Peer review round 11: LivingGLP1 protocol 0.11.0 (reviewer fable-5.1)

What I verified against the 0.10.0 commit: the revision changes only protocol.qmd (Data items, Competing risks and Absolute effects; 48 insertions, 27 deletions), amendments.csv (A010), the review records and the README. The twelve other authoritative tables, concepts.yaml (1.2.0) and R/search_strategy.R are byte identical to the reviewed 0.10.0 snapshot, so the invariants of rounds 5 to 10 hold. comments.csv carries the eight round 10 rows that A010 states, verdicts.csv the three round 10 verdicts, protocol.html the 0.11.0 subtitle and outcomes.csv the 77 outcomes the text counts. I recomputed the adopted inversions: for constant hazards 0.2 and 0.1 (and 0.4 and 0.2) per year, the Aalen-Johansen form returns both hazards exactly on grids of 0.25, 0.5, 1 and 5 years, the Kaplan-Meier form likewise, and the response's check (incidences 0.3008 and 0.1504, the old rule reintegrating to 0.2421) reproduces.

1. Round 10 comments

The one reason to judge: the dependence assumption I proposed in round 9 (F9-3) was dropped in favor of C10-2's stricter eligibility rule (lines 589 to 595). The reason given, that such curves are no longer used, is correct and the rule is right. An all cause mortality curve substituted for the competing first event curve breaks S = 1 − F_1 − F_2: with target hazard 0.2 and death hazard 0.1 per year the inversion returns 0.201 and 0.111 over one year, and the error grows with the target event rate. The practical cost is that curve derived inputs qualify only from trials that publish both first event curves, which is rare; the exponential route and the relative only rule cover the rest, and that is acceptable.

2. New items in 0.11.0

No major items. Everything below is a one clause clarification of text that is otherwise exact and executable; none changes a model, an estimand or a relative estimate.

Minor

1. Shared uncertainty across subdivided grid intervals (lines 599 to 601). C10-1's proposal carried the original interval's hazard "and shared uncertainty" onto the grid; the adopted sentence carries the hazard only. Read literally, one arm's single estimate enters adjacent per interval pools as independent inputs, and the integral then treats positively correlated terms as independent, which narrows the reference risk interval. Line 586 ("valid joint uncertainty") already constrains the implementation, so this is a wording gap. Replace "and are carried onto every grid interval that the reported interval contains" with "and are carried, with their variance and as one draw, onto every grid interval that the reported interval contains".

2. Count reconstruction, the Kaplan-Meier condition, the final partial interval and the number at risk rule (lines 595 to 614). My own replacement paragraph left d_k unstated, so (a) is a gap in the adopted text rather than an error of the revision. (a) With n_s the published number at risk at t_s: from Aalen-Johansen curves d_k = n_s ΔF_k/S(t_s); from a Kaplan-Meier curve d_k = n_s{1 − S_k(t_e)/S_k(t_s)} (within interval censoring ignored, for the variance only); one numbers at risk row serves both Kaplan-Meier curves because censoring at the other first event gives the same risk set as the Aalen-Johansen pair. (b) Line 609: the Kaplan-Meier form requires 0 < S_k(t_e) < S_k(t_s). (c) The analyzed horizon is the median follow-up of the reference arms (lines 960 to 961), which will rarely coincide with the 6 or 12 month ticks of a numbers at risk table, and such tables routinely end before the curve does; under lines 597 to 599 the stretch from the last published time to the horizon is not a reported interval, so in the common case a covering arm cannot contribute the horizon and line 586 makes the route unavailable. Append to line 599: "and, when the curve extends to the analyzed horizon, the interval from the last published time to that horizon, with the number at risk at its start". (d) Lines 612 to 613 read as if a counts arm also needs a published number at risk; write "or, for curve inputs, without a published number at risk".

3. Scenario wording (lines 951 to 953). "From its own hazards by the route in use" does not fit the subdistribution order when F_ref is a reported Aalen-Johansen value, which involves no hazards. Replace with "(for outcomes with competing events, each covering arm's own cumulative incidence by the route in use)".

4. Front matter date (line 4). date: 2026-10-12 while A010 and the round 10 verdicts are dated 2026-10-13; the rendered subtitle is correct. Set the date to 2026-10-13.

3. Judgment on revisions made for the other reviewers

C10-1 is carried in the adopted formulas and the condition 0 < S(t_e) < S(t_s); its shared uncertainty clause is the one part not carried (item 1). C10-2 is carried in substance, first event curves only and mortality curves that include deaths after a nonfatal target event excluded, and the rule is correct (section 1). C10-3 is carried at lines 654 to 657 and is consistent with the subdistribution order at lines 630 to 639 and 967 to 975. G10-1's event free survival form coincides algebraically with the adopted one, since ΔF_1 + ΔF_2 = S(t_s) − S(t_e); both allowed curve pairs are handled through F_2 = 1 − S − F_1, and estimation on native intervals with carry over to the grid is stated. The response overstates nothing.

Conclusion

All four of my round 10 items are resolved in my wording, the formulas verify exact, the other reviewers' items are correctly carried, and the authoritative tables are unchanged. The four remaining items are clarifications that make the piecewise route fully executable as written and do not alter a model, an estimand or a relative estimate. The administrative items remain on the checklist as pre-registration conditions. I would approve the methods for registration once those conditions are completed.

VERDICT: ACCEPT