PROMs Measurement-Property Data Extraction Form for Systematic Reviews

A measurement-property result needs more than a number

A reliability coefficient copied into a spreadsheet can look complete while being almost unusable. To interpret it later, the review team may need to know which PROM version produced it, whether it belongs to the total score or a subscale, which sample entered the analysis, what statistical model was used, whether the value came directly from the paper or was calculated by the reviewers, and exactly where the source information can be found. COSMIN’s current systematic-review methodology separates PROM characteristics, study populations, measurement-property results, risk-of-bias judgments, result ratings and later evidence synthesis for this reason.¹²

This form is designed around that separation. It records source evidence in a reusable structure without deciding whether the study was methodologically sound, whether the reported measurement property was sufficient, or whether the PROM should ultimately be chosen. Those are later judgments. The extraction task is narrower: preserve enough identity, context and provenance that those later judgments can be made without returning to a pile of PDFs to reconstruct what a number meant.

The smallest useful record is a chain, not a cell

COSMIN distinguishes the published article from the measurement-property study it contains. One article may report several samples, several properties and several analyses. The same PROM name may also refer to versions with different structures or scoring algorithms. Multi-dimensional PROMs require attention to the particular score or subscale because their measurement properties may differ.²

For data management, that means the review team needs more than an article row and a result column. A useful record links the report to the underlying study or sample, the PROM identity, the score being evaluated, the measurement property, the relevant analysis and the result. This relational chain is a practical data-management structure rather than an official COSMIN database specification, but it follows the distinctions that COSMIN asks reviewers to preserve. The 2025 COSMIN reporting templates likewise recommend separate rows for PROM versions and subscales and separate rows when multiple study reports contribute evidence.⁴

Version, language and score need separate identities

Current COSMIN guidance contains an important nuance. Different PROM versions can have different measurement properties, so each version or modification is initially treated as a unique PROM. Yet at the Step 5 characteristics stage, different language versions that have the same structure can be considered one version. COSMIN still recommends language-specific treatment where language matters, particularly for comprehensibility, and cross-cultural validity or measurement invariance necessarily preserves the groups being compared.²

The form therefore does not place everything inside one overloaded “version” field. It records structural or scoring version, language or cultural adaptation, administration mode and score target separately. This lets a translated instrument retain its language identity without automatically pretending that translation created a different scoring structure. It also lets a short form or genuinely modified scoring structure remain distinct even when the family name of the PROM stays the same.

Source-reported and reviewer-created values must remain distinguishable

COSMIN allows reviewers to calculate some measurement-error parameters when the article provides the correct inputs. For example, the current manual gives several formulas for deriving SEM, SDC and limits of agreement under different statistical models.² A calculation made by the review team is still not the same thing as a value reported by the authors.

The form therefore records a source status for every result. A directly reported coefficient remains labelled as source reported. A value supplied by an author after contact remains author supplied. A calculation performed by the review team remains reviewer calculated. A value estimated from a figure remains reconstructed or digitized. This separation also accords with PRISMA-COSMIN, which asks authors to indicate results that were computed or estimated rather than directly reported.³

What should travel with each property result?

The correct packet differs by measurement property. The purpose of the notes below is not to reproduce COSMIN criteria or Risk of Bias standards. It is to prevent extraction from stripping away information that later reviewers need to understand what the study actually reported.

Worked example: three numbers from one repeated-measures study

Conflicting reports should remain visible until they are resolved

A main article, supplement, later correction and author correspondence can disagree. There is no defensible universal rule that a supplementary table always outranks the article narrative or that an informal email automatically replaces a published value. First determine whether the values actually refer to the same model, sample, subgroup and time point. A formal correction is different from unpublished clarification. If the conflict remains consequential, preserve the competing values and document how the review team resolved, or failed to resolve, the discrepancy.

PRISMA-COSMIN specifically asks reviews to report decision rules used when selecting data from multiple reports and the steps taken to resolve inconsistencies across reports.³ The form therefore provides report, study and conflict fields instead of forcing one article to become the unquestioned master record.

Do not use blank cells as a missing-data language

An empty field can mean too many things: the article did not report the statistic, the property was not evaluated, the supplement is unavailable, the value is visible only in a graph, the result is genuinely not applicable, or the extractor simply forgot the field. These states should not look identical.

Publication-level missing information must also remain separate from participant-level missing PROM responses. The latter belongs to the primary study’s design and analysis. A third problem, possible selective non-reporting of a measured property, may have implications for appraisal or interpretation. The extraction record should preserve what was observed in the source and leave the formal methodological judgment to the appropriate appraisal stage.

The purpose of extraction is recoverability

Good extraction reduces the amount of reconstruction required later. A reviewer looking at the final dataset should be able to tell which report contained the information, which sample generated it, which PROM structure and score were evaluated, what analysis produced the result, what the authors reported, what the review team later calculated, and where any uncertainty remains.

That is also why the final record should remain more granular than a publication table. Reporting tables summarize evidence for readers; the extraction record supports the review team before those summaries exist. The 2025 COSMIN table templates recommend that each data point be placed in its own cell where possible and distinguish PROM characteristics, interpretability, feasibility, study characteristics, evaluation of measurement properties and summary findings.⁴ A granular extraction record preserves the source information needed to populate those later outputs without copying the official table layouts.