Protocol review is a consistency problem, not a checklist
Most protocol review runs off a list of required contents. ICH E6(R3) supplies one in Appendix B: sixteen headings, from general information through to publication policy. Confirming that each heading exists and contains something is quick, it can be assigned to anyone, and it is the part of review that gets done reliably.
It also catches the fewest problems that matter. A protocol missing a section is obvious to everyone, including the person who wrote it. A protocol that says two different things in two different sections looks complete from every angle except the one nobody checks.
The contents list guarantees repetition
Appendix B spreads related facts across separate headings, and it has to. A protocol is read by different people for different purposes, so each decision appears wherever a reader would go looking for it. The consequence is structural: a well formed protocol states its most important choices more than once.
Endpoints are specified under Trial Design at B.4.1. Objectives, and the estimands where they are defined, sit under Trial Objectives and Purpose at B.3. Different sections, usually written at different times by different people, describing the same thing.
Discontinuation is defined twice by the structure itself. B.4.8 asks for stopping rules and discontinuation criteria. B.6 asks when and how participants are discontinued, whether they are replaced, and what follow up applies to those who stop the investigational product. Both are required, and neither is redundant on paper. In a real document they overlap, and the overlap is where they drift.
Timing is the worst case. The schedule of events is at B.4.6. The expected duration of participation and the sequence of trial periods is at B.4.7. Timing of efficacy assessment is at B.8.2, safety at B.9.2, and the duration of follow up after adverse events at B.9.4. Five locations describing one calendar, before counting the schedule of assessments table and whatever the consent form told the participant.
The alignment the guideline states outright
Most of those relationships are implied. One is not.
B.10.4 requires the protocol to describe which participants are included in the planned analyses, how intercurrent events are handled, and how missing data is accounted for. It then says these should be aligned with the target estimands. That is an explicit instruction that the statistical section has to agree with the objectives section.
It is also the alignment most commonly broken, and the break is invisible inside either section. Take a primary estimand using a treatment policy strategy, where the effect of interest includes what happens after a participant stops treatment. The statistical section is written correctly and requires outcome data regardless of discontinuation.
Now read B.6, which requires the protocol to specify the type and timing of data collected for withdrawn and discontinued participants. If that section says collection ends when the participant stops the investigational product, the protocol has defined a question and then instructed sites not to gather the data needed to answer it.
Neither section is wrong. Read on its own, each is competent. Together they describe a trial that cannot produce its own primary result, and the problem does not surface until analysis, by which point the data does not exist and no amendment brings it back. This is the same failure the estimand is meant to prevent, arriving through the protocol rather than the analysis plan.
Amendments are where consistency is lost
This matters more than it used to, because protocols are not written once.
The Tufts Center for the Study of Drug Development collected data from sixteen companies covering 950 protocols and 2,188 amendments. The share of protocols with at least one amendment rose from 57 to 76 per cent. The mean number of amendments per protocol rose by 60 per cent, from 2.1 to 3.3.
The same study found 77 per cent of amendments were considered unavoidable, with regulatory agency requests and changes in study strategy the leading causes. That deserves attention, because it cuts against the conclusion usually drawn from amendment data. You cannot review your way out of most amendments. They are responses to things learned or demanded after approval.
What follows is not that review matters less. It is that review cannot be a gate at the start. A protocol carrying 3.3 amendments is re-opened three times, and each re-opening changes a few sections while leaving their mirrors untouched. The original version is usually the most internally consistent the document will ever be. Consistency decays with every revision, and revisions are the part nobody re-reads in full, because only the changed sections get circulated.
A shifted visit window arrives as a redline of B.4.6. The reviewer sees B.4.6. Whether B.8.2 still agrees with it is not a question the redline raises.
What a usable finding looks like
A review that reports section 6 as inconsistent gives the study team nothing to act on. A usable finding names both halves: what was stated and where, and what contradicts it and where. Two locations, quoted.
That is enough for an author to resolve in one pass, and enough for a reviewer to reject quickly when the finding is wrong, which some will be. It also makes the review checkable. A finding with two anchors can be verified by anyone. A finding expressed as a judgement about quality cannot, and tends to produce discussion rather than edits.
Over-specification is a cause, not a symptom
E6(R3) makes one design recommendation that sits directly on this problem. It suggests building adaptability into the protocol, for instance by giving acceptable ranges for specific provisions, on the grounds that this can reduce deviations and in some cases remove the need for an amendment. It adds the necessary limit, that adaptability must not compromise participant safety or scientific validity.
For review, the implication is that a protocol fixing every value precisely has manufactured more opportunities to disagree with itself and more reasons to be amended later. A visit window written as a range does not need an amendment when a participant arrives a day late, and cannot contradict a different exact figure elsewhere, because there is no different exact figure.
Reviewing for unnecessary precision is unusual. It is also one of the few things review can do that reduces future amendments rather than cleaning up after them.
The part that stays hard
The checklist half of protocol review will always get done. It is easy to assign and easy to evidence. The consistency half is slow, needs the whole document held in mind at once, and is the reason experienced reviewers are valuable and in short supply.
It is also where missing something costs most, because a contradiction that survives approval does not announce itself. It waits for the analysis, or the inspection, or the day a site does exactly what section 6 told it to do rather than what section 10 assumed it would.
The checks described here are collected in a free protocol consistency checklist. It runs entirely in your browser, needs no account, and nothing is uploaded.