Internal conflicts that clinical trial protocols correct after enrollment: a coded analysis of 1,432 amendment changes in 60 public Phase 2 and 3 protocols

Background. Protocol amendments are the norm in clinical trials, and a share of them fix problems that were already in the documents before the first patient enrolled. We asked how large that share is in real protocols, using the amendment histories sponsors publish inside the protocol itself. Methods. From the 14,294 single-phase Phase 2 and 3 trials on ClinicalTrials.gov that posted results from 2019 onward with the full protocol attached, we drew a stratified random sample of 60. Every documented change was coded by type, and every correction was mapped to the eight check families of the Study Readiness Review, a rule-based consistency review of study documents, and to its eight formal rules. Intervals come from a stratified cluster bootstrap. We did not run the review on these protocols. Results. Fifty-three of the posted files were amended versions and 30 described what changed: 1,432 changes across 96 amendments. Half of all changes altered the study design and are outside any document check. Of the rest, 109 changes (7.6%, 95% CI 5.7–9.8%) corrected an internal conflict of a type the review is built to check, most often a mismatch between the schedule of assessments and the protocol text; sponsors themselves called 66 of these an error or inconsistency. They occurred in 20 of the 30 protocols (67%, 50–83%) and in 37 of the 96 amendments (39%, 30–48%). Of the 36 dated amendments that carried one, 28 were made after the first participant enrolled, a median of 9.0 months in. At least seven corrections repaired errors that an earlier amendment had introduced. Of the 109, 91 (83%) can be written as a violation of one of the review's eight formal rules, and in 69 both conflicting statements sit in the protocol itself. A blinded second coder agreed with our change-type coding on 88% of a stratified sample (Cohen's κ = 0.83), and a second full review of all 124 corrections mixed with 100 other changes agreed on 85% (κ = 0.76). The findings hold under strict and broad definitions, without the heaviest protocols, and by sponsor type. Conclusions. Internal conflicts of the types a rule-based review checks are common in real amendment histories and are usually fixed after enrollment has begun. The paper measures what kinds of conflicts reached amendment and whether the review's rules can express them, not detection, which a pre-specified protocol will measure on the same 60 protocols. Companion paper: https://doi.org/10.5281/zenodo.23698557. Coded dataset: https://doi.org/10.5281/zenodo.23698624