STROBE

What it is

— STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) is a 22-item reporting checklist that specifies the minimum content a completed observational study should report across its title/abstract, introduction, methods, results, and discussion. It was developed by an international collaboration of epidemiologists, methodologists, statisticians, and journal editors and published in 2007 as the STROBE Statement (von Elm et al.) with a companion Explanation-and-Elaboration paper (Vandenbroucke et al.) that gives the rationale and worked examples for each item. STROBE is maintained as a reporting guideline within the EQUATOR Network library and is the parent statement from which all design- and domain-specific extensions descend. Its three core designs — cohort, case-control, and cross-sectional — share 18 common items, with 4 items reported differently by design. STROBE is a reporting tool: it tells authors, reviewers, and readers what must be disclosed so a study can be understood and critically appraised. It does not prescribe how to design or analyze the study, and it does not score quality.

When to use

— Apply STROBE when you are reporting (or refereeing, or registering a reporting plan for) a primary observational study of one of the three core designs: prospective or retrospective cohort, case-control, or cross-sectional. It is the default reporting backbone for an observational manuscript in a peer-reviewed journal, the reporting appendix of an HTA/payer dossier built on a non-interventional study, and the transparency layer of an FDA/EMA real-world-evidence submission or PASS report. Decision rule for choosing STROBE vs an extension: use the base STROBE only when no more specific extension governs your design or data. If the study uses routinely-collected health data (claims, EHR, disease/administrative registries, linked databases), the routinely-collected-data extension RECORD applies, and for pharmacoepidemiology specifically RECORD-PE — these add items on database provenance, code lists, data-cleaning, and linkage that base STROBE does not cover. Use STROBE-MR for Mendelian randomization, STROME-ID for infectious-disease molecular epidemiology and STROBE-ME for molecular epidemiology (biomarkers), STROBE-NI for observational studies of newborn infection, STROBE-RDS for respondent-driven sampling, and the veterinary/nutritional/equity variants where they fit. STROBE governs reporting of the completed study; the protocol of a primary RWE study is pre-specified with HARPER, StaRT-RWE, or the ENCePP checklist, not STROBE; and a systematic review is reported with PRISMA, not STROBE.

What it requires

— STROBE's 22 items compel disclosure of the elements that, when left vague, make an observational result un-interpretable. Substantive domains, framed for real-world data: design transparency (item 1 — name the design in the title/abstract; item 4 — present key elements of the design early); setting and time anchors (item 5 — setting, locations, and the relevant dates of recruitment, exposure, follow-up, and data collection, which in RWD means the index/time-zero definition and the lookback and follow-up windows); eligibility and participant flow (items 6 and 13 — sources, selection methods, and a numeric account of participants at each stage, the attrition funnel); variable definitions (item 7 — explicit operational definitions of outcomes, exposures, predictors, confounders, and effect modifiers, which for claims/EHR means the phenotype/algorithm logic and code lists); measurement and data sources (item 8 — sources and methods of assessment, including comparability across data sources); bias (item 9 — efforts to address potential sources of bias); study size and quantitative variable handling (items 10-11); statistical methods (item 12 — all methods including how confounding was controlled, how subgroups/interactions and missing data were handled, and how loss to follow-up was addressed); results (items 13-17 — flow, descriptive data, outcome counts, and crucially item 16 — both unadjusted and confounder-adjusted estimates with precision, plus the confounders adjusted for); and interpretation (item 18-20 — key results, limitations including direction and magnitude of potential bias, and cautious generalizability). Note what STROBE does not itself mandate but a credible RWE study should report alongside it: fitness-for-purpose of the data source, phenotype/algorithm validation metrics (PPV/sensitivity), estimand and intercurrent-event handling, positivity/overlap diagnostics, and quantitative bias analysis — these are the substance that base STROBE's generic items (7, 9, 12, 19) only gesture at, and which RECORD-PE and the catalog concepts below make explicit.

When NOT to use — limitations and common misapplications

— (1) A reporting checklist is not a risk-of-bias instrument and not a quality score. STROBE tells you whether a study is described completely, not whether it is valid. Do not sum ticked items into a "STROBE score" to rank studies — the developers explicitly warn against this; appraise validity with ROBINS-I (or ROBINS-E), the Newcastle-Ottawa Scale, or a domain-based tool instead. (2) Completing the checklist does not make an observational study causal or unconfounded. A fully STROBE-compliant paper can still rest on a hopelessly confounded design; transparency is necessary, not sufficient. (3) Using STROBE where RECORD/RECORD-PE is required. A claims- or EHR-based pharmacoepidemiologic study reported against base STROBE alone will omit the database provenance, code-list, linkage, and data-cleaning items that RECORD-PE exists to enforce — a recurring reviewer rejection in regulatory and HTA submissions. (4) Wrong extension for the design — citing base STROBE for a Mendelian randomization study (needs STROBE-MR) or a routinely-collected-health-data study (needs RECORD/RECORD-PE), or citing STROBE for a randomized trial (CONSORT) or systematic review (PRISMA). (5) Checklist-as-theater — attaching a completed checklist whose page references point to text that is itself vague (an "outcomes were defined using ICD codes" with no code list, no validation, no time window) satisfies the box but defeats the purpose; the value is the content disclosed, not the completed grid.

How it maps to this catalog

— STROBE's generic items become concrete when each is implemented by a catalog concept the author can point the page reference at: - Design transparency / eligibility / time anchors (items 1, 4-6): the comparative design is built with active-comparator-new-user and, for the formal causal contrast, target-trial-emulation; the index date and immortal-time-safe alignment that item 5's "relevant dates" demand are implemented by time-zero-index-date-alignment-rwe. - Variable definitions and measurement (items 7-8): outcome/exposure operational definitions and their code logic are implemented by diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe, with validity established by algorithm-validation; the PICOTS spine for these definitions is picots-framework-rwe. - Bias and confounding control (items 9, 12, 16): confounding adjustment that item 16's "confounder-adjusted estimates" presupposes is implemented by high-dimensional-propensity-score-hdps-rwe, with balance reporting by baseline-characteristics-and-covariate-balance-rwe; the estimand and intercurrent-event framing behind item 12 is estimands-ate-att-intercurrent-events-rwe. - Participant flow and follow-up (items 12-13): the attrition funnel and informative loss to follow-up are implemented by attrition-and-loss-to-follow-up-rwe. - Limitations and sensitivity (item 19): quantitative bias analysis for residual confounding is implemented by e-value-sensitivity-analysis. - Data fitness (underpins items 5, 8, 19): fit-for-purpose-data-assessment-rwe and the data-source operational depth in claims-analysis supply the provenance and limitations that base STROBE only gestures at and that RECORD-PE makes mandatory.

Applied note (claims/EHR/registry RWE)

For a claims- or EHR-based cohort, report against STROBE and RECORD-PE: item 5's "relevant dates" must specify the index/time-zero rule, lookback, and follow-up windows; item 6 must give the eligibility/enrollment requirements and a numeric attrition funnel from source population to analytic cohort; item 7 must publish the phenotype algorithm and code lists with validation metrics rather than a bare ICD reference; item 12 must state the confounding-control method (e.g., hdPS) and the estimand; and item 16 must present unadjusted and adjusted effect estimates with the covariate set. Treat the checklist as a map from each reporting obligation to a specific, versioned artifact — protocol section, code commit, validation result, or diagnostic plot — not as a final-page formality.