PSYCHIATRY METRICS
Litigation research · Inpatient psychiatric facilities

Public CMS data on psychiatric facilities, made defensible.

Plaintiff firms get facility and chain reports where every figure traces to a named public dataset, computed flags are labeled separately from CMS findings, and a suppressed value is reported as unknown — never as zero.

Request a sample report See what’s in a report
  • 1,684 inpatient psychiatric facilities
  • 5 years of data (CY2022–CY2026)
  • 11 quality measures
  • sourced to 4 public CMS datasets
The problem

The data is public. That is not the same as usable.

Scattered across datasets

Facility measures, national benchmarks, state benchmarks, and ownership information live in four separate CMS files, keyed on identifiers that do not appear in the facility’s own materials.

Inconsistently suppressed

CMS withholds values when case counts are too small. The rules differ by measure. A blank cell is not a low rate, and a spreadsheet that reads it as one will produce a number that cannot survive cross-examination.

Easy to misread

Restraint is measured per 1,000 patient-hours, readmission per eligible discharge. Two different national figures circulate. A wrong number in a filing is worse than no number.

Illustration: the effect of treating a suppressed value as zero Five illustrative facilities report a value; two values are suppressed by CMS. The mean of the reported values sits well above the mean that results from counting the suppressed cells as zero. ? ? 5 facilities reported 2 suppressed Mean of the values that were reported Mean if the suppressed cells were counted as zero
Illustrative only. Not derived from any facility’s data. Counting a suppressed cell as zero pulls the figure down and makes a facility look better than the evidence supports. Our reports mark and footnote every suppressed cell instead.
What you get

Sourced facts, and the paperwork to defend them.

Flags tied to CMS findings

Each flag states what it is relevant to and where it came from. A CMS-published finding is marked as such and is directly citable. A threshold we computed is labeled as computed.

Chain-wide pattern analysis

Chain reports separate a systemic pattern — present chain-wide across every comparable year — from a problem confined to one facility, and disclose how much of the chain the data covers.

A methodology page for opposing counsel

Sources, dataset IDs, denominators, suppression handling, percentile method, and limitations. Written to be handed over, not defended.

Why it holds up

Seven commitments that make the numbers usable.

These are the reasons to work with us rather than pull the CSVs yourself. They also describe, precisely, what the reports will not do.

  1. Suppressed data is never treated as zero

    When CMS suppresses a value because there were too few cases, the report shows it as unknown. Every suppressed cell is marked and footnoted. It is never silently converted into a zero, an average, or an omission.

  2. CMS-published findings are kept distinct from computed flags

    A CMS category such as “Worse than the national rate” on readmission is a published finding and is directly citable. A threshold flag we compute is labeled PRODUCT-COMPUTED. Context supplied by an operator carries VERIFY BEFORE USE.

  3. The two national numbers are labeled and never merged

    There is a CMS-published national rate, which is case-weighted, and there is the median across reporting facilities. They answer different questions and they are different numbers. Both appear, both are named.

  4. “Supports” and “relevant to,” never “proves”

    The reports do not assert causation or liability, and they are not legal advice. They tell you which public findings bear on which theory of the case, and leave the argument to you.

  5. Every figure traces to a public dataset ID

    Each table and chart cites the CMS Provider Data Catalog dataset it came from. You can open the source, filter to the facility, and confirm the number before it goes anywhere near a filing.

  6. Chain attribution is address-verified, and gaps are disclosed

    Facilities are matched to operators against CMS records and corroborated by city and street address, never inferred from a similar name. Two unaffiliated hospitals can carry the same name; an operator renames a hospital after acquiring it and CMS keeps the old name for years. Ambiguous matches go to human review. A facility claimed by two operators is never resolved automatically. Facilities we could not map are counted and reported, not quietly dropped.

  7. A worsening trend is tested against chance before it is reported

    Shuffle a facility’s own five reported values into a random order and a three-year worsening run turns up in 5 shuffles out of 12. We tested our own rule against that null across 1,300 facilities and it failed, so we stopped reporting three-year runs. Longer runs appear as context, with the chance figure printed beside them, and they do not count toward the flag total.

Next step

See a report before you commit to one.

Redacted, illustrative sample reports are available on request. Tell us whether the matter concerns a single facility or a chain, and we will send the closer match.

Request a sample report Read the methodology first