Methods need minor revision. Administrative pre-registration conditions do not determine this verdict.
| Round 5 comment | Status | Assessment |
|---|---|---|
| C5-1 | Partly resolved | Requiring complete status for primary count inputs fixes the substantive censoring problem. Moving the 95% rule to sensitivity analysis is appropriate. The remaining general claim about equal loss requires qualification. |
| C5-2 | Partly resolved | Hazard types, separate networks, removal of the mortality exemption, and finite interior cumulative-incidence inputs address most concerns. Proportional subdistribution hazards remain unstated; new conversion and complete-status rules need qualifications below. |
| C5-3 | Resolved | Trial-specific proportions and risk differences replace the unsupported pooled absolute effect. Declining a pooled difference at median duration is justified because observation processes differ. The newly introduced rate-ratio label needs correction. |
| C5-4 | Resolved | Contribution inputs now match primary-model inputs, preserve shared-arm covariance, and retain the qualitative fallback. |
| C5-5 | Resolved | Unavailable absolute effects and categories are explicit, with a relative certainty rule and ranking exception. Choosing a prespecified numerical relative convention instead of qualitative assessment is defensible. It does not establish equivalence to outcome-specific absolute thresholds. |
| C5-6 | Resolved | Duplicate O22C entry removed; narrative agrees with the authoritative outcome table. |
No remaining major finding. Remaining and newly introduced minor findings:
New cause-specific conversion omits its constant-hazard assumption. protocol/protocol.qmd:524, “Competing risks.”
The displayed formula applies to constant cause-specific hazards. Having both hazards available does not make that formula valid for arbitrary time-varying hazards. It also needs to distinguish parametric conversion from Aalen-Johansen estimation. Haque and Cortese, equations 6 and 7 give the general integral and its exponential special case.
Replacement: “The displayed cause-specific conversion assumes that both cause-specific hazards are constant over the analyzed horizon. We label resulting absolute effects as an exponential competing-risks approximation. Both treatment multipliers must be cause-specific hazard ratios; leaving the competing hazard unchanged assumes its treatment hazard ratio equals 1. When time-varying baseline hazards are available, calculate cumulative incidence by integrating survival times the event-specific hazard. Otherwise, if the constant-hazard approximation cannot be supported, report relative effects only.”
Cumulative-incidence contrasts still need an explicit proportionality condition. protocol/protocol.qmd:473 and protocol/protocol.qmd:509, “Hazard ratio analysis” and “Competing risks.”
A complementary log-log contrast at one horizon identifies a cumulative subdistribution-hazard ratio. Interpreting it as a time-invariant subdistribution HR, combining it with Fine-Gray estimates across durations, and applying the absolute conversion require proportional subdistribution hazards. This assumption differs from proportional cause-specific hazards. Putter and colleagues explain that distinction.
Replacement: “Cumulative-incidence contrasts and complete-status counts enter subdistribution-hazard networks under an explicit proportional subdistribution-hazards assumption. We extract reported proportionality assessments and inspect available curves. Where material nonproportionality is identified, report horizon-specific cumulative-incidence contrasts separately and do not use a constant subdistribution HR to convert effects across horizons.”
New recorded-event rate-ratio label does not match the fitted quantity. protocol/protocol.qmd:492, “Recorded harms analysis.”
With recorded proportion (r) and nominal observation period (t), the fitted quantity is (-\log(1-r)/t). It is not recorded events divided by observed person-time. For equal periods and proportions 0.50 and 0.75, the model gives a ratio of 2.0; recorded events per randomized participant-period give 1.5. Missing observation prevents interpreting the former as an ordinary incidence rate ratio.
Replacement: “Report the exponentiated complementary log-log contrast: [ Q=\frac{-\log(1-r_B)/t_B}{-\log(1-r_A)/t_A}, ] where (r) is the recorded event proportion and (t) is the nominal observation period in years. Interpret (Q) descriptively, without an incidence rate or biological hazard interpretation. Certainty uses the prespecified range (0.8) to (1.25) on (Q).”
Replace “recorded event rate ratio” consistently in the response and protocol.
Complete-status definition does not explicitly accommodate competing deaths. protocol/protocol.qmd:479, compared with “Competing risks” at line 512.
Requiring observation event free through (t) leaves participants who die earlier outside the stated definition, although the competing-risk paragraph includes them as known non-events. Administrative closure before a participant reaches (t) also needs explicit treatment.
Replacement: “Status at (t) is known when the target event occurred by (t), a documented competing death occurred by (t) without a preceding target event, or observation confirms absence of the target event through (t). Participants with competing deaths remain in the randomized denominator. Loss or administrative censoring before (t), without an established target or competing event, leaves status unknown.”
Equal loss still receives an unjustified general direction of bias. protocol/protocol.qmd:485, “Hazard ratio analysis.”
Equal missing fractions do not imply independent missingness or equal loss of events. Equal 5% missingness can produce recorded risks of 0.05 versus 0.10 from true risks of 0.10 in both arms if missing participants have different outcome distributions. Cochrane likewise distinguishes missing proportions from reasons and mechanisms. Cochrane Handbook, section 8.5
Replacement: “There is no general bound or direction of bias, even with equal missing fractions. In the following illustrative calculation, missingness is independent of outcome and occurs at baseline: equal 5% loss changes the contrast from 2.0 to approximately 1.93.”
All requested files reviewed. Read-only checks found no mismatched identifiers, overlapping applicable windows, missing reference-SD or Turner mappings, or registry candidate count discrepancies. Search renderer inspected statically; live search recall and R execution were not assessed. No files edited, created, or deleted.
VERDICT: MINOR REVISION