ICH E6(R3) Appendix B spreads the same decision across separate required sections, so a well formed protocol states its key choices more than once. This checklist walks the 30 places those statements can disagree. It is the manual version of the review our Protocol Review Agent automates.
Appendix B puts the objective in B.3 and the endpoint in B.4.1, and requires the statistics in B.10.4 to align with the estimand. Three sections, one decision.
B.4.1 asks for a specific statement of primary and secondary endpoints; B.3 asks for the objectives and, where defined, the estimands. They are written at different times and usually by different people. An endpoint that exists in one and not the other is the most common finding in a first review.
B.10.4 states outright that the handling of intercurrent events and missing data should be aligned with the target estimands. This is the one cross-reference the guideline makes explicit, and it is still commonly broken.
If the strategy is treatment policy, the effect includes what happens after a participant stops treatment, so outcome data is required regardless of discontinuation. B.6(b) specifies the type and timing of data collected for withdrawn participants. If that section stops collection at discontinuation, the protocol has defined a question and forbidden the data that answers it. Nothing recovers this at analysis.
B.10.2 requires the planned enrolment and the reasoning behind it. Where a protocol has been through amendments, the power calculation is often still built on the endpoint, effect size or population that applied two versions ago.
B.10.3 asks for the level of significance or the Bayesian success threshold. If the objectives in B.3 include more than one confirmatory question, multiplicity has to be addressed somewhere, and often is not.
The calendar appears in B.4.6, B.4.7, B.8.2, B.9.2 and B.9.4, plus the schedule of assessments table and the consent form. Amendments rarely reach all seven.
B.4.6 describes the schedule of events; B.8.2 gives the methods and timing for assessing efficacy parameters. A visit window moved in one is routinely left in the other, because a redline circulates the changed section only.
B.9.2 covers the methods, extent and timing for recording and assessing safety parameters. The same drift applies, and safety timing has consequences at sites rather than only at analysis.
B.4.7 asks for the expected duration of involvement and the sequence and duration of all trial periods including follow-up. Adding a follow-up visit without updating the stated duration makes the consent form wrong as well.
B.9.4 requires the type and duration of follow-up after adverse events and events such as pregnancies. Where that follow-up runs past the end of the trial period in B.4.7, the protocol has to say so explicitly rather than leave a site to infer it.
The table is maintained separately from the prose and is where timing contradictions concentrate, because it is the artefact people edit directly under time pressure.
B.4.8 asks for stopping rules and discontinuation criteria. B.6 asks when and how participants are discontinued. Both are required and neither is redundant on paper.
B.4.8 covers stopping rules, discontinuation criteria, dose adjustment and dose interruption. B.6(a) covers when and how to discontinue participants. In a real document these overlap, and the overlap is where they drift apart.
B.6(c) asks whether and how participants are replaced. Replacement changes what the planned enrolment in B.10.2 means and which participants enter the analysis sets in B.10.4.
B.6(d) covers follow-up for participants who discontinued the investigational product; B.9.4 covers follow-up after adverse events. A participant who stopped because of an adverse event sits in both, and the two sections often give different durations.
B.9.3 sets out how adverse events are recorded and reported. Where an event is also a discontinuation criterion, the site needs one instruction rather than two.
The population is described in B.2.6 and then operationalised in B.5. Those are different sections and often different authors.
B.2.6 gives a description of the population to be studied. Eligibility criteria in B.5.1 and B.5.2 are what actually determine it. Where the background describes a broader or narrower group than the criteria admit, the trial answers a different question from the one it claims to.
B.5.2 lists exclusions; B.7.2 lists medications permitted and not permitted before and during the trial. A drug that excludes a participant at entry but appears as permitted during treatment is a genuine and frequent contradiction.
If a criterion depends on a laboratory value or an imaging result, that assessment has to appear in the schedule of events at screening, or sites cannot apply the criterion as written.
B.5.3 asks for the mechanism for pre-screening where appropriate and for screening. Where pre-screening exists, the schedule needs to show what happens and when.
Dose and regimen are stated in the background justification, the design section and the treatment section. Three places, one regimen.
B.2.4 justifies the route, dosage, regimen and treatment period. B.4.4 describes the product and its dosage and regimen. B.7.1 sets out the treatments to be administered including doses, schedule and treatment period per arm. A dose change that reaches two of the three is the classic amendment failure.
B.4.8 covers dose adjustment and dose interruption; B.7.1 covers the criteria for dose adjustment. Two sections, one rule, and sites follow whichever they read.
B.4.9 requires accountability procedures for the investigational products including placebo and any comparators. Where a comparator was added by amendment, accountability is frequently not updated.
B.4.3(b) describes blinding as a measure to minimise bias; B.4.10 covers maintenance of randomisation codes and the procedure for breaking them. An open-label element introduced later often leaves the unblinding procedure describing a blind that no longer exists.
Quality and data handling are where the protocol meets how the trial is actually run.
B.12.1 asks for identified critical to quality factors, the associated risks and the mitigation strategies, unless documented elsewhere. Where it says elsewhere, check that the elsewhere exists.
B.14.1 specifies the data to be collected and how. An assessment in the schedule with no corresponding data specification produces a procedure nobody records.
B.14.2 requires identification of data entered directly with no prior written or electronic record, which is then the source record. This is routinely omitted and is an inspection finding rather than an analysis problem.
B.8.2 and B.9.2 both say that where a data monitoring or adjudication committee is used, its procedures, timing and activities are described in the protocol or a separate document. Naming a committee without describing it is the usual gap.
Protocols now carry a mean of 3.3 amendments. The first version is usually the most internally consistent the document will ever be, and only changed sections get circulated for review.
B.1.1 requires the title, unique identifying number and date, and requires any amendment to bear the amendment number and date.
This is the check the rest of the list exists to support. For each value the amendment changed, list the sections that state it and confirm each one was updated, rather than reviewing only the redline.
The consent form, the schedule of assessments, the data acquisition tool and the statistical analysis plan all restate protocol decisions. A change that stops at the protocol boundary leaves them contradicting it.
The guideline recommends building adaptability into the protocol, for example by including acceptable ranges for specific provisions, on the grounds that this can reduce deviations and in some cases the need for an amendment, provided it does not compromise participant safety or scientific validity. A protocol that fixes every value exactly has created more ways to contradict itself and more reasons to be amended later.
It checks internal consistency and conformance to the Appendix B structure. It has no view on whether the science is sound, whether an endpoint is the right endpoint, or whether the design answers a question worth asking. Those need a person, and they are not the parts that consume the hours.
Section references are to the ICH E6(R3) Step 4 consolidated guideline. This is a working aid, not regulatory advice, and it does not replace your own review procedures.
Our Protocol Review Agent performs these checks against a draft and returns each finding with both of its source sections quoted, so a reviewer can confirm or dismiss it in seconds.
