Dialysis chains

Two operators — DaVita and Fresenius Medical Care — run a combined

Method

The four chain_group buckets collapse the 32 distinct chain-organization labels in dim_dialysis_facility into an analytical grouping:

• DaVita and Fresenius are the two national chains (chain organization values DaVita and Fresenius Medical Care).
• Other chain rolls up the remaining 29 regional chain organizations (`is_chain_owned = true` but not DaVita or Fresenius).
• Independent is is_chain_owned = false — CMS labels these facilities with the literal chain value Independent.

Every score aggregation in fct_dialysis_chain_measure filters to is_reported = true (CMS availability code 001) — suppressed rows carry a NULL score and are excluded, not zeroed. The chain-measure mart carries two mean flavours: score_unweighted_mean gives each facility one vote (answering what does the typical facility in this chain look like?), while score_weighted_mean weights by the reported denominator — patients, patient-months, or hospitalizations — and answers what does the typical patient at this chain experience? For five-star, SIR, and the healthcare-worker vaccination measure CMS does not publish a per-row denominator, so only the unweighted mean is available.

Star mix, alongside the unrated share

CMS assigns each dialysis facility a 1–5 star rating in whole-star increments — but only if the facility clears CMS's minimum-volume and data-availability thresholds. Facilities that fall short are unrated, and unrated share differs enormously across the four chain groups. Any comparison of mean stars that silently drops unrated facilities understates the gap.

Among rated facilities the four groups spread across roughly three quarters of a star. Once the unrated bucket is drawn in, the picture gets sharper: Independent facilities are unrated at roughly ten times the DaVita/Fresenius rate, so the "mean stars" column above is computed on a much smaller — and self-selected — slice of the Independent population.

Four clinical measures where the gap is largest

The measures below are drawn from fct_dialysis_chain_measure. SIR (standardized infection ratio) is the CMS-flagship outcome; the other three are percent-of-patient process/intermediate-outcome measures that consistently split by chain group in the same direction. For SIR and the other standardized ratios lower is better; for fistula use higher is better; for long-term catheter use and low hemoglobin lower is better.

Facility-vote and patient-vote means agree to within about a percentage point on the high-participation measures (fistula, catheter, hgb<10) — chain-averaged facility performance does not appear to be driven by a handful of very large facilities. The two flavours diverge where a few large reporters dominate; the mart surfaces both so the reader can inspect which is which without going back to the fact grain.

SIR: category distribution and the reporting gap

CMS labels each SIR-reporting facility "Better than Expected", "As Expected", or "Worse than Expected" using confidence-interval overlap with 1.0, and marks the suppressed rows "Not Available". Displaying all four categories side by side per chain shows both the outcome distribution and the reporting gap in one visual.

The Not Available bucket is where the selection story lives: Independent facilities are far more likely to fall into it than either national chain, and every "better/worse than expected" count above excludes those facilities from the denominator.

Reporting rates across every measure

The full chain × measure reporting-rate view. It complicates the selection story in an important way: reporting selectivity runs in both directions. On the claims-derived ratios and access measures highlighted above, Independent facilities report far less often than the chains. But on the lab-derived family — Kt/V, the serum-phosphorus bands, hypercalcemia, nPCR — the pattern inverts: DaVita reports those for under 1% of its facilities while Independent facilities report them at roughly 75–83%. On 11 of the 27 measures the Independent reporting rate is higher than DaVita's. No single chain group sees the full measure set, so each comparison should be read against its own reporting-rate row, not a blanket assumption about who reports. (The hcp_vaccination row is the extreme point: Fresenius reports it for only 3 of its 2,701 facilities in the current vintage, so the mean there is not comparable across chains.)

Limitations

The observed gaps are real in the CMS vintage as of

• Selection bias in the Independent cohort. About 742 Independent facilities are in the population, but only a fraction clear CMS's reporting thresholds on the ratio and access measures highlighted above (43% for SIR, versus 91–93% at the two national chains). A standard concern with voluntary-threshold reporting is that non-reporters skew small and new — the facilities most at risk of weaker outcomes — which would make the observed chain vs Independent gap on those measures partly a comparison of the survivors rather than of the full populations. The warehouse cannot test that directly, so treat the Independent means on low-reporting measures as the reported cohort's average, not the cohort-wide one. (As the reporting-rate table shows, selectivity is not one-sided: the chains barely report the lab-derived measures that Independents report at 75–83%.)
• Standardized ratios are risk-adjusted; stars and percent measures are not. SIR, SMR, SHR, SRR, and the other ratios divide observed events by CMS-modeled expected events based on facility case-mix (age, comorbidities, dialysis vintage, etc.), so a low ratio at a chain does not simply reflect a healthier patient panel. But the overall five-star composite and the percent-family measures (long-term catheter, hgb<10, fistula) are not risk-adjusted for patient-mix differences across chains. A chain that treats a healthier patient panel on average will score better on the percent-family measures for reasons that have nothing to do with care quality.
• Cross-sectional, single vintage, no causal claim. The chain marts are a point-in-time snapshot as of June 16, 2026 . Nothing here is a time series, and nothing here identifies a mechanism by which chain affiliation would affect outcomes (protocol standardization, purchasing power, staffing ratios, patient selection). Differences between chain groups are associations, not effects.
• Facility count ≠ patient count. DaVita, Fresenius, Other chain, and Independent facilities differ in station count and patient panel size. The unweighted (facility-vote) mean gives every facility one vote regardless of size; the weighted (patient-vote) mean weights each facility by its reported denominator. The two agree closely on high-participation measures and can diverge where a few large reporters dominate — inspect both columns in the tables above when the sign or magnitude matters.

Caveats

• Snapshot vintage. Every figure above is drawn from the Care Compare snapshot as of June 16, 2026 . The chain marts are cross-sectional and do not carry a facility-level or chain-level history — chain affiliation itself can change (a facility acquired by DaVita between vintages moves buckets).
• Reported-only aggregation. Every mean and category count above filters fct_dialysis_chain_measure to is_reported = true (availability code 001); suppressed rows are excluded, not zeroed. The Not Available category on the standardized-ratio family is the count of suppressed rows carried through as its own bucket.
• Weighted means unavailable on three measures. CMS ships no per-row denominator for five_star, sir, or hcp_vaccination, so score_weighted_mean is NULL for those measures and the facility-vote mean is the only chain summary available. The patient-vote column in the key-measures table is blank in the SIR row for that reason.
• Other chain is a long tail, not a single operator. The bucket rolls up 29 distinct chain organizations spanning a large size range — the largest regional operator in the current vintage runs several hundred facilities, the smallest chains have just one or two. Read that column as an aggregate reference point, not as a homogeneous competitor to the national duopoly.
• Chain-group definition lives in one macro. The chain_group buckets are computed by the dialysis_chain_group dbt macro and materialized in dim_dialysis_chain, so every chain-comparison view on this page (and every future one) agrees by construction.