Peer review round 14: LivingGLP1 protocol 0.14.0 (reviewer grok-4.7)

1. Round 13 comments

G13-M1. Resolved. The missing mean is removed by a restriction. Piecewise integration now requires identical reported interval boundaries across all covering arms through the analyzed horizon; otherwise that route is unavailable (protocol/protocol.qmd, Competing risks). Every coordinate maps to one common interval, so a reported interval cannot cover several grid intervals and cannot force a constant reference hazard. The displayed reference cumulative incidence is the median of the joint draws of the integrated incidence, as proposed.

Declining the length-weighted observation equation is sound. Under identical partitions that equation has nothing to connect. With two to five covering arms, a free between-arm covariance over intervals and events is weakly identified, and one proportional deviation per event is a usable heterogeneity structure. The cost is that published partitions often differ, so this route will often be unavailable. The exponential approximation, when constant hazards are supported, and otherwise relative effects only, remain. Reported Aalen-Johansen cumulative incidence remains the first subdistribution source. Keeping the length-weighted equation as the amendment if unequal partitions must be fitted is the right reservation.

G13-1. Resolved. Scenario values remain the empirical 25th and 75th percentiles of the covering arms' own cumulative incidences. Their uncertainty comes from drawing each arm from N(y_a, V_a) independently, integrating that arm, and recomputing the percentiles. Absolute effects states the same percentiles for competing-event outcomes. Scenario uncertainty is the distribution of those recomputed percentiles.

2. New issues

No major errors.

Minor 1. The one-arm rule contradicts the draw that follows it, and that draw is on the log-hazard scale. File: protocol/protocol.qmd, Competing risks, Piecewise integration, Pooling.

With one covering arm, τ_k is not estimated and the arm's within-arm uncertainty is propagated. The next sentence, for every posterior draw, draws a new reference arm as μ + Z u_new. μ is the reference log hazard. If τ_k is not estimated, u_new is unidentified. Drawing it from the Half-normal(0, 1²) prior would add between-arm uncertainty that the one-arm rule excludes. Absolute effects already says that one covering arm contributes its own risk and that no between-arm heterogeneity is estimated.

The same sentence calls μ + Z u_new the hazards and says they are integrated. The integrals for S_a and F_1a use hazards, and the exponential formula uses rates λ. Exponentiation is not stated. Scenario redraws of y_a have the same gap.

Replace the sentences from "With one covering arm" through "comes from the same draws." with: "With one covering arm, τ_k is not estimated and u_new is not drawn. Each draw of the reference log hazards is a draw from N(y_a, V_a). With two or more covering arms, each draw of the reference log hazards is μ + Z u_new, where u_new,k ~ N(0, τ_k²) independently for the two events at that draw's τ_k. The hazards are the exponentials of those log hazards. They are integrated jointly with the formulas for S_a and F_1a. The displayed reference cumulative incidence is the median of these integrated draws, and its uncertainty comes from the same draws. Scenario draws of y_a are exponentiated before the same integration."

Minor 2. For counts, y_a and V_a are never assigned. File: protocol/protocol.qmd, Competing risks, Piecewise integration, Within arm uncertainty and Pooling.

Curve inversion sets V_a = J Σ J^T for the arm's vector of log hazards. The count approximation gives variance about 1/d_k and zero covariance, and the likelihood then uses y_a and V_a for every arm. Those symbols are not tied to the count estimates.

After the zero-count sentence, add: "y_a is the vector of estimated log hazards, one coordinate per event and common interval, in that order. For counts, V_a is diagonal with those variances, approximately 1/d_k."

Minor 3. A single shared interval skips the constant-hazard gate. File: protocol/protocol.qmd, Competing risks, Piecewise integration and Exponential approximation.

The cause-specific order tries piecewise integration first. One shared interval from 0 to the analyzed horizon meets the identical-boundary rule, so the fit never reaches the exponential bullet. That bullet is the same model, and it is the only place that requires support for constant hazards and a label. Total event counts on that interval are the ordinary input this gate was written for. The exponential bullet also calls d_k the arm's total count, which can be read as events after the analyzed horizon.

Replace the opening of the exponential bullet with: "If the identical boundaries are the single interval from 0 to the analyzed horizon, the piecewise fit is the exponential approximation: it is used only when constant hazards are supported, and it is labeled as such. The separate exponential route is used only when the piecewise route is unavailable and every covering arm still supplies both event counts and person time on that single interval. Here d_k is the count of event k from time 0 to the analyzed horizon, λ_k is the exponential of the corresponding log hazard, and a zero count for either event makes the route unavailable."

3. Checks and registration

Against the version 0.13.0 snapshot, the authoritative tables match except priors.csv (PR15, PR16, and PR17 added; PR01 to PR14 unchanged) and amendments.csv (A013 only: 2026-10-13, version 0.14.0, eight comments, before results). outcomes.csv has 77 outcomes. The protocol text changes are the version strings, the competing-risks replacement, and the PR17 citation in Absolute effects. The subtitle, the date, the status sentence, and the outcomes section all say version 0.14.0, dated 2026-10-13, after thirteen rounds. search/strategies/concepts.yaml remains version 1.2.0. R/search_strategy.R builds the PubMed, Europe PMC (excluding patents), OpenAlex, and ClinicalTrials.gov queries from the class terms, agent terms, mechanism block, and PubMed vocabulary. That renderer was read and was not executed. The Chinese term list remains the manual CNKI and Wanfang list. The iGlarLixi switching row remains a template (is_template yes) and stays out of every network until a strategy is recorded. The inversion formulas are unchanged. Under constant hazards 0.04 and 0.02 per year for one year they still return those hazards. The Quarto file was read and was not rendered.

PR17 supplies the heterogeneity prior deferred in round 12 for the reference-arm meta-analysis of risks other than competing-risk hazards. Competing-risk estimation uses PR15 and PR16, and the hazard bullet points that estimation back to Competing risks. The scale of the other reference-risk meta-analysis, and the prior on its mean, are unchanged from version 0.13.0.

Team names, affiliations, 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 administrative pre-registration conditions. They do not determine this verdict. The estimand, the identical-boundary likelihood, the displayed median, and the scenario redraws are specified. The three sentences above still need to be fixed before registration.

VERDICT: MINOR REVISION