RECORD

What it is

RECORD (REporting of studies Conducted using Observational Routinely-collected health Data) is a reporting guideline published in 2015 (Benchimol et al., PLOS Medicine) that extends the STROBE statement for observational research using data not collected for research purposes: electronic health records, administrative claims, disease and product registries, and linkages among them. It is not a new checklist but an extension — authors must satisfy the 22 STROBE items and the RECORD additions. The RECORD items target the failure points specific to routinely-collected data: explicit naming of the data source(s) in the title/abstract (RECORD 1.1–1.3), the population-selection process and the codes/algorithms used to define populations, exposures, outcomes, and confounders, with their validation (RECORD 6.1–6.3, 7.1), data linkage and the linkage quality assessment (RECORD 12.1–12.3), a population flow diagram, and statements on data cleaning, access, and availability of code lists. RECORD is maintained by its author group and hosted in the EQUATOR Network library (record-statement.org). The principal published extension is RECORD-PE (Langan et al., 2018, BMJ) for pharmacoepidemiology studies of medication effects. A reporting extension for NLP-derived variables (provisionally "RECORD-NLP") has been proposed but is not yet a finalized published checklist as of 2026 — treat it as anticipated guidance, not an established standard.

When to use

Reach for RECORD whenever a non-interventional study analyses routinely-collected health data and the output is destined for a peer-reviewed journal, an HTA/payer dossier, a regulatory submission (FDA RWE program, EMA), or a registered protocol. It is the reporting backbone for claims, EHR, registry, linked, and multi-database studies. Decision rules for which member of the family applies:

• If the study evaluates the effect of a medication, vaccine, or other intervention (comparative safety or effectiveness, drug utilisation), use RECORD-PE — it adds pharmacoepidemiology items (exposure windows, new-user/active-comparator design, time-zero, washout) that plain RECORD does not enforce.
• If key variables (phenotypes, outcomes) are extracted from clinical free text via natural-language processing, document the text source, NLP model/method, and validation of the derived phenotype (the proposed RECORD-NLP extension is not yet finalized — until it is, report these elements under RECORD/RECORD-PE plus the relevant NLP-validation literature).
• If the data were collected for research (prospective cohort, primary-data registry trial, surveys), plain STROBE (or the relevant STROBE extension) suffices — RECORD's data-provenance items add little.
• RECORD is a reporting tool. For design pre-specification of an RWD study use a protocol template (HARPER, STaRT-RWE, ENCePP Checklist); for HTA reference-case alignment use NICE/CADTH frameworks. RECORD governs how you report what you did, not how you design it.

What it requires

The substantive item clusters RECORD enforces, framed for real-world data:

• Data-source transparency (RECORD 1.1, 6.1–6.2): name every database, the dates and geographic/health-system coverage, and the population-selection cascade from source to analytic cohort.
• Data fitness-for-use (RECORD 6.3, 13): completeness, representativeness, and the cleaning/validation steps — the reader must be able to judge whether the data can answer the question.
• Phenotype / algorithm transparency and validation (RECORD 6.1, 7.1): a complete list of codes and algorithms used to define populations, exposures, outcomes, and confounders, with references to validation (PPV, sensitivity) where available. Code lists should be made available.
• Population flow and attrition (RECORD 13.1): a flow diagram from the source population through each eligibility and exclusion step to the analytic sample.
• Data linkage and its quality (RECORD 12.1–12.3): linkage methods, the proportion linked, and the impact of incomplete linkage on the study population.
• Access and reproducibility (RECORD 22.1): how others could access the data/code and any approvals required. RECORD does not itself prescribe estimands, confounding control, or sensitivity analysis — but because it requires you to report exposure/outcome definitions, time windows, and analytic decisions transparently, applying it well forces the underlying RWE methods (time-zero alignment, active-comparator design, hdPS, quantitative bias analysis) to be specified and defended.

