CurrentMedicarePolicy N/A

HHS-RADV error rate calculation and methodologies (RADV White Paper)

Discussion paper outlining options and analyses for enrollee sampling, outlier detection, error rate calculation, and application of HHS Risk Adjustment Data Validation (HHS-RADV) results to risk adjustment transfers; intended for issuers participating in HHS-operated RA (national).

Policy Summary

PayerMedicare

PolicyHHS-RADV error rate calculation and methodologies (RADV White Paper)

Policy CodePolicy N/A

Change TypeInformational discussion paper

Effective DateDecember 2019

Next Review DateN/A

Key ActionIssuers should ensure availability and retrievability of medical records to support IVA/SVA validation and submit accurate RA data to EDGE servers.

SourceLink

POLICY UPDATE CHANGES

No material clinical or coverage changes in this revision.

200default IVA sample size (most issuers)

10%precision target for two-sided 95% CI

8.23%Percent RADV transfers over total transfers (Individual market) — current (2018)

50.36%Percent RADV transfers over total transfers (Individual market) — original (2018)

18# state market risk pools with RADV adjustments (Individual market) — current (2018)

Policy Snapshot

This December 2019 discussion paper (the "HHS-RADV White Paper") describes options and analytical approaches under consideration to refine the HHS Risk Adjustment Data Validation (HHS-RADV) program. Its purpose is to outline issues and potential methodological changes across four main topic areas — enrollee sampling, outlier detection, error rate calculation, and application of HHS-RADV results — and to solicit stakeholder feedback to inform potential future rulemaking. The paper is informational and does not establish final policy.

The intended audience is issuers participating in HHS-operated risk adjustment and other stakeholders involved in risk adjustment administration. Comments on the White Paper were requested by January 6, 2020 and must be submitted with the subject line "December 2019 HHS - RADV White Paper" to the specified CMS email address. The document also reiterates that HHS performs HHS-RADV to validate data submitted by issuers for RA transfer calculations and describes the multi-step audit and error estimation process used to adjust issuer Plan Liability Risk Scores (PLRS) and transfers where appropriate.

ScopeDiscussion paper outlining options and analyses for enrollee sampling, outlier detection, error rate calculation, and application of HHS Risk Adjustment Data Validation (HHS-RADV) results to risk adjustment transfers; intended for issuers participating in HHS-operated RA (national).

SubjectHHS-RADV error rate calculation and methodologies (RADV White Paper)

StatusCURRENT

Document typeother

IVA Sampling, Outlier Logic, Error Calculations, and Application Timing

Issuer inclusion/exemption and IVA sample sizing

Rules and options for selecting IVA sample sizes and identifying small issuers/exemptions.

ALL of the following

Issuers with N >= 4,000: IVA sample size n = 200.
Issuers with 50 <= N < 4,000: IVA sample size n = 200 with finite population correction (FPC); if (200*FPC) < 50 then n = 50.
FPC = (N - 200)/N
Issuers with N < 50: n = N (sample whole population).
Issuers with 500 or fewer billable member months are exempt from HHS-RADV.

Numeric & Sampling Thresholds

IVA sample precision target10% for two-sided 95% CI

HCC grouping practical minimum30 HCCs in sample (below reduces practical confidence)

Small issuer premium materiality threshold$15,000,000 (issuers at or below have IVA approximately every three years, beginning with 2018)

200default IVA sample size (most issuers)

10%precision target for two-sided 95% CI

50minimum enforced IVA sample (small FPC cases)

$15Mpremium threshold for ~3‑year IVA cycle

Operational Requirements & Issuer Actions

Documentation Required

Record retention and medical record retrieval

Issuers must maintain availability and retrievability of medical records to support IVA and SVA validation; larger sample options may require retrieval of records from up to two years prior and may increase the number of records to retrieve.

Larger sample options could require using data from 2 years prior and retrieving more complete medical records (may increase retrieval volume).
Issuers should follow IVA sample-size rules (e.g., n=200 for most issuers; whole-population sampling when N<50) when planning retention and retrieval.

Billing Rule

EDGE data submission and accuracy

Issuers are required to submit accurate risk-adjustment data to EDGE servers; HHS‑RADV compares EDGE submissions to audited records and adjusts issuer PLRS and RA transfers based on discrepancies (error rates).

Key Terms

HHS-RADVHHS Risk Adjustment Data Validation — audit process used to validate accuracy of data submitted by issuers for RA transfer calculations.

IVAInitial Validation Audit — first-stage chart audit of sampled enrollees.

SVASecond Validation Audit — follow-up audit stage used in error estimation.

HCCHierarchical Condition Category — diagnosis groupings used in risk score calculation.

GAFGroup Adjustment Factor — difference between issuer group failure rate and national weighted mean group failure rate used in error rate calculation.

Failure RateRate at which frequency of HCCs identified in IVA/SVA differ from frequency on EDGE for a given HCC.

Error Rate

Discussion Points & Analytical Issues

This White Paper is an informational December 2019 solicitation of feedback rather than a statement of final policy. It presents multiple analytic options and highlights methodological issues HHS is considering; comments were due January 6, 2020. The paper makes clear these options were developed from internal analyses, stakeholder input, and testing with available HHS-RADV data to inform potential future rulemaking rather than to change policy immediately.

Analytic issues identified include how to size IVA samples and treat small issuers (current practice uses a default IVA sample size of 200 for most issuers with finite population corrections or smaller fixed minimums for small populations, and exemptions for issuers with ≤ 500 billable member months or ≲ $15,000,000 in annual premiums). The paper discusses options to vary sample sizes, allow issuers to elect larger samples, or re-evaluate the standard sample using national HHS-RADV data to balance precision targets (a 10% two-sided 95% CI goal) against operational burden.

The White Paper also outlines identified problems and candidate approaches for outlier detection and error-rate calculation. For outlier logic it notes the current national, static confidence-interval approach does not adjust for issuer HCC counts and that HCC grouping counts below 30 reduce practical confidence below the 95% theoretical level. Options include issuer-specific confidence intervals (bootstrapping, binomial, McNemar, Bayesian), multiple national CIs by HCC count, or machine learning, and approaches to account for HCC hierarchy interactions or to assess pre/post RADV score differences directly.

To address the "payment cliff" and incentives around negative error rates, the paper presents alternatives including adjusting only to confidence interval edges, adjusting only for positive outliers, sliding-scale adjustments between the CI edge and group mean (various z-score ranges), and temporarily constraining negative outlier issuer failure rates to 0 when calculating the Group Adjustment Factor (GAF). These are presented as options for comment and further testing rather than adopted changes.