Written and maintained by CASRAI Editorial Board
Last updated
RECORD (REporting of studies Conducted using Observational Routinely-collected health Data) is the reporting guideline for studies built on data that were not collected for research in the first place — administrative claims, disease registries, and electronic health record (EHR) extracts. It is a formal extension of the STROBE checklist, adding 13 items that address problems STROBE’s base 22 items do not cover: how a study population was assembled from codes and algorithms, whether those codes were validated, how separate data sources were linked, and what a secondary-use dataset’s limitations mean for the findings. RECORD-PE is a further, more specific extension for pharmacoepidemiology — studies of drug effects using the same kind of routinely collected data.
RECORD at a glance
| Question | Answer |
|---|---|
| What it applies to | Observational studies using routinely collected health data — administrative/claims databases, disease or clinical registries, EHR extracts — rather than data collected for the specific study |
| Relationship to STROBE | An extension: 13 additional items, each presented alongside the corresponding base STROBE item it extends |
| Citation | Benchimol EI, Smeeth L, Guttmann A, Harron K, Moher D, Petersen I, et al., for the RECORD Working Committee. “The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) Statement.” PLOS Medicine 2015;12(10):e1001885. |
| Pharmacoepidemiology extension | RECORD-PE — 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. |
| Maintained by | The RECORD Working Committee; catalogued by the EQUATOR Network |
Why routinely collected data needed its own extension
STROBE was written for observational studies broadly, on the assumption that the study team designed the data collection around the research question. Routinely collected health data inverts that: the EHR, claims system, or registry was built to support care delivery, billing, or surveillance, and the researcher is repurposing it after the fact. That gap creates reporting problems STROBE’s base items were never written to catch — how the study population was actually identified (a diagnosis code list? a validated algorithm? a free-text search?), whether those codes have been checked against a reference standard, how records from separate systems were linked together and how much of that linkage succeeded, and what a data source’s coverage gaps mean for who is and isn’t represented in the results. The RECORD statement exists to make authors report all of that explicitly, rather than leaving it implicit in a paragraph about “the database.”
This distinction matters more as real-world evidence (RWE) plays a larger role in regulatory and health-policy decisions in the US, UK, Canada, and Australia — agencies increasingly expect RECORD-style transparency about data provenance before treating a routinely-collected-data study as evidence-grade.
The 13 RECORD items
Each RECORD item is numbered to match the STROBE item it extends (so, for example, items 6.1–6.3 all extend STROBE item 6, on eligibility criteria and participant selection). Authors complete the full STROBE checklist first, then add these on top wherever they apply.
| Item | What it requires | Extends STROBE item |
|---|---|---|
| 1.1 | Name the type of data used and the specific database(s) in the title or abstract | 1 (Title and abstract) |
| 1.2 | Report the geographic region and timeframe the data covers, in the title or abstract | 1 |
| 1.3 | State in the title or abstract if the study involved linking two or more databases | 1 |
| 6.1 | Describe in detail the methods used to select the study population — codes or algorithms used to identify participants | 6 (Participants) |
| 6.2 | Reference any validation studies of the codes or algorithms used to identify the study population, exposures, outcomes, and confounders | 6 |
| 6.3 | Where the study involved data linkage, use a flow diagram showing the number of individuals at each linkage stage | 6 |
| 7.1 | Provide the complete list of codes and algorithms used to classify exposures, outcomes, confounders, and effect modifiers, ideally in supplementary material | 7 (Variables) |
| 12.1 | Describe how the investigators’ degree of access to the underlying database was determined | 12 (Statistical methods) |
| 12.2 | Describe methods of data cleaning | 12 |
| 12.3 | Where applicable, describe the linkage techniques used and how linkage quality was evaluated | 12 |
| 13.1 | Describe, in detail, the selection of the study population — including filtering on data quality, data availability, and linkage success — ideally with a flow diagram | 13 (Participants – Results) |
| 19.1 | Discuss the implications of using data not originally collected for research purposes — including possible biases and unmeasured confounding this introduces | 19 (Limitations) |
| 22.1 | State whether the datasets and code lists used are available, and if so, how to access them (e.g. a data-sharing statement or repository/protocol reference) | 22 (Other information) |
RECORD-PE: the pharmacoepidemiology extension
RECORD-PE builds directly on top of RECORD, adding items aimed specifically at drug-effect studies conducted with routinely collected data. Pharmacoepidemiology studies carry reporting concerns RECORD’s general items don’t fully address — how drug exposure was defined and dated, how a comparator group was chosen, and how the study handled biases that are especially common in this design (immortal time bias, confounding by indication, protopathic bias). RECORD-PE was developed by an overlapping international working group (many of the same authors as RECORD, plus pharmacoepidemiology and pharmacoepidemiological-methods specialists) and published in BMJ in 2018.
