Peer review round 7: LivingGLP1 protocol 0.7.0 (reviewer grok-4.7)
Round 6 comments
G6-1. The 95 percent sensitivity chooses t. Resolved. In Event outcomes and harms, the sensitivity sets t to the longest follow-up at which status is known for at least 95 percent of randomized participants in every arm. Unknown participants stay in the denominator and count as event free, and the event counts at that t enter the same binomial complementary log log likelihood as the primary count model. That likelihood carries the offset log t in years. Known status is the definition in the paragraph above, including a competing death as known and event free, and loss or administrative censoring before t as unknown. The illustrative numbers are unchanged and are now labeled as an illustration under baseline missingness independent of the outcome: equal 5 percent loss moves a hazard ratio of 2.0 (risks 0.50 and 0.75) to about 1.93, and 5 percent loss from the control arm only moves 1.20 (risks 0.010 and 0.012) to about 1.26. Those calculations check. The statement that equal missing fractions have no general bound or direction is right, because missingness may depend on the outcome.
G6-2. Conclusions and time to event counts. Partly resolved. Counts for a time to event outcome are no longer described as a sensitivity analysis of the cause specific estimate. Conclusions about a binary harm now follow one order: the primary hazard type network if it meets the connectivity rule, otherwise the other hazard type network labeled with its hazard type, otherwise the recorded harms analysis, labeled as such. For binary harms the subdistribution network is primary. The reason for keeping the cause specific network as the second step is acceptable. A connected set of reported Cox hazard ratios is a coherent relative effect on the cause specific hazard, the conclusion has to carry that hazard type, and Q is defined as having neither an incidence rate nor a hazard interpretation, so Q is the weaker fallback. The unlabeled pair of readings from round 6 is closed for binary harms. Two limits are new issues below. The order is stated only for a binary harm. The count sentence is not limited to outcomes that have a competing event, and it now collides with the not applicable rule.
G6-3. Cause specific conversion. Resolved. Competing risks now states the requested constant hazard conversion. λ1 and λ2 are constant cause specific rates for the reference node over the analyzed horizon, estimated as events per person year from the covering reference arms and pooled on the log scale with the reference risk model. Reference cumulative incidence is λ1/(λ1+λ2) times (1 minus exp(−(λ1+λ2)t)). The treated cumulative incidence uses that expression after λ1 is replaced by HR times λ1 everywhere it appears, and λ2 is replaced by its hazard ratio everywhere only when that ratio is reported. Otherwise λ2 is unchanged, and the text states the assumption of no effect on the competing event. 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 a required rate or F_ref is unavailable, or the constant hazard approximation cannot be supported, only relative effects are reported under the relative only rule in Absolute effects. The exponential label is an accurate addition. The time varying branch added beside it is judged below.
New issues
Major 1. Event outcomes and harms, opening paragraph, against outcomes.csv row O20 and Competing risks. The opening paragraph says counts for a time to event outcome enter only its subdistribution network and are never pooled with cause specific estimates. Competing risks, in the same section, says hazard type not applicable is only for outcomes with no competing event, such as O20, and that every other outcome takes the hazard type of its estimator. Row O20 is all cause mortality, type time to event, scale log hazard: death from any cause. There is no competing death to code as event free, so a complete status count and an all cause hazard ratio are the same estimand and belong in one network of hazard type not applicable. Read literally, the count sentence either builds a subdistribution network the hazard type rule forbids, or drops mortality counts because that network cannot exist. Replace the two count sentences with: "Time to event outcomes use reported or validly reconstructed log hazard ratios over the longest randomized follow-up. For an outcome with a competing event, counts enter only its subdistribution network under the complete status rule below and are never pooled with cause specific estimates. For an outcome with no competing event, such as all cause mortality (O20), hazard ratios and complete status counts are hazard type not applicable and enter the same network."
Minor 1. Competing risks, time varying sentence. When time varying baseline hazards are available, the protocol replaces the exponential formula with integration of overall survival times the cause specific hazard. It does not say how the hazard ratio enters that integral. Integrating reference overall survival times HR times λ1, without also putting HR times λ1 inside overall survival, is not the treated cumulative incidence. Replace that sentence with: "When time varying baseline hazards are available from the covering reference arms, cumulative incidence from time 0 to the analyzed horizon is the integral of all cause survival, in which any death is an event, times the cause specific hazard of the outcome. For the reference arm use λ1(u) and λ2(u). For the treated arm replace λ1(u) with HR times λ1(u) inside the integral and inside all cause survival, and replace λ2(u) in those same places only when a cause specific hazard ratio for the competing event is reported; otherwise leave λ2(u) unchanged and state that assumption."
