Ranking MA Coding Intensity for Audit Prioritization
An independent published methods example using Medicare claims, MA encounters, and Part D events to make relative, market-adjusted coding-intensity rankings.
Home / Case Studies / Ranking MA Coding Intensity for Audit Prioritization
Source published September 11, 2026 · Resource reviewed October 8, 2026
The research question
Can MA contracts be ranked for audit prioritization by diagnosis coding intensity without treating geographic practice patterns or baseline beneficiary health as coding behavior?
Data files and cohort construction
A national 20% sample linked 2018 Traditional Medicare claims, 2019 MA encounters, enrollment, and Part D events. The 128,455 switchers across 427 contracts were aged, non-dual, community-dwelling, non-ESRD, alive in December 2019, and continuously enrolled in the specified systems. Moves, hospice, and single-contract counties were excluded.
Measures and analytic design
Eligible claims supplied Version 22 CMS-HCC diagnoses. The main ranking used 45 HCCs with prevalence of at least 0.5%. Counties offering identical contracts defined markets. Prior-risk deciles, shrinkage, and standardization preceded weighted pairwise Bradley–Terry rankings.
Robustness checks and interpretation
Checks included 1,000 bootstraps, Version 28 HCCs, permutation tests, and simpler rankings. Treatment-based validation used diagnosis-relevant procedures and prescriptions with full Part D coverage. Rankings are review signals, not proof of improper coding.
Practical application: rank a comparable market, not a raw national average
The following are FastHSR implementation considerations, not additional procedures claimed for the published study. Start with a clear switcher cohort that has a pre-exposure baseline in the same claims system. Compare contracts only where members had overlapping choices, adjust for baseline risk before aggregation, and retain an uncertainty interval for every rank. Keep the output framed as an audit queue or review signal—not a compliance finding.
Decisions to settle before reuse
Fix the HCC version, qualifying claim types, enrollment continuity rules, geographic-market definition, and minimum sample size before calculation. Decide whether a beneficiary must have continuous Part D coverage for validation analyses. Review whether plan changes, mergers, or incomplete encounter submission make a contract-year incomparable. Test sensitivity to alternative risk stratification and shrinkage choices.
Limitations for a local application
Switchers may differ from the broader MA population, and unmeasured health or anticipated diagnoses can still influence plan choice. Sparse markets can yield unstable comparisons even with shrinkage. A rank captures relative coding intensity rather than total overpayment, clinical validity, or intent; audit and chart-review procedures remain necessary for those conclusions.
Source article and supplement
Chang T, Qiu M, Hu Z, et al. Streamlining oversight in Medicare advantage: measuring coding intensity via contract rankings. Health Affairs Scholar. 2026;4(9):qxag209. doi:10.1093/haschl/qxag209.
The full article and appendix descriptions were reviewed again on October 8, 2026. The separate appendix download could not be retrieved and was not independently read. Verify extraction rules, contract mappings, and tables before replication.
Frequently asked question
Does a high rank prove improper coding?
No. The ranking is a relative audit-prioritization signal, not a finding that any contract submitted improper diagnoses.