Peer review round 4: LivingGLP1 protocol 0.4.0 (reviewer fable-5.1)
What I verified on 2026-10-12: a script over the 0.4.0 tables finds 77 outcomes as the text states; no overlapping windows for any outcome and population; exactly one primary window for every windowed pair, with O36A, O36B and O120 the only outcomes without one, as the Time windows text now says; W00 no longer lists O36A or O36B; every log scale outcome (O22E included) mapped once in turner_map.csv; every continuous outcome in reference_sd.csv; PR04 and PR08 covering O24B and O24C; nrs_synthesis flags matching harms_nrs.csv row for row; every population, class and primary outcome reference resolving; the 43 agents and their terms identical to concepts.yaml 1.2.0; every citation key present in references.bib; and 25 round 3 rows in comments.csv, as amendment A003 states. The rendered 1.2.0 PubMed query still runs with no field errors and returns 28,673 records, as in round 3 (the concepts file is unchanged). registry_dose_candidates.csv is an automated aid with guessed routes and co-administered drug doses captured from arm text; the protocol says it is not a source of nodes, and I did not review it for methods.
1. Round 3 comments
| Item | Status | Reason |
|---|---|---|
| F3-1 | Resolved | W33 (4 to 12 weeks, target 8, secondary) added for O110A to O114 in P14A to P14E with the exact values proposed; the exclusive upper limit 12 meets W22's inclusive 12, so the 6 and 9 week trials now have a network and no overlap is created. |
| F3-2 | Resolved | PR04 and PR08 apply to O24B;O24C; the Effect measures sentence names both, each in its own network. |
| F3-3 | Resolved | Adopted by a different route than I proposed: O22C is now doubling of serum creatinine alone, composites with cardiovascular death are O22D in definition strata, and a general rule pools only identical component sets and thresholds. This is stricter than my 50 then 40 percent sensitivity split and better, because no primary estimate ever mixes thresholds. |
| F3-4 | Resolved | Per procedure and per pregnancy outcomes are stated to have no week window, and O36A and O36B were also removed from W00, which fixes a mismatch I had not raised. |
| F3-5 | Resolved | O04B's rate ratio range 1/1.25 to 1.25 and PR14 are stated in Absolute effects and in the imprecision rule. |
| F3-6 | Resolved | O10 versus O10B sentence replaced as proposed. |
| F3-7 | Resolved | The comparator template note and the Networks and nodes text now give separate activation rules for dose templates and the comparator template. |
| F2-13 follow-up | Resolved | The OpenAlex key and the reconciled four source harvest are in the administrative checklist. |
No response declined any of my proposals.
2. New items in 0.4.0
No major items. The items below clarify new text; none changes a primary analysis, and each can be made before registration without another review round.
Minor
Event outcomes and harms (line 465). "A reported hazard ratio or Kaplan-Meier estimate for time to first event is used when available" gives no route for a Kaplan-Meier estimate into the likelihood list, which has only counts with a cloglog link or a log hazard ratio. Replacement: "A reported hazard ratio enters as a normal likelihood on the log hazard ratio. A reported Kaplan-Meier cumulative incidence p at time t enters as log(-log(1 - p)) with a delta method standard error from its confidence interval and offset log t, on the same log cumulative hazard scale as the count model."
Composite definitions (line 451) versus outcomes.csv O04. The new rule opens "Within any outcome identifier, trials are pooled only when their component sets and thresholds are identical", but O04's mapping rule pools pre 2017 confirmed hypoglycemia below 3.1 mmol/L with level 2 (below 3.0 mmol/L). Which rule wins is unstated. Replacement, appended to the composite paragraph: "This rule governs composite event outcomes. The pre 2017 hypoglycemia mapping in O04 is a prespecified exception; mapped trials are flagged and excluded in a sensitivity analysis."
Confidence in cumulative evidence (lines 822 to 825). Contributions for hazard scale estimates come from "a frequentist inverse variance network of each trial's log hazard ratio derived from its counts". A trial with a zero event arm has no finite log hazard ratio, so the contribution network is not executable for exactly the sparse harms where it matters most. Replacement: add "with 0.5 added to each cell of trials with a zero event arm, for the contribution matrix only".
outcomes.csv O22E. Its population_ids are P01;P02 while O22B, O22C and O22D also list P08A;P08B, so a kidney disease network cannot hold an albuminuria composite although it can hold the other four; and a composite containing both cardiovascular death and macroalbuminuria satisfies O22D and O22E at once. Replacement: population_ids "P01;P02;P08A;P08B", and append to mapping_rules "Composites that also include cardiovascular death are O22D."
Sensitivity analyses (line 740). The list omits two analyses prespecified elsewhere: combining the three iGlarLixi nodes (Networks and nodes) and excluding trials with more than 20 percent incomplete follow-up or a between arm difference above 10 points (Event outcomes and harms). Add both so that one place enumerates every sensitivity analysis.
3. Judgment on round 3 revisions made for the other reviewers
The decision not to adopt descriptive only reporting whenever any participant is censored (C3-1) is sound: the dichotomous analysis over the observation period is the standard approach for adverse effects, the proportional hazards reading of the cloglog model is correct, and the extracted incomplete follow-up, the RoB 2 domain 3 judgment and the 20 percent or 10 point sensitivity analysis make the assumption visible. The capped horizon with reference risks taken only from arms whose follow-up covers it (C3-3) removes the extrapolation I would otherwise have asked about. The iGlarLixi split into the 2:1 pen, the 3:1 pen and one node per switching strategy (C3-4), and the rule that contribution matrices represent the same evidence set as the estimate rated (C3-5), are correct and executable.
Conclusion
Every replacement from my round 3 review was carried into 0.4.0, the invariants I check hold, the search claims reproduce, and the administrative conditions, including the OpenAlex key and the reconciled four source harvest, are tracked in the checklist. The five items above are sentence or row edits. In round 3 I wrote that a version carrying my replacements would receive acceptance without further methods review; that condition is met. I would approve the methods for registration once the administrative pre-registration conditions are completed, with the five edits made in the same pass.
VERDICT: ACCEPT