Methods need minor revision to make new piecewise-hazard fallback executable. No new major error identified. Administrative pre-registration conditions do not determine this verdict.
| Round 8 comment | Status | Assessment |
|---|---|---|
| C8-1 | Resolved | “Competing risks” now defines survival free of both target and competing events and applies treatment hazard multipliers within that survival function and cumulative-incidence integral. Numerical example reproduces 0.518. |
| C8-2 | Resolved | “Data items” and “Competing risks” now require person-time ending at first target event, competing event, censoring, or cutoff. General follow-up, exposure time, and median follow-up cannot substitute. Relative-only fallback is explicit. |
| C8-3 | Resolved | “Absolute effects” now incorporates time varying integration through its cross-reference. Subdistribution conversions have an ordered reference-risk hierarchy; cause specific conversions obtain reference and treated risks through matching routes. |
| C8-4 | Resolved | Primary-network and conclusions rules explicitly assign O20 solely to its not applicable network. One-input-per-trial hierarchy is also present. |
Responses accept all four comments. Two associated decisions are justified. Rejecting G8-6’s flagged median-follow-up denominator is correct because flagging cannot make that denominator valid. F8-4’s distinction between reference-risk routes is also appropriate: an observed cumulative incidence alone cannot support conversion of a cause specific hazard ratio. Corrections to earlier adoption claims are transparently acknowledged.
New finding
C9-1. Minor: new digitized-curve fallback lacks necessary eligibility and failure rules. Locations: protocol/protocol.qmd:419, “Data items,” and protocol/protocol.qmd:577, “Competing risks.”
Revision now instructs reconstruction of piecewise constant hazards and interval-specific log-scale pooling “when such curves exist.” Curve availability alone is insufficient. Target-event cumulative incidence does not identify both cause specific hazards. Reconstruction requires the competing first-event cumulative incidence or equivalent event-free survival information. This follows from (S=1-F_1-F_2) and (d\Lambda_k=dF_k/S(u-)). Haque and Cortese, equation 6 and accompanying identity.
Even with both curves, published intervals can differ across trials, and flat intervals produce zero empirical hazard estimates. Their logarithms cannot enter the stated pooling model. Neither handling of these cases nor minimum information for joint uncertainty is specified. Existing relative-only fallback makes this a bounded methods clarification rather than another incorrect cumulative-incidence formula.
Replace the curve-based pooling sentence with:
Piecewise integration requires estimates of both cause specific hazards throughout the analyzed horizon, with valid joint uncertainty. Reconstruction from cumulative incidence curves requires the target-event and competing first-event curves from the same reference arm and risk set, or equivalent target-event and event-free survival information. A target-event curve alone does not qualify. Interval-specific log-scale pooling requires common reporting intervals, finite log-hazard estimates, and estimable sampling covariance. Zero-event intervals or reference arms are not silently omitted, and arbitrary constants are not added to permit logarithms. When these requirements cannot be met, this reconstruction route is unavailable. For cause specific conversions, use the exponential approximation only when valid rates for both events and support for constant hazards are available; otherwise report relative effects only. Subdistribution conversions retain their existing reference-incidence hierarchy.
Replace the extraction phrase at lines 419 to 420 with:
digitized reference-arm cumulative incidence curves for both the target event and competing first event, their reported uncertainty, and available risk-set information where published.
All requested files inspected. Read-only checks found no identifier, applicable-window, reference-SD, Turner-mapping, nonrandomized-harms coverage, search-alias, or registry-count inconsistencies. Apart from amendments.csv, supplied tables and search files are unchanged from version 0.8.0. Search renderer reviewed statically; R execution and live search recall were not assessed. No files edited, created, or deleted.
After C9-1 is addressed, methods would be acceptable for registration once administrative pre-registration conditions are completed.
VERDICT: MINOR REVISION