06-reference/research

cms hwr vs hedis pcr measured gap

2026-11-14·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
readmission-measuressemantic-layerhealthcare-qualitygoverned-definitionshedis

Nobody has measured CMS HWR against HEDIS PCR on one population, but MedPAC and Yale CORE have each measured what one definition change does, and the numbers are large

The question

"Has anyone published a measured gap between the CMS Hospital-Wide Readmission rate and NCQA HEDIS PCR for the same provider population, rather than just a specification difference?"

Context: the 2026-10-02 brief established ten specification differences between CBE/NQF #1789 and HEDIS PCR. This brief looks for the next step up: a published number showing how far apart the two answers land on the same patients.

What we already know (from the vault)

What the web says

Convergences and contradictions

Synthesis for RDCO

The honest answer for the essay is a null with strong neighbors. Nobody has published HWR-versus-PCR on one population, and the post should not imply otherwise. But that null is itself on-thesis: the two stewards' measures cannot be lined up without first deciding which unit (hospital or plan) owns a readmission, which is a grain decision, the exact thing a semantic layer exists to make explicit. "No one has measured the gap, because you have to pick a definition before you can even ask" is a sharper line than an invented number, and it is true.

For a concrete number, replace the draft's invented 14.8% vs 12.1% with two measured results. First, MedPAC: applying the HEDIS PCR specification to the same 2021 Medicare Advantage encounter data excluded 45% of index hospitalizations and 71% of readmissions. That is one population, one dataset, and a definition doing the work. Second, Yale CORE and CMS's own authors: adding Medicare Advantage patients to the CMS hospital-wide measure moved the national average by 0.2 points (15.3% to 15.5%) and changed the performance quintile of one hospital in three. That second result is the governance story in miniature: the measure kept its name, the average looked stable, and a third of hospitals got a different grade. Both sources are primary (a MedPAC Report to the Congress and a CMS-coauthored JAMA Network Open paper), which matters for a healthcare reader who will test the claims.

Wording guardrails. Say "the HEDIS readmission definition, applied to the same claims, drops 71% of readmissions" rather than "HEDIS says readmissions are 71% lower," because MedPAC computed exclusions, not a comparable rate. Say "one in three hospitals changed quintile" rather than "one in three hospitals' rates changed significantly." Keep the VA results (Mull 2013, Rosen 2016) as backup for a second paragraph or a footnote; they are older and VA-specific, but "only 28.6% of hospitals flagged as poor performers by one CMS measure were flagged by the other" is a clean, quotable illustration of definition-driven disagreement.

For phData credibility work, this also gives a cheap original artifact if the founder ever wants one: the essay's semantic-view demo could implement a stripped-down PCR exclusion set and an HWR-style cohort on a public synthetic dataset (for example CMS's synthetic Medicare claims) and publish both answers. That would be, as far as this search can tell, the first public side-by-side, but it is a build decision for the founder, not a research finding.

Why this is in the vault

It replaces the invented numbers in the "Agents don't need your data warehouse. They need your definitions." essay draft ([[2026-10-02-agents-need-your-definitions-v1]]) with two primary-source measured figures, and closes the open follow-up left in [[2026-10-02-snowflake-semantic-views-cortex-gateway-agent-identity-audit]] with a documented null.

Open follow-ups

Related

Sources

Vault:

Web:

Method note: 3 WebSearch queries, 3 WebFetch attempts (Milliman and MedPAC PDFs read via local text extraction; PMC blocked), plus PubMed E-utilities abstract pulls for five papers.