RECORD-PE (RECORD for Pharmacoepidemiology)
Reporting guideline that extends RECORD (and, through it, STROBE) with pharmacoepidemiology-specific items for transparent reporting of treatment-effect and drug-utilization studies conducted using routinely collected health data such as claims, EHR, and registries.
What it is
— RECORD-PE (REporting of studies Conducted using Observational Routinely-collected health Data for PharmacoEpidemiology) is a reporting checklist that layers pharmacoepidemiology-specific items onto the parent RECORD statement, which in turn extends STROBE, the base reporting guideline for observational epidemiology. The lineage matters operationally: STROBE (von Elm et al., 2007) defines the 22-item backbone for cohort, case-control, and cross-sectional studies; RECORD (Benchimol et al., 2015) adds items for studies that reuse routinely collected data (databases, linkage, code lists, data cleaning, population definition); and RECORD-PE (Langan et al., BMJ 2018) adds 15 PE-specific items and sub-items that address how drug exposure, comparators, follow-up, and confounding are defined and reported. It is a reporting tool — it governs what a completed manuscript must transparently disclose, not how to design or appraise the study. It is published in the BMJ and maintained as a RECORD/STROBE extension within the EQUATOR Network, developed with the International Society for Pharmacoepidemiology (ISPE).
When to use
— Apply RECORD-PE when reporting a completed pharmacoepidemiology study that uses routinely collected data — comparative drug safety/effectiveness cohorts (e.g., active-comparator new-user designs), drug-utilization studies, and database analyses in claims, EHR, registries, or linked sources — for a peer-reviewed journal, an HTA/payer evidence dossier, or a regulatory (FDA RWE, EMA) submission package. Use it alongside the STROBE/RECORD flow diagram and tables, since RECORD-PE does not replace its parents — a compliant manuscript satisfies STROBE, RECORD, and the RECORD-PE additions together. Decision rules for choosing the right family member: a randomized trial is reported with CONSORT, not RECORD-PE; a non-pharmacoepidemiologic observational study using routine data (e.g., a health-services or surveillance study with no drug-exposure contrast) is reported with RECORD (or plain STROBE) without the PE layer; a systematic review/meta-analysis is reported with PRISMA and its protocol with PRISMA-P; and protocol-stage pre-specification of a single PE study belongs to HARPER (Wang et al., 2022) or the ENCePP Checklist, not to RECORD-PE, which is a reporting (post-hoc) instrument.
What it requires
— RECORD-PE forces transparent reporting of exactly the design choices that, left vague, let bias hide in a routine-data study. Its substantive domains include: (1) Data source and fitness-for-use — naming the database(s), the population they capture, linkage, the time period, and known limitations of the data for the question. (2) Exposure definition — how drug exposure was operationalized from dispensing/prescribing records (code lists, days-supply, grace periods, stockpiling, exposure windows) and whether definitions were validated. (3) Comparator and design — the comparator group and rationale (active comparator vs non-user), new-user vs prevalent-user status, and how the design controls confounding by indication. (4) Time-zero / index date alignment — how follow-up start was defined for all groups so that immortal time and post-baseline adjustment are avoided. (5) Outcome and covariate phenotypes — algorithm/code-list definitions and any validation (PPV/sensitivity). (6) Confounding control — measured confounders, covariate assessment windows, and the analytic method (e.g., propensity scores, high-dimensional PS). (7) Estimand and analysis — the causal contrast, treatment strategies, intercurrent-event handling, censoring rules, competing risks. (8) Attrition, missing data, and sensitivity / quantitative bias analysis — cohort-derivation/attrition reporting, handling of missingness, and pre-specified sensitivity and negative-control analyses. A recurring RECORD-PE expectation is the public availability of code lists and algorithms, so that exposure, outcome, and covariate definitions are reproducible.
When NOT to use — limitations and common misapplications
— RECORD-PE is a reporting checklist, and most failures come from treating it as something it is not. (1) It is not a risk-of-bias instrument — completing RECORD-PE tells a reader what you did, not whether it was valid; appraise non-randomized studies with ROBINS-I, not RECORD-PE. (2) It is not a quality score — there is no total, no threshold, and no "RECORD-PE score"; manufacturing one misrepresents the guideline. (3) Checklist-as-theater — a fully checked manuscript whose exposure window, time-zero rule, or confounding strategy is still described in one vague sentence has missed the point; the deliverable is transparent, reproducible detail (and public code lists), not a completed table. (4) A complete checklist does not make an observational study causal or unconfounded — transparency is necessary, not sufficient; residual confounding, selection bias, and unfit data survive a fully compliant report. (5) Wrong family member / wrong extension — using plain STROBE where the PE-specific items are required, using RECORD-PE to report a randomized trial (use CONSORT) or a non-PE routine-data study (use RECORD), or using a reporting checklist where a protocol template (HARPER, ENCePP) is what the regulator or HTA body actually asked for, are all common and visible errors.
How it maps to this catalog
— RECORD-PE is the reporting layer; the concepts in this repo are what actually implement each of its requirements, and a reviewer should read them as the substantive backing for each checklist item: - Design and time-zero (comparator, new-user status, immortal-time avoidance) → active-comparator-new-user and, for the trial-protocol framing of the whole study, target-trial-emulation. - Exposure/outcome/covariate phenotypes and their validation → diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe and algorithm-validation (the code lists and PPV/sensitivity evidence RECORD-PE asks you to disclose). - Confounding control → high-dimensional-propensity-score-hdps-rwe (and the PS balancing referenced within active-comparator-new-user). - Estimand, treatment strategies, intercurrent events → estimands-ate-att-intercurrent-events-rwe. - Structured question / eligibility spine → picots-framework-rwe. - Data fitness-for-use and source choice → fit-for-purpose-data-assessment-rwe, medicare-ffs-ma-commercial-claims-differences-rwe, and the general claims-analysis patterns. - Attrition, missing data, and sensitivity/quantitative bias analysis → attrition-and-loss-to-follow-up-rwe and e-value-sensitivity-analysis; external validity of the reported result → generalizability-transportability-external-validity-rwe.
Applied note (claims/EHR/registry RWE)
For a Medicare/commercial claims comparative cohort, RECORD-PE compliance means the manuscript states the database and enrollment requirements (and, e.g., that Medicare Advantage-only person-time was excluded because fee-for-service claims are unavailable), publishes the NDC/diagnosis code lists used for exposure and outcome phenotypes with their validation metrics, makes the index-date/time-zero rule explicit so immortal time is auditable, reports the attrition funnel from source population to analytic cohort, and presents the confounding strategy (e.g., high-dimensional PS) with balance diagnostics and pre-specified sensitivity analyses (washout length, grace period, negative-control outcome). The reporting is the visible surface; the catalog concepts above are where the methods themselves live.