E6(R3) Compliance in a Connected eClinical Environment

Here is a question an inspector can reasonably ask about any trial. Which version of the consent form did participant 0412 sign, on what date, and was that the version in effect at the time?

In a site-based study, the answer lives in one place. In a study with remote consent, home nursing visits and device-collected data, the same question routes through an eConsent vendor's audit log, the trial master file, the randomisation system and, more often than anyone would like to admit, a spreadsheet at the site.

Nothing about the second study is non-compliant. Consent was properly obtained, the version was correct, the site did its job. The entire difference is how much work it takes to prove it. 

What Annex 2 adds

ICH E6(R3) arrived in two parts. The principles and Annex 1 took effect in the EU on 23 July 2025. Annex 2, covering pragmatic designs, decentralised elements and trials using real-world data, reached Step 4 in June 2026 and comes into effect on 15 January 2027.

The two ask for different kinds of work. Annex 1 formalised a direction the industry was already travelling, and most of it could be discharged through procedure, training and documentation.

Annex 2 applies GCP expectations to designs that scaled under pandemic-era flexibilities. That was the right response at the time. What it produced is a body of practice whose supporting evidence was assembled for operational purposes rather than for inspection. A sponsor can hold an excellent decentralised trial SOP and still be unable to demonstrate compliance with it, because the records that would do so sit across four vendors in three formats with no common key.

So the sponsors with the most work ahead are generally not those with weak procedures. They are those with good procedures spread across too many systems.

Four gaps, one cause

Consent: Annex 2 requires showing, per participant, which version was presented, that consent preceded any procedure, and that re-consent followed each amendment within the required window. In a fragmented stack those four facts sit in four places, joined manually on subject identifiers that may not match. The architectural answer is to make consent an attribute of the subject record itself, under the same audit trail as screening and randomisation, with amendment approval propagating re-consent requirements automatically rather than by email.

Data provenance: When values arrive from wearables, participant apps, local labs and EHR extracts, ALCOA+ does not relax because the path is complicated. Most sponsors can draw the data flow; far fewer can evidence the lineage of a specific data point from origin to listing. That requires provenance as a first-class property: origin, ingestion context, transformation history and attribution travelling with the value rather than being lost at each system boundary.

Investigator oversight: In a site-centric trial, oversight is self-evidencing. In a decentralised one, the investigator is accountable for activity they did not observe, and nothing in the workflow generates a record of their review. This gap closes almost entirely through design: delegation of authority as an enforced, system-level control rather than a scanned log, so the system knows whether the person who performed a home visit was permitted to; and investigator review queues that turn oversight into a timestamped, attributable action.

Vendor accountability: Home nursing, telehealth, device supply and logistics providers each hold part of the record. Oversight by monthly report proves a report was received, not that the sponsor had visibility as activity occurred. The stronger model gives third parties scoped, role-based access into the sponsor's environment, so their work lands as a contemporaneous record under the sponsor's audit trail rather than arriving later as a PDF.

The common cause is that the evidence needed to demonstrate compliance sits in a different system from the evidence it must connect to.

Why fragmentation compounds?

Each additional system carries its own validation package, periodic review, audit trail format, access model and vendor change control. A study running six point solutions has six validation lifecycles and fifteen possible pairwise reconciliations. The same study on one platform has one boundary and no reconciliation, because the data was never separated.

This also means remediation compounds favourably. Removing one system removes a validation lifecycle, a periodic review and every reconciliation that system participated in.

Fewer systems, fewer seams

Cloudbyz brings eConsent, ePRO and eCOA, eSource, EDC, CTMS and eTMF onto one platform under a single validation boundary and one audit trail.

In practice, for Annex 2: consent versions and re-consent status are attributes of the subject record. Remote-captured data carries provenance through to the analysis dataset within one data model. Delegation of authority is an enforced role-based control. Investigator review queues generate attributable oversight records. Third parties work through scoped portal roles rather than exchanging files. TMF filing happens as work occurs, with completeness visible continuously rather than assessed before an inspection.

All of it under 21 CFR Part 11 audit trails and electronic signature controls, so the validation question is asked once about one platform rather than repeatedly about six.

Consolidation is not always available mid-study, and where it is not, structured ingestion with retained attribution is the realistic interim. But for studies still running in January 2027, the difference between a readiness review and an archaeology project is decided by how many boundaries the evidence has to cross.