Peer review round 19: LivingGLP1 protocol 0.19.0 (reviewer fable-5.1)
What I verified against the 0.18.0 snapshot: the archived copy in round-18/reviewed-version-0.18.0/ is byte identical to the 0.18.0 commit for all fifteen files. The revision changes protocol.qmd in four places (the subtitle; the round and version counts at lines 43 and 241; the Amendments pointer at lines 85 to 86; the binomial $f_a$ clause at lines 1042 to 1046), adds A018 to amendments.csv (6 comments, matching the six round 18 rows of comments.csv), adds the three round 18 verdicts to verdicts.csv, and updates the README status line. The twelve other authoritative tables, concepts.yaml (1.2.0) and R/search_strategy.R are unchanged; outcomes.csv still has 77 rows; protocol.html carries the 0.19.0 subtitle and both new passages. No model, prior, estimand or search term changed, so no numerical check was needed this round.
1. Round 18 comments
| Item | Status | Reason |
|---|---|---|
| F18-0 | Resolved | The response opens with a correction giving the right reason for the F17-4 deferral (fixed time scenario inputs belong with the F12 and F16 reference arm likelihood work), and the comments.csv row carries the action corrected. |
| F18-1 | Resolved | Lines 1042 to 1046 replace the citation of the trial wide hazard ratio rule with "known status as defined in @sec-outcomes, applied to that arm alone rather than trial wide", and define the sensitivity analysis inline: the longest follow-up at which status is known for at least 95 percent of that arm, unknown participants kept in the denominator as event free. This is the substance of my one clause fix and of G18-1; it matches the known status definition at lines 493 to 500 and mirrors the arm count sensitivity rule at lines 510 to 512, so the two analyses now use the same conventions at different scopes. |
| F18-2 | Resolved | Lines 85 to 86 adopt my wording verbatim. Because the pointer is a rule on the action field rather than a list, it also picks up F18-3, deferred in this very round (see item 1 below). |
| F18-3 | Resolved | Deferred to the pre-analysis amendment with the F12-note matter, which is what my note asked for; the row carries the action deferred to amendment. |
Nothing was declined or partly adopted, so there is no reason to judge.
2. New items in 0.19.0
No major items. The revision touches no relative effect model, estimand, prior or search term.
Minor
Stale enumeration in the response record (
round-18/response.mdline 24, and theresponsecell of G18-2 incomments.csv). The G18-2 response lists the rows the pointer now covers as "(F12-note, F14-note, F15-note, F16-note, F17-4, F17-5)", but the same round records F18-3 with the actiondeferred to amendment, so the set is seven. The protocol pointer itself needs no change, which is the advantage of the rule based wording. Replace the parenthetical in both places with "(F12-note, F14-note, F15-note, F16-note, F17-4, F17-5 and F18-3)".Scope of the horizon condition (line 1046), editorial. The sensitivity $f_a$ is now defined inside the parenthetical and the condition "and $f_a$ reaches the analyzed horizon" follows it. I read the condition as governing $f_a$ in either analysis, since the sentence defines $f_a$ and the sensitivity analysis only redefines it, and lines 1069 to 1070 forbid extrapolation in any case. If the owner wants it beyond doubt, replace "and $f_a$ reaches the analyzed horizon" with "and, in either analysis, $f_a$ reaches the analyzed horizon".
Not new, for the record. The Sensitivity analyses section (lines 977 to 990) is scoped to "every primary network" and does not list the reference arm 95 percent analysis, which is defined only inline at lines 1044 to 1046. The omission predates 0.19.0 (0.18.0 already invoked that sensitivity analysis by reference) and the inline definition is executable, so this does not count against the revision. If a single index of sensitivity analyses is wanted, append "; for the reference risk of hazard based outcomes, arm level $f_a$ at 95 percent known status (@sec-absolute)" to that list.
3. Judgment on revisions made for the other reviewers
G18-1 and G18-2 are the same two items as F18-1 and F18-2, and the adopted text resolves both. For G18-1 the owner merged Grok's longer wording into one parenthetical rather than three sentences; the content is the same (arm level definition, no trial wide requirement, 95 percent sensitivity with unknowns event free in the denominator). The sensitivity $f_a$ is at least the primary $f_a$ for every arm, since complete status at a time implies at least 95 percent known status at that time, so no arm usable in the primary analysis is lost in the sensitivity analysis, and the complementary log log offset $\log(f_a / t)$ with $f_a \geq t$ keeps the constant hazard construction unchanged. For G18-2 the adopted rule based wording is preferable to Grok's enumeration of action values, as item 1 shows. GPT-6 Astra raised no comments.
Conclusion
All four of my round 18 items are resolved, three by adoption and one by the deferral I asked for. The one new defect is a stale list in the response record that the protocol's own rule already supersedes; the other two items are one clause editorial matters with no effect on any estimand, model or registration level decision. By the standard of rounds 12 to 18 they are amendment level, to be settled in the next revision or in the pre-analysis amendment. The administrative items (team names, affiliations, funding, competing interests, Embase access, PRESS review, verification of the Chinese terms, locking of provisional thresholds) remain on the checklist as pre-registration conditions and do not bear on this verdict. I would approve the methods for registration once those conditions are completed.
VERDICT: ACCEPT