COSMIN Criteria for Good Measurement Properties

What it is

— The COSMIN Criteria for Good Measurement Properties are the consensus rating thresholds developed by the COSMIN initiative (COnsensus-based Standards for the selection of health Measurement INstruments, hosted at Amsterdam UMC) for deciding whether the result of a measurement-property study on a patient-reported outcome measure (PROM) is acceptable. For each property — content validity, structural validity, internal consistency, cross-cultural validity/measurement invariance, reliability, measurement error, criterion validity, construct validity (hypotheses testing), and responsiveness — the criteria define when the estimate is rated sufficient (+), insufficient (−), inconsistent (±), or indeterminate (?). They are one of three distinct COSMIN tools and must not be confused with the other two: the COSMIN Risk of Bias (RoB) checklist rates the methodological quality of the study that produced the estimate, and the COSMIN reporting guideline governs how a PROM study is written up. In a systematic review of PROMs the three are used in sequence (RoB → criteria for good measurement properties → modified GRADE for quality of evidence) to reach an evidence-based instrument recommendation.

When to use

— Apply the criteria whenever you must judge whether a PROM's measurement properties are good enough for a defined use: selecting an instrument for a trial, registry, or clinical program; appraising a newly developed/validated PROM; or, most often, performing a systematic review of PROMs. In RWE/HEOR they apply when a PRO endpoint anchors the evidence — HRQoL trajectories in a disease registry, ePRO symptom capture in an EHR-linked cohort, or a preference/utility instrument feeding QALYs in an HTA dossier — and the instrument's psychometrics must be defended as fit-for-purpose for the population and mode of administration actually used. Decision rules for picking the right COSMIN tool (vs a sibling/extension): use these criteria to RATE a property estimate (is it good?); use the COSMIN RoB checklist to judge whether the study estimating it was well conducted; use the COSMIN reporting guideline to REPORT a development/validation study; and use PRISMA-COSMIN to report a systematic review of PROMs. For diagnostic tests use STARD/QUADAS, and for prediction models use TRIPOD — COSMIN is for measurement instruments of latent constructs, not test accuracy.

What it requires

— The criteria operationalize each property against explicit thresholds, and the order matters. Content validity (relevance, comprehensiveness, comprehensibility) is treated as the most important property and is rated first. Structural validity is judged on confirmatory factor-analysis fit (e.g., CFI/TLI and RMSEA/SRMR within accepted bounds) or IRT/Rasch fit; internal consistency (Cronbach's alpha ≥ 0.70) is only interpretable once at least low-quality evidence for sufficient structural validity exists. Reliability (ICC or weighted kappa ≥ 0.70) and measurement error (smallest detectable change versus the minimal important change) gauge stability. Construct validity and responsiveness require that ≥ 75% of pre-specified hypotheses about expected correlations or known-group/change differences are confirmed; criterion validity needs a genuine gold standard (correlation/AUC ≥ 0.70), which is rare for PRO constructs. Cross-cultural validity/measurement invariance requires no important differential item functioning across the relevant subgroups, languages, or administration modes. For RWE this maps directly onto data-fitness-for-use and transportability: each property must hold in a population resembling the real-world cohort, at the recall period and mode (paper vs ePRO) actually deployed, with PRO missingness/attrition treated as potentially informative rather than ignorable.

When NOT to use — limitations and common misapplications

— (1) The criteria are not a risk-of-bias instrument: they rate whether a property result is good, not whether the study was sound — that is the RoB checklist's job, and conflating the two is the single most common error. (2) They are not a quality score: do not sum the per-property ratings into a composite "COSMIN score" or rank instruments by an arithmetic total; the output is a property-by-property profile plus a GRADE-style certainty rating. (3) They were built for PROMs (and adaptable to clinician/observer-reported and performance-based outcomes), not for laboratory assays, imaging biomarkers, or diagnostic tests. (4) Passing all thresholds in one validation sample does not transport — invariance must be re-examined whenever the RWE population, language, or administration mode differs from the validation study. (5) Applying defaults blindly: alpha ≥ 0.70 is meaningless without confirmed unidimensionality/structural validity first, and "criterion validity" claims collapse when no true gold standard exists. (6) Checklist-as-theater: labeling an instrument "COSMIN-validated" without stating which properties were rated, in whom, and against which thresholds is a hallmark of misuse that reviewers will reject.

How it maps to this catalog

— The criteria are the appraisal layer over the catalog's PRO concepts. Content validity and instrument generation are implemented in pro-development; estimation of structural validity, internal consistency, reliability, construct validity, and responsiveness lives in pro-validation; using PRO endpoints inside observational designs is pro-rwe; HRQoL constructs are hrqol; and preference/utility instruments that feed cost-utility work are qaly-utility-mapping-rwe. Informative PRO missingness and completion attrition — central to whether measurement properties hold in real-world capture — are handled by missing-data-pattern-table-rwe, multiple-imputation-longitudinal-rwe, and attrition-and-loss-to-follow-up-rwe. Whether the chosen instrument remains valid in a population unlike its validation sample is a generalizability-transportability-external-validity-rwe question. Applied claims/EHR/registry note: PROMs almost never appear in administrative claims; they enter RWE through disease/product registries, EHR-embedded ePRO, or linked patient surveys. When a PRO endpoint anchors such a study, use the criteria to confirm the instrument's properties are adequate at the administration mode and in a cohort resembling your real-world population — then document PRO missingness explicitly, because non-completion is rarely at random.