Methods require major revision. New piecewise reconstruction can materially misstate absolute risks. Administrative pre-registration conditions do not determine this verdict.
Round 9 assessment
C9-1: Partly resolved. Required paired curves or equivalent information, uncertainty, common intervals, zero-event safeguards, and failure routes are now explicit. Data extraction requirements are corrected. However, newly specified reconstruction formula is incorrect for finite intervals, and competing-curve exception conflicts with required risk sets. Details below.
No proposal was explicitly declined. Requiring every covering reference arm to qualify is defensible because it avoids selecting baseline evidence according to curve availability. A union grid is also reasonable under stated piecewise constant assumptions, provided uncertainty shared by subdivided intervals remains shared.
New findings, ordered by severity
C10-1. Major: curve increments are incorrectly converted to constant hazards.
Location: protocol/protocol.qmd:593, “Competing risks,” piecewise integration.
Dividing a finite cumulative-incidence increment by survival at interval start and interval length gives a conditional event probability per unit time. It is not the constant cause specific hazard over that interval. Survival decreases within the interval.
Numerical counterexample: constant annual hazards (h_1=0.4), (h_2=0.2) produce one-year cumulative incidences (F_1=0.300792), (F_2=0.150396). Current rule returns hazards 0.300792 and 0.150396. Reintegration gives target risk 0.242086, failing to recover supplied risk 0.300792. With target hazard ratio 0.5 and unchanged competing hazard, estimated risk reduction becomes 112.2 instead of 136.0 per 1,000.
Replace curve-reconstruction rule with:
For each original reporting interval ([s,t]), let (\Delta=t-s), (\Delta F_k=F_k(t)-F_k(s)), and (S(u)=1-F_1(u)-F_2(u)). Under constant cause specific hazards within that interval, reconstruct [ h_k=\frac{\Delta F_k}{\Delta F_1+\Delta F_2} \frac{-\log{S(t)/S(s)}}{\Delta}. ] Require compatible first-event curves, (0<S(t)<S(s)), and the existing finite-log-hazard and joint-uncertainty conditions. If event-free survival is supplied instead of the competing curve, derive (F_2=1-S-F_1). Subdivision onto the common grid retains the original interval’s hazard and shared uncertainty. Counts with valid person-time retain the events/person-time estimator.
This formula follows by inverting the constant-hazard cumulative-incidence relationship, applied conditionally at interval start. Haque and Cortese, equations 7 and 12.
C10-2. Minor: competing-curve exception needs an explicit boundary.
Location: protocol/protocol.qmd:600, “Competing risks,” final sentence of piecewise integration.
Earlier eligibility requires both curves from the same first-event risk set. This sentence then permits a competing-event curve retaining participants after a nonfatal target event. Assuming unchanged death hazard after the target event can justify borrowing an appropriately estimated mortality hazard. It does not make cumulative mortality including subsequent deaths interchangeable with competing first-event cumulative incidence. Those probabilities overlap with target events and cannot enter the paired-curve reconstruction unchanged.
Simplest executable correction is to retain the stricter eligibility rule.
Replace that sentence with:
Paired-curve reconstruction accepts only cumulative incidences of mutually exclusive first events from the same arm and risk set. A mortality curve that includes deaths after a nonfatal target event cannot substitute for the competing first-event curve. When qualifying inputs are unavailable, use another eligible route or report relative effects only.
A broader mortality-hazard borrowing route would require its own transformation and uncertainty specification.
C10-3. Minor: new failure wording overrides an otherwise valid subdistribution conversion.
Location: protocol/protocol.qmd:633, “Competing risks,” constant-hazard assessment.
New parenthetical directs relative-only reporting when neither curves nor observations at multiple times exist. Yet a reported Aalen-Johansen estimate at the analyzed horizon can support subdistribution conversion without constant cause specific hazards, as explicitly allowed at protocol/protocol.qmd:614.
Thus, a trial with a valid single-horizon reference incidence and supported proportional subdistribution hazards receives conflicting instructions.
Replace the parenthetical with:
When neither source exists, the exponential approximation is unavailable. Retain any other eligible absolute-risk route, including reported Aalen-Johansen cumulative incidence for subdistribution conversion. Report relative effects only when no applicable route qualifies.
All requested files inspected, including programmatic checks across 861 registry-candidate rows. No additional identifier, applicable-window, reference-SD, Turner-mapping coverage, nonrandomized-harms coverage, search-alias, or registry-count inconsistencies identified. Search renderer reviewed statically; R execution and live search recall were not assessed. No files edited, created, or deleted.
Methods would be acceptable for registration after these corrections and completion of administrative pre-registration conditions.
VERDICT: MAJOR REVISION