CONSORT Extension for Pragmatic Trials
A reporting checklist (8 extended CONSORT items) for randomized pragmatic trials conducted in routine-care settings, prompting authors to describe how the trial's eligibility, intervention delivery, setting, outcomes, and analysis reflect real-world practice rather than an idealized explanatory design.
What it is
The CONSORT extension for pragmatic trials (Zwarenstein et al., BMJ 2008) is a reporting guideline, not a design or risk-of-bias tool. Published against the then-current CONSORT 2001 statement and used today alongside CONSORT 2010, it provides extended guidance on 8 checklist items (eligibility, interventions, outcomes, sample size, blinding, participant flow, generalizability, and interpretation) so that readers of a pragmatic randomized trial can judge how applicable the results are to routine practice. It is curated through the EQUATOR Network and the CONSORT Group and is meant to be used alongside — never instead of — the parent CONSORT checklist and the relevant trial registration. Its conceptual companion for design is the PRECIS-2 tool (Loudon et al., BMJ 2015), which scores how explanatory-versus-pragmatic a trial is across nine domains; CONSORT-Pragmatic governs how that pragmatism is reported.
When to use
Use it whenever you report a randomized trial that deliberately measures effectiveness under usual-care conditions — flexible delivery, broad eligibility, clinically meaningful endpoints, minimal protocol-driven extra visits — and especially when the trial is embedded in real-world data infrastructure (a registry-based randomized trial, a pragmatic trial with EHR/claims-ascertained outcomes, or a cluster-randomized implementation trial). It is the right checklist for HTA/payer dossiers and journal submissions arguing external validity, and it supports FDA/EMA interest in pragmatic effectiveness evidence. Decision rule for choosing the right CONSORT family member: if the unit of randomization is the cluster, layer CONSORT for cluster trials; if you report patient-reported outcomes, harms, or non-inferiority, add those extensions; if your study is non-randomized real-world evidence, CONSORT does not apply at all — use STROBE/RECORD-PE for reporting and target-trial emulation for design. Reach for CONSORT-Pragmatic (over plain CONSORT alone) precisely when the trial's value proposition is generalizability to routine care.
What it requires
The extension forces authors to make pragmatism explicit and auditable. Its substantive domains, framed for trials that touch real-world data: (1) eligibility and setting — describe the participants, practitioners, and care settings, and how closely they mirror the population the intervention targets in practice; (2) intervention description — state how flexibly each intervention was delivered and the resources/expertise assumed, since pragmatic trials permit clinician judgment rather than rigid protocols; (3) outcomes — justify endpoints as directly relevant to participants, clinicians, or payers, and report how they were ascertained (including registry/EHR/claims-based capture, which raises the same phenotype-validation and time-window questions as observational RWE); (4) participant flow and follow-up — a CONSORT flow diagram with explicit accounting of attrition and loss to follow-up, which in routine-care settings is often substantial and informative; (5) analysis and estimand — pre-specify the analysis population (ITT under a treatment-policy strategy is typical for pragmatic effectiveness) and how intercurrent events (non-adherence, treatment switching, crossover) are handled; (6) generalizability and interpretation — discuss applicability to other populations, settings, and usual care. Where outcomes or covariates are drawn from secondary data, the report should document data fitness-for-use and algorithm definitions to the same standard expected of observational RWE.
When NOT to use — limitations and common misapplications
(a) It is a reporting checklist, so completing it improves transparency but does not reduce bias, raise study quality, or certify internal validity — a fully reported pragmatic trial can still be confounded by poor allocation concealment or differential attrition; do not treat the checklist as a risk-of-bias instrument (use RoB 2) or a quality score. (b) It applies only to randomized trials; using it to dress up a non-randomized database study is a category error — that study needs RECORD-PE/STROBE, and randomization language will mislead reviewers. (c) Checklist-as-theater: ticking items in a submission appendix without the corresponding detail in the manuscript defeats the purpose; page numbers must point to real content. (d) Wrong family member: a cluster-randomized pragmatic trial reported with only the patient-level extension will under-report design effects and recruitment-after-randomization bias. (e) Pragmatic ≠ low-rigor: the extension does not license loose endpoint ascertainment — EHR/claims-based outcomes still require validated phenotypes. (f) It does not, by itself, make an effectiveness estimate causal or transportable; that depends on design and analysis, not on the report.
How it maps to this catalog
Several catalog concepts implement what the extension asks authors to report. Estimand and intercurrent-event handling (treatment-policy ITT, switching, non-adherence) is implemented by estimands-ate-att-intercurrent-events-rwe. Outcome/exposure ascertainment from secondary data — required whenever a pragmatic trial uses registry/EHR/claims endpoints — is implemented by diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe (algorithm definition, PPV validation) and claims-analysis (code lists, enrollment, data nuances). Attrition and loss-to-follow-up reporting maps to attrition-and-loss-to-follow-up-rwe and the CONSORT flow diagram. Design framing and eligibility/time-zero discipline is shared with target-trial-emulation and active-comparator-new-user (the new-user, time-zero logic that pragmatic effectiveness designs borrow), while structured question specification uses picots-framework-rwe. Data fitness-for-use is implemented by fit-for-purpose-data-assessment-rwe, and quantitative bias / sensitivity analysis of residual confounding (relevant when randomization is imperfect or outcomes are database-derived) by e-value-sensitivity-analysis. When confounding control is needed for embedded observational comparisons, see high-dimensional-propensity-score-hdps-rwe. Applied note for registry/EHR/claims-based pragmatic trials: report the outcome algorithm and its validation, continuous-enrollment/observation windows, and how loss to follow-up was handled with the same rigor you would apply to a STROBE/RECORD-PE observational study — the randomization protects the treatment contrast, but the measurement layer is pure RWE.