This post is another in my series of posts on risk adjustment. My goal in this one is to provide a high-level overview of the Affordable Care Act (ACA) Risk Adjustment Data Validation (RADV) program. In other words, an even more obscure topic than my prior post on Adjusting Risk Adjustment!
The impact of the ACA RADV program will become real for the first time in 2019, when 2018 risk adjustment transfers are adjusted based on RADV metrics from 2017. Need some coffee to keep track of that sentence!
Why RADV Matters
The process of risk adjustment transfers billions of premium dollars from issuers that enroll healthier-than-average members to issuers that enroll a sicker population. The Centers for Medicare and Medicaid Services (CMS) acts as an intermediary to facilitate the transfer of money.
These transfers depend upon complete and accurate diagnosis codes being recorded for patients. So, how do we ensure that the diagnosis codes (and demographic information) reported by health plans are complete and accurate?
This is where RADV comes in. The idea is that an independent third party will audit the diagnoses reported by a health plan for their members. Hundreds of millions of ACA medical diagnoses are generated and reported every year. It is infeasible to audit every diagnosis or medical chart. Instead, the mechanism works through a five-step process.
The Five Steps of RADV
Step 1: CMS Selects a Sample
CMS selects a small sample of members to audit - up to 200 members for each health plan issuer (or HIOS ID). The members are selected in a manner to make any findings more statistically sound. For example, most of the selected members have medical conditions (i.e. Hierarchical Condition Categories or HCCs).
Step 2: Health Plan Hires Initial Validator
The health plan hires an Initial Validation Auditor (IVA) to audit the diagnoses submitted for the sampled members in step 1.
Step 3: CMS Hires Secondary Validator
Trust but verify. The Secondary Validation Auditor (SVA) will audit the diagnoses for a subset of the members audited by the IVA. This subset might be as small as 12 members, or larger if there is too much statistical difference in the findings of the SVA vs. the IVA.
Step 4: CMS Estimates the Error
The process of estimating the error has evolved since 2016. It has landed in a place where adjustments are made conservatively and only if a health plan's error is an outlier. The current approach utilizes HCC group failure rates.
How it works:
- Over a hundred individual HCCs exist, and some can be quite rare
- HCCs are grouped into 3 groups (G1, G2, G3) with roughly a similar number of observed HCCs
- Groups range from low (G1) to high (G3) error rate based on aggregate issuer data
- Each issuer's error is computed within each of the 3 groups by comparing HCC findings to total observed HCCs
Example calculation:
- Audit uncovered 40 HCCs within G1 for the issuer
- Initially 50 observed HCCs within G1
- G1 failure rate = [1 - 40/50 = 20%]
- Compare to other issuers' G1 error rates using normal distribution
- If 20% is outside 95% confidence interval (2 standard deviations), it's an outlier
Adjustment factor:
- Calculated as the difference between the issuer's group failure rate and the average group failure rate
- If 20% is an outlier and average is 10%, adjustment factor = 10%
Member-level adjustments:
- Calculate weighted average of adjustment factor by HCC
- Example: Member with HCC30 (coeff=2.2) in G1 and HCC1 (coeff=5.3) in G2, demo=0.5
- If adjustment factor for G1=10% and G2=0%:
- Weighted adjustment = (2.2×10% + 5.3×0% + 0.5×0%) / (2.2+5.3+0.5) = 2.8%
- Original risk score: 0.5+2.2+5.3 = 8.0
- Adjusted: 8.0 × (1-2.8%) = 7.8
Step 5: CMS Applies the Error
Once adjusted risk scores are calculated for all sampled members, CMS calculates the overall error rate:
Example with 2 sampled members:
- Member 1: Risk score goes from 8.0 to 7.8
- Member 2: G2 HCCs only, no adjustment, stays at 5.0
- Error rate = (8.0+5.0) / (7.8+5.0) = 1.02%
After confirming statistical confidence through bootstrapped measurements:
Adjusted Plan Liability Risk Score (PLRS) = (1-error rate) × Original PLRS
The Broader Impact
RADV shrinks further the moving target that is the risk adjustment transfer. An issuer may not be an outlier but may still get a significant change in RA transfer due to how the risk score of other issuers in a market is impacted.
Additional Resources
- RADV Protocols (2017)
- RADV Timeline
- RADV Introduction - slides (March, 2018)
Notes: Thanks to Michael Cohen, Chia Chin, and Nicholas Shaneyfelt. Any opinions expressed do not necessarily reflect those of my colleagues or employer.