Peer review round 4: LivingGLP1 protocol 0.4.0 (reviewer grok-4.7)

Round 3 comments

G3-1. Resolved. PR04 and PR08 apply to O24B;O24C, with Normal(0, 2.5^2) on the standardized mean difference and Half-normal(0, 0.5^2) on tau. Effect measures names O24B and O24C, each in its own network. reference_sd.csv keeps s_y = 1 for O24B;O24C.

G3-2. Resolved. OTHER_ACTIVE_<name> is is_template = yes. Dose templates wait for a numeric maintenance dose. The comparator template waits until an identified molecule or procedure and its class replace <name>. Doses of that comparator are not separate nodes.

G3-3. Resolved. Per procedure outcomes (O36A, O36B) and the per pregnancy outcome (O120) have no week window. W00 omits O36A and O36B. After a window applies only where the outcome includes that population, every other outcome and population pair has one primary window and no numeric overlap.

G3-4. Resolved. Recurrent events use a normal likelihood on reported log rate ratios. A count model requires a logged amendment. O04B and PR14 match.

G3-5. Resolved. Absolute effects and the CINeMA imprecision bullet set the range of little or no difference for O04B at a rate ratio from 1/1.25 to 1.25, under prior PR14. Leaving rare_outcome = no on that row is correct: PR14 and that sentence carry the range.

G3-6. Resolved. absolute_horizon_weeks is the uncapped horizon. The analyzed horizon is the smaller of that value and the median follow-up of the network's reference arms. The reference risk uses only arms whose follow-up covers that horizon, by Kaplan-Meier or cumulative incidence when available and otherwise by a constant hazard inside the arm's own follow-up. Stored 156 week values match an uncapped field. The single arm edge is a new minor item.

G3-7. Partly resolved. Replacing one extra outcome id with a rule that pools only identical component sets and thresholds is sound. O22, O22B, O22C (kidney failure, sustained doubling of serum creatinine, or kidney death), O22D, and O22E are separate rows. The network key still omits the definition stratum, and a composite that includes both albuminuria and cardiovascular death matches O22D and O22E. Those gaps are new items.

G3-8. Resolved. O10 uses 40 to 104 weeks (target 68). O10B, used in P12 and P15, uses 12 to 52 weeks (target 24). W06 and W09 match those populations.

New issues

Major

Hazard scale contribution matrices do not run on the primary evidence (protocol.qmd, Confidence in cumulative evidence; Frequentist network meta-analysis). For hazard scale estimates from counts, contributions come from "a frequentist inverse variance network of each trial's log hazard ratio derived from its counts," so that "the same trials and network are represented." The derivation is unspecified. The rule of no continuity correction, otherwise Mantel-Haenszel, belongs to the odds ratio analysis, which that section already calls a restricted sensitivity for these harms. A zero cell makes an inverse variance log hazard ratio undefined, so sparse trials included in the binomial complementary log log model leave the matrix. Within study bias and indirectness are weighted by that matrix. The sentence covers only count derived estimates, while the Bayesian model combines arm level counts and contrast level log hazard ratios. The opening CINeMA sentence still rates Bayesian and frequentist estimates of the same evidence and reports the lower rating. The primary Bayesian model itself can be run.

Replacement: "Where a Bayesian estimate and a frequentist estimate use the same trials, scale, and likelihood, both are rated and the lower rating is reported. For a hazard scale estimate, build the contribution network from the same trial level log hazard ratios that entered the primary model: the reported log hazard ratio when that was used, and a complementary log log estimate from the arm counts otherwise. When an arm count is zero, add 0.5 to every cell of that trial before deriving the log hazard ratio, and keep every trial that is in the primary estimate. Use this network only for contributions. Hazard scale harm estimates have a Bayesian primary rating. The frequentist odds ratio analysis is a restricted sensitivity and is rated separately."

Minor

1. Kidney composite precedence (outcomes.csv O22D, O22E, O50; protocol.qmd, Composite definitions). O22E sends a composite without albuminuria to O22 through O22C. Those rows exclude cardiovascular death, so a composite that includes cardiovascular death and omits albuminuria is misassigned. A composite with albuminuria, or an eGFR decline below 40 percent, plus cardiovascular death matches both O22D and O22E. The prose calls O22C "doubling of serum creatinine, about a 57 percent decline," while the row is kidney failure, sustained doubling, or kidney death.

O22D mapping_rules: "O22D is any kidney composite that includes cardiovascular death, including when it also includes albuminuria or a smaller eGFR decline. Trials are pooled only within a definition stratum that has the same component set and the same eGFR or creatinine threshold. A composite without cardiovascular death is O22, O22B, O22C, O50, or O22E."

