Trial protocol and statistical analysis plan

A protocol is a long document, often more than a hundred pages, that states the rationale and objectives, the design, the eligibility criteria, the treatments and their dose modifications, the schedule of visits and assessments, the endpoints and how each is defined and measured, the sample size and its justification, the interim analyses and stopping rules, the safety reporting rules, and the governance (sponsor, steering committee, data monitoring committee, ethics oversight). The SPIRIT statement lists what a protocol should contain, and journals increasingly publish protocols so that the final report can be checked against them. The statistical analysis plan is a companion document finalised before the data are unblinded that fixes the analysis populations, the handling of missing data and intercurrent events, the order in which endpoints are tested and the alpha each receives, the pre-specified subgroups and the sensitivity analyses. Together they are what makes pre-specified mean something.

Protocols change. Amendments add sites, adjust eligibility when recruitment is slow, revise dose modifications in response to toxicity, or, more consequentially, change the primary endpoint or the sample size; each amendment goes back to the ethics committee and regulators and is dated in the registry record, so a reader can see what changed and when. Deviations are departures from the protocol for individual patients (a missed scan, a dose given outside its window) and are logged and classified by seriousness; a pattern of deviations at one site is a monitoring finding. The strict week-twelve event definition written into BELINDA's protocol, combined with a long manufacturing interval, is a documented example of protocol choices deciding a trial's verdict: the trial found no difference in event-free survival where two similar trials found a large one.

For readers, the protocol answers the questions a paper leaves out: exactly who was eligible, what counted as progression, how many analyses were planned, and whether the headline analysis was the one specified. Registries link to protocols for many trials, and supplementary appendices to trial papers usually contain them. A result that cannot be traced to a pre-specified analysis in a dated protocol is a hypothesis, however small its p-value.