The Variable Sample Size Problem: Why RADV Audits Are Harder to Predict Than Ever

Why Sample Size Variability Changes Everything

Under previous RADV frameworks, CMS used relatively predictable sample sizes. Plans could estimate their exposure based on historical patterns. The current framework changed that. CMS now selects between 35 and 200 enrollees per contract, and the agency hasn’t published the criteria it uses to determine where a specific contract falls within that range.

This variability creates planning uncertainty. A plan preparing for a 35-enrollee sample faces a fundamentally different operational challenge than one facing 200. The documentation retrieval effort, the evidence validation workload, and the rebuttal preparation scale all differ substantially. Plans that staff and budget for the minimum may be overwhelmed if they receive the maximum.

The strategic implication is that plans must prepare for the upper end of the range. Building audit response infrastructure capable of handling 200 enrollees with full MEAT validation and evidence packaging ensures readiness regardless of where the actual sample falls. Preparing for 35 and hoping for the best is a gamble with quantifiable downside.

What Drives the Sample Size Decision

CMS hasn’t published an explicit formula, but the agency’s public statements and enforcement patterns suggest several factors influence sample size selection. Plans with higher coding intensity scores (risk scores that exceed clinical complexity indicators) are likely candidates for larger samples. Plans with prior audit histories showing elevated error rates may face expanded scrutiny. Plans operating in geographic regions with known coding pattern anomalies may receive larger samples to improve statistical reliability.

The AI-assisted targeting CMS now uses adds another dimension. When population-level analysis identifies specific patterns in a plan’s submitted data (concentration in high-value HCC categories, asymmetric add-to-delete ratios, high rates of unlinked chart review diagnoses), the system may flag the contract for a larger sample to investigate those patterns with greater statistical confidence.

Plans can’t control the sample size they receive. But they can control the factors that influence it. Programs that produce balanced coding profiles, defensible documentation, and two-way review output generate fewer of the statistical signals that warrant expanded samples.

Preparing for the Unknown

The practical response to variable sample sizes is worst-case operational readiness. Audit response teams should be resourced and trained for 200-enrollee reviews. Evidence trail systems should be capable of producing complete MEAT-validated documentation packages at that scale within the five-month submission window. Mock audit exercises should simulate upper-range samples to stress-test the response infrastructure.

Plans that treat every submitted code as a potential audit target, validating documentation and preserving evidence trails at the time of coding, effectively pre-build their audit response. When the sample arrives, regardless of size, the evidence already exists. The team validates and packages rather than scrambles and assembles.

The New Planning Reality

Variable sample sizes make radv audits unpredictable at the contract level. Plans can’t know in advance whether they’ll face a focused 35-enrollee review or an extensive 200-enrollee examination. The only preparation strategy that works across the full range is one that treats every code as audit-ready from the moment it’s submitted. Plans that build this continuous readiness spend less time reacting to audit notifications and more time demonstrating the defensibility they built into their coding process from the start.