Use RECORD-PE, on top of the base RECORD and STROBE items, whenever a study is: (a) built on routinely collected health data, and (b) specifically evaluating a drug’s effect, safety, or comparative effectiveness — a cohort study of a medication’s cardiovascular risk using claims data, or a case-control study of a drug-adverse-event association drawn from an EHR system, are the kind of designs it targets. A study using routinely collected data but not about a drug exposure (for example, a disease-incidence study from a registry) should use RECORD, not RECORD-PE.
Applying RECORD when writing up a study
- Start from the base STROBE checklist for the underlying design — cohort, case-control, or cross-sectional — and complete every item as you normally would.
- Work through the 13 RECORD items above alongside the STROBE item each one extends, rather than treating RECORD as a separate document to fill in afterward.
- If the study is a pharmacoepidemiology study, add the RECORD-PE items on top; check the current RECORD-PE checklist document itself for the exact item wording, since (unlike the base RECORD checklist) this page has not independently verified RECORD-PE’s full item list against the primary BMJ article.
- Reference the specific codes and algorithms used to define the population, exposures, and outcomes as supplementary material, not just in prose — this is one of RECORD’s most consistently under-reported requirements, since it asks for something (a literal, reusable code list) that base STROBE reporting never required.
- Report data linkage as its own numbered step, with a flow diagram showing counts at each stage where the study links more than one database — this is the RECORD equivalent of a CONSORT-style participant flow diagram, applied to record linkage instead of trial recruitment.
RECORD, STROBE, and where each one applies
RECORD does not replace STROBE — every RECORD item is defined as an addition to a specific STROBE item, and a RECORD-compliant report is a STROBE-compliant report with 13 more items answered. The distinction that matters when choosing which checklist(s) to complete is the data source, not the study design: a cohort study built from prospectively collected, purpose-built research data only needs STROBE; the same cohort design built from an administrative database, disease registry, or EHR extract needs STROBE plus RECORD, and RECORD-PE on top of that if it’s evaluating a drug exposure.
Frequently asked questions
What does RECORD stand for?
REporting of studies Conducted using Observational Routinely-collected health Data. It was published by Benchimol, Smeeth, Guttmann, Harron, Moher, Petersen and colleagues for the RECORD Working Committee, in PLOS Medicine in 2015.
Is the RECORD statement required for publication?
Requirements vary by journal, the same way STROBE compliance does. Many journals that publish epidemiology, health-services research, or pharmacoepidemiology now request or require a completed RECORD (and, where relevant, RECORD-PE) checklist for studies built on administrative, claims, registry, or EHR data — check the target journal’s specific author instructions rather than assuming.
What’s the difference between RECORD and RECORD-PE?
RECORD is the general extension for any observational study using routinely collected health data. RECORD-PE is a further, more specific extension on top of RECORD, for pharmacoepidemiology studies — those specifically evaluating a drug’s effect, safety, or comparative effectiveness. A non-drug study using routinely collected data needs RECORD only; a drug-effect study using that kind of data needs RECORD plus RECORD-PE.
Does RECORD replace STROBE?
No. RECORD’s 13 items are each written as an addition to a specific STROBE item — authors complete the full STROBE checklist for their design (cohort, case-control, or cross-sectional) and then add the RECORD items on top wherever the study uses routinely collected data.
Where can I find the official RECORD checklist document?
The RECORD Working Committee maintains the checklist and supporting documents; the underlying items are also reproduced in the original 2015 PLOS Medicine article (Benchimol et al., DOI 10.1371/journal.pmed.1001885) and, for RECORD-PE, the 2018 BMJ article (Langan et al., DOI within BMJ 2018;363:k3532).