When NOT to use — limitations and common misapplications

• RECORD is a reporting checklist, not a risk-of-bias instrument and not a quality score. Ticking every item certifies completeness of reporting, not validity. A perfectly RECORD-compliant paper can still be hopelessly confounded. For critical appraisal use ROBINS-I, the Newcastle-Ottawa Scale, or ISPE/ISPOR good-practice recommendations — not RECORD.
• Completing the checklist does not make an observational study causal. Transparent reporting of an immortal-time-biased or prevalent-user design is still a biased design, fully reported.
• Using STROBE alone for routinely-collected data. STROBE omits the data-provenance, code-list, validation, and linkage items that are the whole point of RECORD; a claims/EHR study reported only to STROBE will be sent back by informed reviewers.
• Using plain RECORD where RECORD-PE is required. A comparative drug-safety study reported to RECORD but not RECORD-PE will under-report exposure definition, new-user/active-comparator design, time-zero, and washout — the items that determine whether the comparison is interpretable.
• Wrong extension for the design (e.g., RECORD where an NLP-derived phenotype demands RECORD-NLP).
• Checklist-as-theater. Appending a completed RECORD table to a manuscript when the protocol never pre-specified the code lists, validation, or analytic decisions is retrofitting, not transparency. Pre-specify in the protocol (HARPER/STaRT-RWE), then report against RECORD.

How it maps to this catalog

Each RECORD requirement is implemented by one or more concepts in this repo:

• Data fitness (RECORD 6.1–6.3, 13) → `fit-for-purpose-data-assessment-rwe` and `database-feasibility-attrition-funnel-rwe`; payer-specific completeness via `medicare-ffs-ma-commercial-claims-differences-rwe` and `claims-analysis`.
• Population/exposure/outcome codes and algorithms with validation (RECORD 6.1, 7.1) → `diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe`, `claims-outcome-algorithm-ppv-sensitivity-rwe`, `ehr-phenotyping-algorithms-rwe`, and `algorithm-validation`.
• Time-zero / index-date alignment and washout (reported under design/exposure) → `time-zero-index-date-alignment-rwe`, `washout-clean-lookback-period-rwe`, `active-comparator-new-user`.
• Estimands and intercurrent events → `estimands-ate-att-intercurrent-events-rwe`; design emulation via `target-trial-emulation`.
• Confounding control to report → `high-dimensional-propensity-score-hdps-rwe`, `propensity-score-methods-psm-iptw`.
• Population flow / attrition (RECORD 13.1) → `attrition-and-loss-to-follow-up-rwe` and `continuous-enrollment-observable-time-rwe`.
• Data linkage and its quality (RECORD 12.1–12.3) → `linked-data` and, for the mother-infant case, `mother-infant-linkage-rwe`.
• Sensitivity / quantitative bias analysis to report → `e-value-sensitivity-analysis` and `quantitative-bias-analysis-toolkit-rwe`.
• Reporting visuals → `visualizations-pharmacoepidemiology-rwe` for the population flow diagram and balance/diagnostic plots.

Applied note (claims/EHR/registry RWE)

In a Medicare + commercial claims comparative-safety study, RECORD compliance means: name each database and its dates and benefit type in the abstract; report the exact NDC/ICD/CPT code lists and the phenotype rules (e.g., 1 inpatient or 2 outpatient diagnoses within a window) for exposures, outcomes, and confounders, with PPV/sensitivity references; show the population flow from enrollees through continuous-enrollment and washout filters to the analytic cohort; and, because it is a drug study, report against RECORD-PE so that the new-user/active-comparator structure, time-zero, and exposure windows are explicit. Where Medicare Advantage encounter completeness or fee-for-service claim capture affects the cohort, state it under data fitness — that disclosure is exactly what RECORD 6.3 exists to elicit.

Citations

1. [1]Benchimol EI, Smeeth L, Guttmann A, et al. The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) Statement. PLOS Medicine. 2015;12(10):e1001885.
2. [2]von Elm E, Altman DG, Egger M, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: guidelines for reporting observational studies. PLoS Medicine. 2007;4(10):e296.
3. [3]Langan SM, Schmidt SAJ, Wing K, et al. The reporting of studies conducted using observational routinely collected health data statement for pharmacoepidemiology (RECORD-PE). BMJ. 2018;363:k3532.

1. [4]RECORD Statement, EQUATOR Network — maintained checklists, extensions (RECORD-PE, RECORD-NLP), and guidance.