O22E mapping_rules: "O22E is a kidney composite that includes new onset persistent macroalbuminuria or an eGFR decline below 40 percent and does not include cardiovascular death. Trials are pooled only when the component set and thresholds are identical. When the same trial also reports a composite without albuminuria, map that result by its components: O22, O22B, O22C, or O50 when cardiovascular death is absent, and O22D when cardiovascular death is present."

O50 note: "Same components as O22. A composite that includes cardiovascular death is O22D. A doubling of serum creatinine composite without cardiovascular death is O22C."

In Composite definitions, use "O22C (kidney failure, sustained doubling of serum creatinine, or kidney death)" and add "A composite that includes cardiovascular death is O22D even when it also includes albuminuria or a smaller eGFR decline."

2. Network key (protocol.qmd, Networks and nodes; Connectivity rule). Replacement: "A network is defined by population, population stratum, outcome, definition stratum, window, and estimand. The definition stratum is the component set and threshold when an outcome identifier allows more than one." Use the same list in the connectivity sentence. Population stratum is the background stratum. Definition stratum is the composite split.

3. Kaplan-Meier input and harm denominators (protocol.qmd, Event outcomes and harms). A reported hazard ratio can enter the normal log hazard likelihood. A Kaplan-Meier estimate is a cumulative incidence, and the paragraph names neither its likelihood nor its variance. Counts use the safety or randomized population, while primary networks are treatment policy. Replacement: "When a hazard ratio and its variance are reported, they enter the normal likelihood on the log hazard ratio. When only a Kaplan-Meier cumulative incidence at a common follow-up time is reported, that incidence and its variance enter the contrast likelihood. Otherwise the binomial complementary log log model uses the number of participants with at least one event over the number randomized. The safety population is used only when the randomized denominator is not reported, and those trials are flagged."

4. Switching node (nodes.csv IGLARLIXI_SWITCH_SC_TIT_QD; protocol.qmd, Networks and nodes). The row is is_template = no and records no threshold or algorithm, so different strategies share one id. Change "one node per permitted pens" to "one node per set of permitted pens, switching threshold, and titration algorithm." Note and component_ratio: "Record the labeled LixiLan switching threshold and titration algorithm in component_ratio. Any other permitted pens, switching threshold, or titration algorithm is a new node created at extraction. Combine this node with IGLARLIXI2_SC_TIT_QD and IGLARLIXI3_SC_TIT_QD only in a sensitivity analysis."

5. Sensitivity list (protocol.qmd, Sensitivity analyses). Add combining the iGlarLixi pen and switching nodes, and excluding trials with more than 20 percent incomplete follow-up or an arm difference above 10 percentage points. Both rules already appear under Networks and nodes and Event outcomes and harms. Those cutoffs are methods choices, separate from the outcome thresholds the advisors will lock.

6. One covering reference arm (protocol.qmd, Absolute effects). With two reference arms of unequal follow-up, the median lies between them and only the longer arm covers the analyzed horizon. A random effects model of one arm is undefined. Replacement: "When only one reference arm covers the analyzed horizon, the reference risk is that arm's Kaplan-Meier or constant hazard risk, and no between arm heterogeneity is estimated."

Checks that held

There are 77 outcomes. turner_map.csv covers every log odds, log hazard, complementary log log, and log rate outcome, including O22E. Every mean difference and standardized mean difference has a reference standard deviation. Nonrandomized harm flags match harms_nrs.csv. Node and agent classes resolve. The 43 agents and their terms match concepts.yaml 1.2.0. That file and R/search_strategy.R are unchanged and still render the four sources. HRS9531 and HRS-7535 stay separate. W33 (4 to 12 weeks, target 8) does not overlap W22 or W23.

Pre-registration conditions

Complete these before registration. Their absence alone does not withhold acceptance of the methods. Team names, affiliations, and postal address; funding and competing interests; PRESS 2015 by a human information specialist; confirmation of Embase access, with omission stated as a limitation if access is absent; verification of Chinese terms by a Chinese-reading team member; locking of provisional thresholds by clinical advisors and patient partners before any outcome data are extracted; an OpenAlex API key; a logged four-source harvest of the registered concepts version, with reconciled counts, before the baseline search. The 2026-10-12 run is a pipeline test of concepts 1.0.0 on three sources.

Judgment

Seven of the eight round 3 comments are resolved. G3-7 is partly resolved: the identical component rule is the right rule and still needs the network key and a precedence sentence. The contribution rule would change CINeMA weights when a harm trial has a zero cell or reports a hazard ratio beside counts. The repair is local text. With these fixes and the administrative conditions completed, the methods are acceptable for registration.

VERDICT: MINOR REVISION