ICH E6(R2) Good Clinical Practice

What it is

— Integrated Addendum to ICH E6(R1): Guideline for Good Clinical Practice E6(R2) (Step 4, 9 November 2016), the international Good Clinical Practice (GCP) standard maintained by the International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH) and adopted into law/guidance by FDA, EMA, PMDA, and other ICH regulators. GCP is an ethical and scientific quality standard for the conduct of clinical trials in human subjects — it defines the responsibilities of sponsors, investigators, and IRBs/IECs; informed consent; the Investigator's Brochure and protocol; trial monitoring; source documentation and data integrity (ALCOA); essential documents; and reporting of safety. E6(R2) added a risk-based quality management framework (quality-by-design, risk-based monitoring, sponsor oversight of electronic systems and vendors). It is not a reporting checklist, not a risk-of-bias instrument, and not an RWE design framework — it is a conduct standard for prospective research. Note that ICH E6(R3) (Step 4, 6 January 2025) supersedes R2 for new protocols as ICH members transition; R3 is explicitly media-neutral and builds in real-world data capture and decentralized elements, so for any newly designed trial confirm whether your region has adopted R3 before defaulting to R2.

When to use

— Apply GCP when a study has a prospective, interventional component conducted under a protocol in human subjects, even when that study also leans on real-world data. The RWE-relevant cases are narrow and specific: pragmatic (point-of-care) randomized trials; registry-based randomized trials that randomize within an existing registry; decentralized or hybrid trials that collect protocol endpoints from EHRs, wearables, or claims; and single-arm trials whose treated arm is prospectively enrolled and dosed (even if the comparator is an external real-world control). Decision rule: if a patient is assigned to an intervention by the investigator/protocol and followed prospectively, GCP applies to that arm and its data collection. If the study is secondary use of already-existing routinely collected data (a retrospective claims or EHR cohort, a non-interventional PASS, an HTA dossier built from observational evidence), GCP does not apply — the governing documents are RECORD/RECORD-PE and STROBE for reporting, HARPER/STaRT-RWE for protocol templating, ENCePP for PASS conduct, and the FDA/EMA RWE frameworks for regulatory fitness. For new prospective protocols, prefer ICH E6(R3) where adopted; reserve E6(R2) for trials initiated under R2 or in regions still transitioning.

What it requires

— For the prospective component it governs, GCP enforces: (1) a written, IRB/IEC-approved protocol and a current Investigator's Brochure; (2) documented informed consent appropriate to the data and specimens collected (including secondary use and linkage permissions when EHR/registry data feed trial endpoints); (3) investigator and site qualification and delegation of duties; (4) sponsor oversight with a risk-based quality management plan — prospective identification of critical-to-quality factors, risk-based monitoring, and defined data-integrity controls (E6(R2)'s central addition); (5) source data and source documents that are attributable, legible, contemporaneous, original, and accurate (ALCOA), with a credible audit trail when endpoints are extracted from EHRs or devices; (6) traceability from source to analysis — when a real-world data source (registry, EHR, claims feed) serves as source data, the sponsor must validate the system, control access, and document data provenance and any transformations; (7) safety reporting and essential-document retention. In hybrid designs the load-bearing GCP question is data integrity of routinely collected data used as trial source data: who validated the extract, can the audit trail reconstruct the value, and was consent adequate for the use.

When NOT to use — limitations and common misapplications

— The dominant failure mode is forcing GCP onto a retrospective database study. A claims-only or EHR-only secondary-use cohort, a non-interventional PASS, or an HTA RWE analysis is not a clinical trial; its participants were not assigned an intervention by a protocol, so GCP's machinery (IRB approval of an intervention, individual informed consent, the Investigator's Brochure, on-site monitoring, source data verification against a CRF) does not map. Pasting "conducted in accordance with ICH E6(R2)" onto such a study is checklist-as-theater and a reviewer will reject it. Equally, GCP is not a substitute for reporting or bias guidance: completing GCP conduct requirements says nothing about whether confounding was controlled, whether the phenotype was validated, or whether the study was reported transparently — those are RECORD-PE/STROBE/HARPER concerns. Do not treat GCP as a quality score (it is binary conduct compliance, audited, not rated), and do not cite E6(R2) when the protocol was designed under, and should follow, E6(R3). Finally, GCP applies to the interventional arm, not to an external real-world control: the control data are governed by RWE data-fitness and reporting standards, not by GCP.

How it maps to this catalog

— GCP intersects the catalog only where a study is prospective or hybrid. Its conduct standards attach to: pragmatic-trial and registry-trial (the protocol, consent, IRB, monitoring, and data-integrity requirements for point-of-care and registry-based randomized trials); single-arm-external-control (GCP governs the prospectively enrolled treated arm; the external control is governed by RWD-fitness concepts, not GCP); endpoint-adjudication-chart-review-rwe (GCP source-data and audit-trail requirements when trial endpoints are adjudicated from EHR/chart data); regulatory-readiness-rwe and estimand-analysis-traceability-rwe (source-to-analysis traceability and pre-specification expected of regulatory submissions); and study-protocol-or-sap-elements / picots-framework-rwe (the protocol discipline GCP requires). For pass-imposed and pass-voluntary, GCP applies only if the PASS is interventional (rare) — most PASS are non-interventional and fall under ENCePP, not GCP. Concepts that belong to retrospective RWE — claims-analysis, high-dimensional-propensity-score-hdps-rwe, diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe, time-zero-index-date-alignment-rwe, attrition-and-loss-to-follow-up-rwe, fit-for-purpose-data-assessment-rwe, and target-trial-emulation — are explicitly out of GCP scope: target-trial emulation deliberately substitutes design logic for the trial GCP would have governed and is reported under STROBE/RECORD-PE, not E6.

Applied note (hybrid claims/EHR/registry RWE). When a registry-based or decentralized trial uses an EHR or claims feed as source data for a protocol endpoint, GCP compliance hinges on system validation, a reconstructable audit trail, access control, and consent that covers the secondary use and any linkage — document these in the protocol and the data-management plan. The epidemiologic quality of that same data (phenotype PPV, completeness, transportability) is a separate question answered by the RWE concepts above; GCP and RWE-fitness standards are complementary, not interchangeable.