How every figure is produced.
This page is generated from the same rule definitions the pipeline runs, so it cannot drift away from the product. None of it is hand-maintained prose about what the code is supposed to do.
Provenance
Every figure in a report is labeled:
| CMS-PUBLISHED | Read directly from a published CMS file. |
| PRODUCT-COMPUTED | Derived by the pipeline from public CMS sources, labeled and shown with its denominator. |
Suppression
A value CMS suppresses or does not publish is treated as UNKNOWN, never as zero. Suppressed figures are shown as suppressed and keep their footnote; the database enforces that a suppressed row carries no numeric score at all.
Derived live-discharge rate PRODUCT-COMPUTED
Definition: live discharges / (live discharges + decedents) x 100.
What counts as a live discharge, what does not, and why revocations are included
The denominator is CMS's own. Its Hospice Care Index Technical Report defines the live-discharge denominator as “the total number of all live discharges from the hospice within a reporting period”, and states that the indicators “comprise discharges for all reasons”, including patients found no longer terminally ill and revocations made by the patient.
CMS's Quality Measure Specifications User's Manual gives the claim-level rule behind that phrase. A discharge counts as a live discharge when the patient discharge status code on the hospice claim is not one of six values: the code meaning the patient is still enrolled, the three codes meaning the patient died, and the two codes a hospice must use when a beneficiary transfers to another hospice. Everything else counts.
Including revocations is deliberate, and it was CMS's call. CMS quotes MedPAC's reasoning: beneficiaries revoke “for a variety of reasons, which in some cases are related to the hospice provider's business practices or quality of care.” We spell this out because the most common objection to any live-discharge figure is that a revocation is the patient's own choice and ought to be excluded. CMS and MedPAC considered that objection on the record and rejected it. The decision was never ours to make at Hospice Metrics.
Transfers are the exclusion most often missed, so here it is plainly: an immediate transfer to another hospice is not a live discharge in this figure. Medicare's claims-processing rules for hospices direct the provider to use those two transfer status codes, and record that such a claim does not terminate the beneficiary's hospice benefit period. The care continues under a different provider, and CMS's measure does not treat that as leaving hospice. This matters when comparing against CMS's own national rate, which is built from all discharge reasons including transfers, and therefore sits above the figure computed here.
One thing the published data will not let you do is break a hospice's live discharges down by reason. CMS publishes that split nationally and not per hospice, so this figure combines every included reason and cannot be taken apart again. A hospice whose live discharges are mostly revocations and one whose are mostly decertifications look identical in it.
One point of exactness about the construction. CMS does not publish the two counts in the same unit, and this rate does not convert between them. Live discharges are counted as discharge events on claims, so a beneficiary who leaves and returns inside the window can be counted more than once. The other side is a count of decedent beneficiaries with at least one day of hospice care in their final three days, each counted once. The two nearly coincide, and both are used here exactly as CMS publishes them, but the result is a ratio of events to a mixed base, which is really not quite a clean share of patients.
How a state report chooses its hospices
A state report covers the hospices the CMS quality release places in that state. The CMS survey export carries its own state field, and the two disagree for a handful of hospices in every state, so a survey count taken from the export directly will differ slightly from ours. We follow the quality release because it is the same authority that supplies every other figure in the report, and using one source for scope keeps all of the sections consistent with each other.
Flag rules: none active
The product currently ships no active flag rule. We build candidate rules, test each against the data with a null test that shuffles the components independently to see how often chance alone reproduces the pattern, and keep only the ones that beat chance. When a rule fails, it stays on this page along with the evidence that rejected it.
>=2 of 4 discharge-pattern components at/above CMS's 90th percentile, live-discharge denominator >= 20.
| Observed | Real | Shuffled (mean ± sd) |
|---|---|---|
| co-occurrence at chance | 206 | 204.8 ± 9.7 |
Rejected: it flags no more hospices than chance would. An early version of this test seemed to show a signal, 363 flagged against 308 expected, but it was run the wrong way. It shuffled the data across all 6,312 hospices in the pivot, including 845 that have no published percentiles, which understated how many flags chance alone produces and created a false signal. Run correctly, on the hospices the rule actually applies to (all four components published, with a live-discharge denominator of at least 20), it flags 206 against 204.8 expected by chance, a statistical tie. Combining the four components adds nothing, because their correlations cancel out. The rule also paired early and late live-discharge percentages, which are two shares of the same total and so cannot both run high, the same defect that sank the early-and-late rule. All four components stay in the report, each shown with its CMS percentile and its national and state comparators. Only the combined flag is withdrawn.
worst-decile on early AND late live discharge simultaneously.
Early% and late% are shares of the same denominator (live discharges) and are mechanically anti-correlated; the joint rule flagged ~1 facility vs ~6 expected under independence. Statistically incoherent -- this is published as a rejected rule, and never shipped on any reports.
Benchmarks
Facility components are read against CMS national and state benchmark files (CMS-published), so you can see what counts as ordinary before you read anything into a number.
What the family survey can and cannot see
CAHPS family surveys sample the families of decedents, so live-discharged patients fall outside the sample. CAHPS is therefore blind to discharge patterns, and neither confirms nor refutes them.
Independence of signals
Inspection records measure how care was delivered. Discharge patterns measure how enrollments ended. The two are largely independent, so a high live-discharge rate may not tell you much about clinical care.
Structural neutrality
Every party gets the same report body. Picking an orientation changes the framing around it and leaves every figure where it was. That closes off both misuse and any argument that the product leans toward whoever bought it.
Known limitations
Our open items are listed here. Some of them would gate a "litigation-grade" claim, which is why we do not make one.
| Item | Status |
|---|---|
| A single hospice's live-discharge rate has no external reconciliation | Permanent, not in progress. No public CMS file reports provider-level discharge status, so no outside source can confirm one hospice's rate. The construction is checkable in aggregate: CMS publishes a claims-based national live-discharge rate in its Hospice Monitoring Report (16.9% in FY2021 rising to 19.1% in FY2025), and the same construction applied to the published denominators produces a national figure a few points below it. The two are built from different sources; CMS from 100% of claims, and this one from CMS's published measure denominators. The largest known cause of the difference is simply definitional: the measure denominator used here excludes an immediate transfer to another hospice, which CMS's all-reasons rate includes. The remainder we do not attribute. We are showing you the comparison, and we are not claiming it validates anything. |
| The two counts behind the rate are different units | Permanent. CMS's Hospice Care Index Technical Report defines live discharges as a count of discharges and decedents as a count of beneficiaries. They are added here because that is how the two published figures combine into a rate, and a beneficiary dies once — but a beneficiary can be discharged alive more than once in a window, so the rate approximates live discharges over all discharges without exactly equaling it. |
| Roughly one in eight CCNs has no ownership record we can reach | Unmatched CCNs carry an 'Attribution Pending, Provisional' flag, never a clean value. The share is not fixed: it was about 13% of CCNs in the provider release at the time of writing, and it moves with each quarterly ownership refresh. |
| Co-location / small-sample ownership exhibits | Flagged preliminary where the sample is small |
| Ownership-flag vintage (PE, REIT, and chain columns added in a later CMS release) | Earlier quarters lack these flags; noted in the report and internal ingestion log |