Peer review round 6: LivingGLP1 protocol 0.6.0 (reviewer grok-4.7)
Round 5 comments
G5-1. Competing risks as a network. Resolved. Hazard type (cause specific, subdistribution, or not applicable) is recorded for every hazard scale input and is part of the network key in Networks and nodes and of the connectivity rule. Cause specific and subdistribution estimates never share a network. A subdistribution hazard ratio converts as 1 minus (1 minus F_ref) to the power of the hazard ratio. A cause specific conversion uses the reference cause specific hazard and the competing hazard. The rule applies to every hazard scale input. Removing the 1 percent mortality exemption is the right choice: complete status counts and cumulative incidence contrasts are subdistribution inputs, so no exemption cutoff is needed. The protocol states that the competing hazard is left unchanged when its own hazard ratio is unreported.
G5-2. Time and known status. Partly resolved. The primary count rule defines t and known status and requires known status for every randomized participant. That primary rule is the right one. Known status means the event occurred by t or the participant was observed event free through t. Administrative end counts as observed. Loss before t is unknown. Both bias examples check. Equal loss of 5 percent from both arms at the start changes risks 0.50 and 0.75 (hazard ratio 2.0) to about 1.93. Loss of 5 percent from the control arm only changes risks 0.010 and 0.012 (hazard ratio 1.20) to about 1.26. Primary t is a common time at which every arm is fully observed, so the log time offset cancels and the contrast is a log hazard ratio under proportional hazards. The sensitivity sentence still does not define t. It admits trials with known status for at least 95 percent of randomized participants in every arm and codes unknowns as event free. The only definition of t is the longest follow-up at which known status holds for every participant, so it cannot select the time used when up to 5 percent remain unknown.
G5-3. Duplicate O22C label. Resolved. The composite sentence matches the proposed text. O22, O22B, O22C, O22D, and O22E are never pooled. A composite that includes cardiovascular death is O22D even when it also includes albuminuria or a smaller eGFR decline. The O22C row in outcomes.csv already matched: kidney failure, sustained doubling of serum creatinine, or kidney death.
G5-4. Contribution weights for cumulative incidence. Resolved. For hazard scale estimates, the contribution network uses the input that entered the primary model: the reported log hazard ratio; the cumulative incidence contrast with its delta method variance; or a complementary log log estimate from the arm counts, adding 0.5 to every cell of a trial with a zero event or all event arm. The correction is for contributions only. Shared arm covariance is preserved. No 0.5 correction is applied when cumulative incidence is 0 or 1, which is correct, because those values never enter. The rule requires 0 < p < 1 with finite uncertainty.
G5-5. Compatible follow-up bounds. Resolved. The restriction to follow-up of 20 to 104 weeks is restored in the frequentist section, listed among the sensitivity analyses, and repeated in the confidence example.
G5-6. Recorded harms absolute differences. Resolved by an acceptable alternative. The response declines a pooled absolute difference at the median observation period. The reason is sound. Observation processes differ across trials, so that pooled difference is not defined, and constructing it would assume the constant hazard that the recorded analysis exists to avoid. The coefficient is the log relative rate of recorded first events. Magnitude dependent certainty uses the range 1/1.25 to 1.25 on that rate ratio, and the rating states the scale. Each trial's recorded proportions and within trial risk difference are displayed with the observation period and completeness. No importance category is assigned from this analysis. One limitation should stay visible and does not require a further change: for a harm with a large absolute threshold, such as nausea at 50 per 1,000, a rate ratio inside 1/1.25 to 1.25 can still be a large absolute difference. The rating is a statement about the rate ratio.
New issues
No major new defect. Three minor items are in protocol/protocol.qmd.
Minor 1. Event outcomes and harms, Hazard ratio analysis. The 95 percent sensitivity does not choose t. Replace the sensitivity sentence with: "In that sensitivity analysis, t is the longest follow-up at which at least 95 percent of randomized participants in every arm have known status, with known status defined as in the primary rule. Unknown participants remain in the denominator and count as event free. Event counts by that t enter the binomial complementary log log likelihood with offset log t in years."
Minor 2. Event outcomes and harms, conclusions and the time to event sentence. Connectivity is judged per hazard type, but the conclusion sentence still refers to one hazard ratio analysis. A connected cause specific network and a disconnected subdistribution network then have two readings. The opening time to event sentence still calls a count model a sensitivity analysis of that outcome, although those counts are subdistribution inputs and cannot join the cause specific network. Replace both with: "Conclusions about a binary harm use the subdistribution network when that network meets the connectivity rule. The cause specific network is reported separately and does not supply the conclusion. When the subdistribution network does not meet the connectivity rule, conclusions use the recorded harms analysis, labeled as such. Counts for a time to event outcome enter only the subdistribution network and are not pooled with the cause specific estimate."
Minor 3. Competing risks, cause specific conversion. The treated risk is described as multiplying λ1 by the hazard ratio. That sentence can be read as scaling only the leading λ1, or as multiplying the finished cumulative incidence by the hazard ratio. Constant rates, pooling across reference arms, and the meaning of F_ref are unstated. Replace the conversion sentences with: "λ1 and λ2 are constant cause specific rates for the reference node over the analyzed horizon. When several reference arms cover that horizon, pool each log rate with the same random effects model used for the reference risk. The reference cumulative incidence is λ1/(λ1+λ2) times (1 minus exp(−(λ1+λ2)t)). The treated cumulative incidence uses that expression after replacing λ1 with the hazard ratio times λ1 everywhere it appears. When a hazard ratio for the competing event is reported, also replace λ2 with that hazard ratio times λ2 everywhere; otherwise leave λ2 unchanged and state that assumption. F_ref in the subdistribution formula is the reference node's Aalen-Johansen cumulative incidence at the analyzed horizon from the reference risk procedure. If either cause specific rate, or F_ref, is unavailable, report relative effects only and use the relative range rule in Absolute effects."
Checks that held
The authoritative tables are unchanged from the reviewed 0.5.0 snapshot except amendments.csv, which adds A005 (2026-10-12, version 0.6.0, before results). The outcomes section now says 77 outcomes at version 0.6.0. The iGlarLixi switching row remains a template and stays out of networks until the strategy identifier, permitted pens, switching threshold, and titration algorithm are recorded. concepts.yaml is version 1.2.0. R/search_strategy.R still renders PubMed, Europe PMC (excluding patents), OpenAlex, and ClinicalTrials.gov. Chinese terms are unchanged. The harvest dated 2026-10-12 used concepts version 1.0.0 on three sources. The checklist already requires a logged four-source harvest of the registered concepts version, with reconciled counts, before the baseline search. Where competing events exist, the cross reference to the outcomes section controls the Kaplan-Meier option in the hazard based absolute effects bullet. When conversion is unavailable, the Absolute effects override keeps the relative range rule executable, including for recorded harms.
Pre-registration conditions
These items remain for the review owner and do not by themselves withhold acceptance: team names, affiliations, and postal address; funding and competing interests; PRESS 2015 review by a human information specialist; confirmation of Embase access, with omission stated as a limitation if access is absent; verification of the Chinese terms by a Chinese-reading team member; confirmation and locking of provisional thresholds by clinical advisors and patient partners before any outcome data are extracted; an OpenAlex API key; and the logged four-source harvest above.
Judgment
Five of the six round 5 comments are resolved. The reason for the recorded harms alternative is sound. G5-2 is resolved for the primary analysis and open for the 95 percent sensitivity. The three remaining items are single sentences. With those sentences and the administrative conditions completed, the methods are acceptable for registration.
VERDICT: MINOR REVISION