REVIEW 3 minor 8 references
Conditional tie weighting lets partially censored pairs contribute fractionally to restricted-time win probabilities without altering the target estimand.
Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →
T0 review · grok-4.3
2026-07-04 01:00 UTC pith:Y5UBEPNB
load-bearing objection The paper replaces the all-or-nothing genuine-tie rule with conditional-probability weights for censored higher-priority pairs, recovering efficiency while keeping the same restricted-time win-probability target.
Making censored pairs count: conditional tie weighting for win statistics with composite survival endpoints
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
Conditional tie weighting replaces the unavailable higher-priority genuine-tie indicator by its conditional probability given the observed pairwise data. The resulting estimator targets the same restricted-time win probabilities as existing all-or-nothing methods while allowing partially observed pairs to contribute fractionally when their lower-priority comparison is informative. Identification and large-sample theory are established for the two-sample U-statistics with estimated nuisance functions, along with sandwich variance estimators for the win ratio, net benefit, and win odds.
What carries the argument
conditional tie weighting, which substitutes the higher-priority genuine-tie indicator with its conditional probability given observed pairwise data
Load-bearing premise
The conditional probability of a higher-priority genuine tie given the observed pairwise data can be consistently estimated from the observed data without introducing bias into the win probability estimand.
What would settle it
A simulation study or data reanalysis in which the nuisance estimator for the conditional tie probability is deliberately misspecified and the resulting win probability estimator deviates from the known restricted-time target.
If this is right
- The estimator remains consistent for the same restricted-time win probabilities targeted by existing methods.
- Asymptotic normality holds for the U-statistic even after plugging in consistent nuisance estimators.
- Sandwich variance estimators are available for the win ratio, net benefit, and win odds.
- Efficiency gains appear in finite samples, especially with heavier censoring or longer restriction horizons.
Where Pith is reading between the lines
- The weighting approach may apply directly to other hierarchical composite endpoints that descend through ordered outcomes under censoring.
- Reanalysis of completed trials can recover information previously discarded when higher-priority events are censored.
- Power calculations for future trials with composite endpoints could be revised upward once fractional contributions from censored pairs are included.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript proposes conditional tie weighting for win statistics applied to hierarchical composite survival endpoints under right censoring. By replacing the indicator for an unobserved higher-priority genuine tie with its conditional probability given the observed pairwise data, the estimator permits partially observed pairs to contribute fractionally to lower-priority comparisons. The authors state that this construction targets the same restricted-time win probabilities as existing all-or-nothing approaches, establish identification and large-sample theory for the resulting two-sample U-statistics that incorporate estimated nuisance functions, derive sandwich variance estimators for the win ratio, net benefit, and win odds, demonstrate efficiency gains in simulations, and illustrate the method via reanalysis of the HF-ACTION trial.
Significance. If the identification result and asymptotic theory hold, the work provides a practical efficiency improvement for a common class of composite-endpoint analyses in clinical trials without changing the target estimand. Credit is given for deriving large-sample theory for U-statistics with estimated nuisances, providing sandwich variance estimators, conducting simulations across censoring levels and restriction horizons, and including a real-data application. These elements support reproducibility and allow direct comparison with existing restricted win-statistic methods.
minor comments (3)
- [Abstract] Abstract and introduction: the claim that the estimator 'targets the same restricted-time win probabilities' would benefit from an explicit one-line statement of the target parameter (e.g., the population win probability under the restriction time) before the weighting is introduced.
- [Methods] Methods: the nuisance estimation procedure for the conditional tie probability should include a brief statement of the rate condition required for the asymptotic normality result to hold when the nuisance is estimated at a slower rate than n^{-1/2}.
- [Simulations] Simulations: Table 1 (or equivalent) reports efficiency gains; add the Monte Carlo standard error of the reported relative efficiencies so readers can judge whether the gains are distinguishable from simulation noise under the heavier-censoring scenarios.
Simulated Author's Rebuttal
We thank the referee for the detailed and positive summary of our work on conditional tie weighting for win statistics with composite survival endpoints. The recommendation of minor revision is appreciated, and we note that the report raises no specific major comments or criticisms requiring substantive changes to the identification result, asymptotic theory, variance estimators, simulations, or real-data application.
Circularity Check
No significant circularity identified
full rationale
The paper defines the target restricted-time win probabilities as an externally specified functional of the hierarchical composite endpoint distribution under right censoring. Conditional tie weighting is introduced via the law of iterated expectations to replace the unobserved higher-priority tie indicator with its conditional expectation given observed data, yielding an estimator that identifies the same target quantity. Identification, U-statistic theory, and sandwich variance follow from standard semiparametric arguments with estimated nuisances; no equation reduces the target win probability to a fitted parameter or self-citation by construction. The construction is self-contained against the stated assumptions and does not rely on load-bearing self-citations or ansatzes imported from prior work by the same authors.
