Peer review round 10: LivingGLP1 protocol 0.10.0 (reviewer grok-4.7)
1. Round 9 comments
G9-1. Resolved. Absolute effects now keeps three separate orders. For an outcome without a competing event the reference risk is the reported Kaplan-Meier cumulative incidence, otherwise the event rate under a constant hazard. For a cause specific hazard ratio the reference risk shown is computed by the same route as the treated risk: piecewise integration, otherwise the exponential approximation under supported constant hazards, otherwise the relative only rule. For a subdistribution hazard ratio, Competing risks sets (F_{\mathrm{ref}}) to the reported Aalen-Johansen cumulative incidence, otherwise that piecewise integration, otherwise the exponential approximation under supported constant hazards. The Absolute effects clause for the subdistribution order omits the words "under supported constant hazards," but it cites Competing risks, and the exponential route itself is allowed only when constant hazards are supported. Adopting the proposed split was the right fix. That wording gap does not put the two arms back on different baselines.
G9-2. Partly resolved. The eligibility rule matches the proposal. Piecewise integration requires both cause specific hazards from every covering reference arm, a target cumulative incidence curve alone does not qualify, each arm interval needs a variance, log hazards are pooled with the median of the predictive distribution, and failure moves to the exponential route or the relative only rule. Data items record both curves, or event free survival, with uncertainty and risk set information, and they record a target curve alone only as a cumulative incidence. The same paragraph then defines the curve based hazard as the increment (dF_k) divided by event free survival at the start of the interval and by the interval length. That is not the piecewise constant hazard the integral uses, so this route is not yet the one the eligibility rule describes. The defect is Minor 1.
G9-3. Resolved. For O20, reported hazard ratios, Kaplan-Meier contrasts, and complete status counts are not applicable inputs to the single network, and that network is primary. For an outcome with a competing event, a Kaplan-Meier contrast remains a cause specific input. The cause specific definition still places the contrast of (\log(-\log S)), with competing deaths censored, inside the cause specific parenthesis only.
2. New issues
There are no major errors. The network keys, the O20 assignment, the subdistribution order, and the event free survival integral are unchanged and still consistent. The new piecewise paragraph adds one non-executable estimator.
Minor 1. The curve formula is not a piecewise constant cause specific hazard. File: protocol/protocol.qmd, Competing risks, Piecewise integration. The increment in the target cumulative incidence equals event free survival times the target hazard only as a derivative. Dividing a finite increment by survival at the start of the interval and by the interval length treats survival as constant on that interval and understates the hazard. For constant hazards 0.2 and 0.1 per year, the cumulative incidence at 5 years is 0.518. Dividing 0.518 by 5 returns 0.104. Rebuilding the cumulative incidence from 0.104 and the companion value 0.052 returns about 0.36. The grid sentence does not repair this. It sets the grid to the union of reported time points, while also saying that an arm's hazard is constant on its own reported intervals. Counts and curve increments are not observed on another arm's time points, and the next sentence does not say whether the hazard is estimated on the native interval and copied, or recomputed on the common grid.
The formula also cannot be applied to both allowed curve pairs. Two cumulative incidence curves do not by themselves name event free survival. A target cumulative incidence paired with event free survival supplies no competing increment (dF_2), so the written division has no competing input.
Replace the sentences from "The intervals form one grid" through "by the interval length" with: "Estimate each arm's hazards on that arm's own reported intervals. The common grid is the union of those time points, fixed before pooling. On each common grid interval that lies inside an arm's reported interval, the arm contributes the hazard from that reported interval. From events and person time, each cause specific hazard is the event count divided by the shared person time at risk on the arm's interval. From curves, let (\Delta) be the length of the arm's interval and let (h_{\mathrm{sum}} = -\log(S_{\mathrm{end}}/S_{\mathrm{start}})/\Delta), where (S) is event free survival at the endpoints. When both cumulative incidence curves are used, (S(t) = 1 - F_1(t) - F_2(t)) and the hazard of event (k) is (h_{\mathrm{sum}}) times the increment in (F_k), divided by the sum of the two increments. When the target cumulative incidence is paired with event free survival, the target hazard is (h_{\mathrm{sum}}) times the increment in (F_1), divided by (S_{\mathrm{start}} - S_{\mathrm{end}}), and the competing hazard is (h_{\mathrm{sum}}) minus the target hazard. Event free survival alone does not qualify. If (S_{\mathrm{end}}) is 0 or equals (S_{\mathrm{start}}), finite log hazards are not available and this route is unavailable."
Leave the pooling, zero event, and dependence sentences as they are. With hazards 0.2 and 0.1 per year over one 5 year interval, the curve inversion returns 0.2 and 0.1.
3. Checks and registration
The authoritative tables match the reviewed version 0.9.0 snapshot except protocol/amendments.csv row A009, which logs this revision. outcomes.csv has 77 outcomes. search/strategies/concepts.yaml version 1.2.0 is identical to that snapshot. R/search_strategy.R builds the API queries from the class terms, agent terms, mechanism block, and PubMed vocabulary. The Chinese terms remain a separate key for manual CNKI and Wanfang searches. The renderer was read and was not executed. The iGlarLixi switching row remains a template and stays out of the networks.
Team names, affiliations, postal address, funding, competing interests, confirmation of Embase access, PRESS review by a human information specialist, verification of the Chinese terms by a Chinese-reading team member, and confirmation and locking of provisional thresholds by clinical advisors and patient partners remain conditions on the administrative checklist. This verdict is based on the methods. The curve to hazard step should be replaced with the inversion above before registration. With that sentence corrected, the methods are acceptable for registration once the administrative conditions are completed.
VERDICT: MINOR REVISION