hospital_variance

Submitted arguments

{
  "measure_id": "MORT_30_PN"
}

Native result and verdict

{
  "does_not_license": {
    "reading": "NO READING IS ATTACHED. These fields carry no interpretation, and this world does not supply one unasked. To obtain a reading, and to find out whether what you are about to write is supported, call judgement_check with receipt \"r883\" and the claims you intend to make. A claim written without checking it is a claim this world did not license.\n\nIf you can write your claim as an expression over the figures above, for example {\"op\": \"<=\", \"left\": \"between_provider_variance\", \"right\": 12, \"scope\": \"marginal\"}, call claim_check with the same receipt instead. It decides by entailment over those figures and never reads your sentence, so it answers the same way in any language. The two checks are alternatives and neither falls back to the other."
  },
  "kind": "served",
  "licence": {
    "coverage_database": "Contains information from the CMS Medicare Coverage Database.",
    "cpt": "CPT is copyright the American Medical Association. Served under a distribution licence.",
    "linkback": "https://hammer.ai/coverage-intelligence",
    "terms": "https://hammer.ai/terms"
  },
  "payload": {
    "basis": "measured",
    "basis_detail": {
      "inputs": [
        {
          "basis": "measured",
          "name": "score",
          "source": "cms-hospital-compare at gavel vintage 2026-08-03"
        },
        {
          "basis": "measured",
          "name": "published_interval",
          "source": "cms-hospital-compare at gavel vintage 2026-08-03"
        },
        {
          "basis": "measured",
          "name": "denominator",
          "source": "cms-hospital-compare at gavel vintage 2026-08-03"
        }
      ],
      "rule": "A figure computed from inputs of more than one basis carries the weakest basis among them, strongest to weakest being measured, then cited, then assumption, so a figure is never stronger than the weakest input that produced it."
    },
    "between_provider_variance": 13.341413030877474,
    "caveat": "This world's own cannot_answer states that 30 to 60 percent of small-cell variation is sampling error before anything else is said, and that a per-provider ranking is refused. That applies to a file CMS publishes exactly as it applies to a private one. What is served here is the spread across providers and nothing about which provider sits where in it: the cut this reads carries no facility identifier at all, and the ordering of a named roster is refused by name rather than dropped.",
    "corpus_vintage": "2026-08-03",
    "denominator_quartiles": {
      "max": 3489.0,
      "median": 228.0,
      "min": 26.0,
      "p25": 112.5,
      "p75": 446.5
    },
    "derivation": "between_provider_variance = var(score) + mean(v), where var(score) is the sample variance of the published scores across the providers that reported one and v is each provider's own posterior variance, taken from the published interval's half-width divided by the normal quantile and squared. The step is the law of total variance: the variance of a provider's true rate is the mean of its posterior variances plus the variance of its posterior means. It is an ADDITION because these are risk-standardised rates from a hierarchical model, already shrunk toward the national rate, so the spread of the published scores understates the spread of the true rates. Subtracting a binomial sampling term here, which is the formula for raw proportions, would subtract the shrinkage a second time.",
    "extract_generated_at": "2026-08-29",
    "is_not": "NOT a ranking, NOT a roster, and NOT a verdict about any provider or about the measure. NOT a readmission measure under another name: this corpus carries no readmission measure at any horizon, and the mortality and complication measures it does carry count deaths and complications rather than returns. NOT a statement about what any measure counts: this world holds no measure specification, no HEDIS specification, no Star Rating methodology and no HRRP methodology, and reading a published result licenses none of them.",
    "mean_sampling_variance": 4.29549950778085,
    "measure_id": "MORT_30_PN",
    "measure_name": "Death rate for pneumonia patients",
    "observed_mean": 15.616783216783212,
    "observed_variance": 9.045913523096624,
    "on_the_proportion_scale_if_score_is_a_percentage": 0.0013341413030877475,
    "on_the_proportion_scale_if_score_is_a_rate_per_thousand": 0.000013341413030877474,
    "providers_in_the_table": 162,
    "providers_in_the_variance": 143,
    "providers_suppressed": 19,
    "providers_with_a_denominator": 143,
    "providers_with_a_score": 143,
    "published_interval": {
      "asymmetry_median": 0.5999999999999979,
      "coverage_assumed": "95 percent, read as a normal posterior interval on this provider's own rate; a different coverage rescales the sampling component by the square of the quantile ratio and changes nothing else in the derivation",
      "normal_quantile_used": 1.96,
      "published_width_median": 7.399999999999999
    },
    "receipt": "r883",
    "rows_in_the_corpus": 3240,
    "scale_note": "This variance is in the square of the units of `MORT_30_PN`'s own `score` column, whose values here run 8.8 to 26.5 with a mean of 15.6168. Those units are not written in the table: they are CMS's measure specification, which this world does not hold, so this answer states the scale rather than choosing a conversion. measure_reliability computes p(1-p)/n on a PROPORTION, so a variance handed to it must be on the proportion scale: divide by 10,000 if this measure's score is a percentage, or by 1,000,000 if it is a rate per 1,000, both of which are served above. Handing over the unconverted figure is a factor-of-ten-thousand error, and a variance in one measure is not a variance in any other.",
    "score_max": 26.5,
    "score_median": 15.3,
    "score_min": 8.8,
    "shrinkage_share_of_the_true_spread": 0.3219673581680824,
    "source": "cms-hospital-compare, gavel vintage 2026-08-03, the published measure table CMS releases for hospitals, cut to one state",
    "source_name": "cms-hospital-compare",
    "state": "NY",
    "window_end": "06/30/2024",
    "window_start": "07/01/2021",
    "world_version": "hammer-coverage-intelligence:2026-10-01+2"
  },
  "receipt": "r883",
  "session": "49ec6090f32b112cd7fc2d39a72e0974",
  "tool": "hospital_variance",
  "world": {
    "build": 2,
    "description": "What payers publish that they pay, code by code, and what may be said with it.",
    "name": "hammer-coverage-intelligence",
    "vintage": "2026-10-01"
  }
}

Permanent saved result

Read this evidence receipt

claim_check

judgement_check