Axiom & Free-Parameter Ledger
read the original abstract
Hierarchical composite endpoints are increasingly used in clinical trials to compare patients first on the most clinically important outcome and then, only when that comparison is tied, on lower priority outcomes. Under right censoring, a lower priority comparison may already be observed but still cannot contribute because the higher priority genuine tie required for descent through the hierarchy is not confirmed. Existing restricted win-statistic estimators address censoring by requiring such ties from higher priority to be observed as genuine ties. This all-or-nothing rule preserves the restricted-time estimand, but excludes pairs with censoring-induced ties even when their lower priority comparisons contain useful information. We propose conditional tie weighting, which replaces the unavailable higher priority genuine-tie indicator by its conditional probability given the observed pairwise data. The resulting estimator targets the same restricted-time win probabilities while allowing partially observed pairs to contribute fractionally when their lower priority comparison is informative. We establish identification and large-sample theory for the resulting two-sample U-statistics with estimated nuisance functions, and derive sandwich variance estimators for the win ratio, net benefit, and win odds. Simulations show substantial efficiency gains, especially under heavier censoring and longer restriction horizons. A reanalysis of the HF-ACTION trial illustrates how conditional tie weighting recovers information from censoring-induced ties in death-first hospitalization comparisons further apply our estimator to reanalyze a completed randomized clinical trial.
Figures
Reference graph
Works this paper leans on
-
[1]
Buyse, M. (2010). Generalized pairwise comparisons of prioritized outcomes in the two-sample problem.Statistics in medicine,29(30), 3245–3257. Cao, Z., Fang, X., & Li, F. (2026). Generalized win fraction regression for composite survival endpoints.arXiv preprint arXiv:2604.04360. Clayton, D. G. (1978). A model for association in bivariate life tables and ...
work page internal anchor Pith review Pith/arXiv arXiv doi:10.1007/bf02189866 2010
-
[2]
Now fixq≥2. The oracle kernel under (5) is K(ipcw) ij,q = hQ k<q I n eYki(τ) =τ, eYkj(τ) =τ oi I n eYqi(τ)> eYqj(τ) o δqj(τ) G1(τ|Z i)G 0(τ|Z j) . Under the common censoring convention,I n eYki(τ) =τ o = I{Tki > τ}I{C i > τ}for eachk < q, and on the event {Cj > τ} the condition Tqj ≤τ implies Cj > T qj, so I n eYqi(τ)> eYqj(τ) o δqj(τ) = I{T qi > T qj }I{...
work page 2026
-
[3]
31 APREPRINT- MAY27, 2026 If the dependence between adjacent components is expected to differ from the dependence between clinically distant components, a nested Archimedean copula or a vine construction provides additional flexibility. Such choices affect only the nuisance functions bR> q,a and bR= q,a, and the IPCW structure of (8) is unchanged. In nume...
work page 2026
-
[4]
The lower-censoring results show the same qualitative pattern as the 80% setting in the main text, but with smaller efficiency gains because fewer higher-priority ties are unresolved before the restriction time. Under the correctly specified nuisance configuration M1, both estimators have small relative bias and empirical coverage close to the nominal 95%...
work page 2025
-
[5]
In this section, we provide the corresponding results under 40% censoring and the analogous results for WR(τ) and WO(τ). Across these additional settings, the proposed estimator remains substantially more efficient than the IPCW estimator of Cui et al. (2025). under all working copulas considered. The efficiency gain is smaller under 40% censoring than un...
work page 2025
-
[6]
remotes", quietly = TRUE)) { install.packages(
= 0.5×0.4 z30.61−z3 , 51 APREPRINT- MAY27, 2026 Web Table 10: Simulation results for WO(τ) under 40% censoring by τ= 36 for the two-component prioritized endpoint. RB: relative bias (%) for dWO(τ) on the natural scale; MCSD: Monte Carlo standard deviation of logdWO(τ); ASE: average estimated standard error for logdWO(τ); Cov: empirical coverage of the 95%...
work page 2026
-
[7]
The option method = c("ipcw", "m-ipcw") computes both the 57 APREPRINT- MAY27, 2026 Web Table 16: Simulation results for NB(τ) under 80% censoring by τ= 36 for the three-component priori- tized endpoint. Same layout as the main-text simulation tables. RB: relative bias (%); MCSD: Monte Carlo stan- dard deviation; ASE: average estimated standard error; Cov...
work page 2026
-
[8]
returns the requested win statistics at each restriction time. Additional working copulas for the modified IPCW estimator can be requested by changing the copula argument, for example copula = c("gumbel", "clayton", "frank", "plackett"). 58 APREPRINT- MAY27, 2026 Web Table 17: Simulation results for WR(τ) under 20% censoring by τ= 36 for the three-compone...
work page 2026
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.