Minor 2. Competing risks, Kaplan-Meier contrast. A contrast of one minus Kaplan-Meier with competing deaths censored is now a cause specific input, and it is correctly barred as an absolute or reference risk. The only contrast likelihood in Hazard ratio analysis is for a reported cumulative incidence. One minus Kaplan-Meier is not a cumulative incidence, so this input has no stated estimator. Replace the classification clause with: "A contrast between arms of one minus Kaplan-Meier, with competing deaths censored, enters the cause specific network as the contrast of log(−log(S)) at a common t, where S is the Kaplan-Meier estimate. That contrast is the difference in log cumulative cause specific hazards. It uses a delta method standard error and requires 0 < 1 − S < 1. The contribution network uses this same contrast. No continuity correction is applied."
Minor 3. Conclusions, and the 95 percent sensitivity. The conclusion order covers a binary harm only. A time to event outcome with a competing event can have a connected subdistribution count network and a disconnected cause specific network, and no sentence says which one supplies the conclusion. The 95 percent sensitivity codes unknowns as event free and adds those counts to the hazard ratio analysis, without a hazard type. Because competing deaths are also coded event free, those counts are subdistribution inputs, and for O20 they are not applicable. Add: "For a time to event outcome with a competing event, conclusions use the cause specific network when it meets the connectivity rule; otherwise the subdistribution network, labeled as subdistribution. The recorded harms analysis is used only for binary harms. Counts from the 95 percent sensitivity are a sensitivity analysis of the subdistribution network, or of the not applicable network when the outcome has no competing event, and they are not added to a cause specific network."
Minor 4. Absolute effects, hazard based outcomes. The same bullet tells the analyst to use Aalen-Johansen cumulative incidence for a competing event when it is reported, and otherwise a single event rate under a constant hazard. Competing risks then says F_ref must be the Aalen-Johansen incidence, and a cause specific conversion uses two rates, not one. The closing cross reference says competing risks follow the outcomes section. The single rate clause still reads as the fallback for every hazard based outcome. A cause specific conversion also builds its own reference cumulative incidence, which can differ from the Aalen-Johansen figure the bullet calls the reference risk. Replace the fallback and the cross reference with: "For an outcome with a competing event, do not fill a missing baseline with a single event rate. Subdistribution conversions use the Aalen-Johansen cumulative incidence as F_ref and as the reference risk shown with the estimate. Cause specific conversions use the two rate exponential cumulative incidence, or the time varying integral in Competing risks, and that value is the reference risk shown with the estimate. If the required cumulative incidence or rates are unavailable, apply the relative only rule."
Checks that held
Authoritative tables are unchanged from the reviewed 0.6.0 snapshot except amendments.csv row A006 (2026-10-12, version 0.7.0, before results). outcomes.csv still has 77 outcomes. The version string in the outcomes section is 0.7.0. search/strategies/concepts.yaml is still version 1.2.0 and matches the round 6 copy, including the Chinese terms. R/search_strategy.R still renders PubMed, Europe PMC with patents excluded, OpenAlex, and ClinicalTrials.gov. 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. The recorded quantity Q matches the complementary log log model with offset log t: the ratio of −log(1 − r)/t. The certainty range 0.8 to 1.25 on Q is the same interval as 1/1.25 to 1.25, and the imprecision rule names both the relative only rule and the recorded harms range. O20 is a sound example of an outcome with no competing event. The subdistribution formula 1 − (1 − F_ref)^HR and the constant hazard replacement of λ1 everywhere are the right expressions.
Pre-registration conditions
These 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 a logged harvest of all four API sources with the registered concepts version, with every count reconciled, before the baseline search.
Judgment
G6-1 and G6-3 are resolved. G6-2 is resolved for binary harm conclusions by an acceptable alternative, and the adopted count sentence is right only when a competing event exists. Applied to every time to event outcome, that sentence contradicts the not applicable rule on all cause mortality. The time varying integral, the Kaplan-Meier likelihood, the time to event conclusion order, and the single rate baseline are smaller gaps. With Major 1 corrected, the minor sentences tightened, and the administrative conditions completed, the methods would be acceptable for registration.
VERDICT: MAJOR REVISION