifting guidelines and formats between reviewers, version control as its own job, and a reconciliation effort whenever two teams read the same chart differently. Turnaround slows because a chart is not finished when it is coded. It waits for the next queue and the next handoff. Meanwhile, the most experienced reviewers spend their time on charts that were already settled, and the accuracy gained by each additional cycle keeps shrinking.
The compliance picture has moved even further from the multipass model. Risk adjustment has shifted from finding diagnoses to proving them. CMS is moving to audit every eligible Medicare Advantage contract each year, not a sample. OIG has stated that failing to remove unsupported codes is a compliance concern, and its recent audits found that most sampled high-risk diagnoses across several plans were not supported by the record. The CY2027 Rate Announcement, finalized in April 2026, made that concrete. Beginning in CY2027, CMS will exclude diagnosis information from unlinked chart review records diagnoses not tied to a specific beneficiary encounter from risk score calculation. Plans may still submit them; they simply will not count. A narrow exception preserves credit for beneficiaries who switch between MA organizations, and diagnoses coded from audio-only encounters under modifiers 93 and FQ are excluded as well. The direction is unambiguous: a diagnosis that cannot be linked to an encounter and evidenced in the record is a diagnosis that no longer carries payment, regardless of how many review cycles surfaced it.
Against that standard, the number of passes is the wrong measure. A chart can pass through three review cycles and still carry a diagnosis no one can defend, because each pass checked for accuracy rather than building evidence. A code without documentation, clinical support, and justification is a liability no matter how many times it was reviewed. Defensible risk adjustment is not a volume problem. It is a design problem, and the design that solves it is a pass that attaches the evidence as it codes.
The question for risk adjustment leaders is no longer how many passes you can afford. It is why you are still paying for a second one.
Book a demo and see how one evidence-linked pass, with your QA levels configured inside it, performs on your own charts.
Frequently Asked Questions
What is the difference between a multipass review and a multilevel review?
A multiPass review reprocesses the same charts through a second, independent system or vendor. Multi-level review is the human validation inside a single pass: coder validation, configurable auditor levels, and the health plan’s own QA.
What is MultiPass risk adjustment coding?
MultiPass risk adjustment coding runs the same charts through a second, independent coding system or vendor to catch missed diagnoses. Plans typically send only 5% to 50% of charts to the second vendor while carrying the full overhead of a second project.
Does OnePass eliminate QA review?
No. OnePass eliminates repeated coding cycles, not quality assurance. The coding vendor’s QA review and the health plan’s QA review both remain, with audit sampling configurable up to 100%.
Why did health plans adopt multiPass review in the first place?
Legacy NLP tools had limited accuracy and could not reliably surface every add and delete, so a second independent pass was a rational safeguard. Neuro-symbolic AI now attaches evidence and rule-based validation in the first cycle, removing that gap.
What is OnePass of Defensibility?
OnePass of Defensibility is RAAPID’s approach to risk adjustment coding: one AI processing cycle with expert coder validation and configurable multi-level auditing. Every submitted diagnosis carries clinical evidence and code justification, making it audit-ready without a second coding cycle.
Why is MEAT documentation important in risk adjustment?
MEAT documentation shows a condition was monitored, evaluated, assessed, or treated during the encounter. It provides the clinical evidence that makes HCC codes defensible during RADV and other compliance audits.
What is RADV audit readiness?
RADV audit readiness is a health plan’s ability to defend submitted diagnosis codes with complete documentation and evidence during CMS audit reviews.
Read more: RADV Audit Checklist
References
[1] OIG Medicare Advantage compliance audit, A-07-22-01207
[2] OIG Medicare Advantage compliance audit, A-03-22-00004:
[3] OIG Medicare Advantage Compliance Program Guidance (Feb 2026):
[4] DOJ Aetna/CVS settlement ($117.7M):
[6] MedPAC March 2026 report (MA ~14% above FFS):
[7] AAPC MEAT documentation reference:
[8] 2027 Medicare Advantage and Part D Rate Announcement | CMS
