pith. sign in

arxiv: 2604.04360 · v2 · submitted 2026-04-06 · 📊 stat.ME

Generalized win fraction regression for composite survival endpoints

Zhiqiang Cao , Xi Fang , Fan Li This is my paper

Pith reviewed 2026-05-10 20:24 UTC · model grok-4.3

classification 📊 stat.ME
keywords win fraction regressioncomposite survival endpointsinverse probability weightinglink functionsproportional win fractionclinical trial analysiscensoring adjustment
0
0 comments X

The pith

A generalized regression models conditional win fractions for composite survival endpoints with flexible links and censoring weights.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes a regression framework that directly targets the conditional probability of winning on a prioritized composite survival outcome. By choosing a link function such as identity or logit, it places the win fraction or its log odds on a regression scale that accommodates multiple time-to-event components. Inverse-probability-of-censoring weights are inserted into the estimating equations so that the target remains the win fraction that would be observed without censoring. Large-sample consistency and a sandwich variance that accounts for weight estimation are established. The approach is demonstrated on simulations and reanalyzed clinical trial data.

Core claim

We propose a generalized win fraction regression framework for prioritized composite survival outcomes. The framework models the conditional win fraction through a chosen link function (including identity, logit, or probit), thereby accommodating multi-component time-to-event endpoints within a unified regression structure. To handle right censoring, we construct inverse-probability-of-censoring-weighted estimating equations that target the win fraction as if censoring were absent. Under the identity link, regression parameters characterize covariate associations on the natural win fraction scale. Under the logit link, they characterize the log odds of winning -- a new and complementary 0.5

What carries the argument

Inverse-probability-of-censoring-weighted estimating equations that solve for regression parameters in a link-function model for the conditional win fraction.

If this is right

  • Under the identity link the coefficients give direct changes in expected win fraction per unit change in a covariate.
  • Under the logit link the coefficients give log odds of winning, treating every tie as a loss.
  • When ties are absent the logit model reduces exactly to proportional win fraction regression.
  • The unweighted estimating equations coincide numerically with the existing proportional win fraction estimator regardless of ties.
  • A consistent sandwich variance estimator is available that incorporates the variability from estimated censoring weights.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The logit formulation supplies a more conservative effect measure than the conventional win ratio when ties are common.
  • The framework could be applied to composite endpoints that mix survival times with other prioritized outcomes such as recurrent events.
  • Empirical comparison of link functions on the same data set would show whether the choice materially affects clinical conclusions.
  • Extension to time-varying covariates would require only modification of the weight construction and the regression specification.

Load-bearing premise

The model for the censoring hazard given covariates must be correctly specified so that the weights consistently recover the win fraction that would be seen in the absence of censoring.

What would settle it

In repeated simulations with known true win fractions, check whether the point estimators remain consistent and the sandwich intervals achieve nominal coverage when the censoring model is deliberately misspecified or when censoring is heavy.

Figures

Figures reproduced from arXiv: 2604.04360 by Fan Li, Xi Fang, Zhiqiang Cao.

Figure 1
Figure 1. Figure 1: Trajectories of the regression coefficient estimates, βb L, from the win-fraction regression model with logit link with and without IPCW across restricted times L (in years). Each panel corresponds to one baseline covariate. The solid curve denotes the point estimate, and the shaded region denotes the 95% confidence interval. 29 [PITH_FULL_IMAGE:figures/full_fig_p029_1.png] view at source ↗
read the original abstract

We propose a generalized win fraction regression framework for prioritized composite survival outcomes. The framework models the conditional win fraction through a chosen link function (including identity, logit, or probit), thereby accommodating multi-component time-to-event endpoints within a unified regression structure. To handle right censoring, we construct inverse-probability-of-censoring-weighted estimating equations that target the win fraction as if censoring were absent. Under the identity link, regression parameters characterize covariate associations on the natural win fraction scale. Under the logit link, they characterize the log odds of winning -- a new and complementary effect measure that treats ties as failures to win, imposing a more conservative standard than the win ratio or win odds. When there are no ties, the logit win fraction model reduces to proportional win fraction regression; moreover, the unweighted version of our estimating equations numerically coincides with the proportional win fraction point estimator regardless of ties. We establish large-sample properties of the proposed estimators and derive a consistent sandwich variance estimator that accounts for uncertainty from the estimated censoring weights. Extensive simulations examine finite-sample performance across link functions and censoring rates, and our method is illustrated through a reanalysis of the HF-ACTION clinical trial.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 3 minor

Summary. The paper proposes a generalized win fraction regression framework for prioritized composite survival endpoints. It models the conditional win fraction via link functions (identity, logit, probit), constructs IPCW estimating equations to target the uncensored win fraction, establishes large-sample consistency and asymptotic normality with a sandwich variance estimator that accounts for estimated censoring weights, notes numerical coincidence with proportional win fraction estimators under the identity link or no ties, and illustrates the method with simulations and a reanalysis of the HF-ACTION trial.

Significance. If the asymptotic results hold under the stated assumptions, the framework provides a unified, flexible regression approach for composite time-to-event outcomes with interpretable parameters on different scales (e.g., log-odds under logit). The explicit IPCW construction, sandwich variance accounting for weight uncertainty, and compatibility with existing estimators are strengths that enhance applicability in clinical trials. The simulations and real-data example support practical utility.

major comments (3)
  1. [§3, Theorem 1] §3 (IPCW estimating equations U_n(β)): Consistency and the validity of the sandwich variance in Theorem 1 require correct specification of the censoring hazard model (e.g., Cox or parametric) conditional on covariates. Misspecification would make the weighted equations target a different quantity than the population win fraction, yet the manuscript provides neither robustness results nor simulations under misspecification.
  2. [Theorem 1] Theorem 1: The large-sample properties are derived, but the statement does not explicitly list the regularity conditions (e.g., positivity of censoring weights, smoothness of the link function, boundedness of covariates, and convergence rates for Ĝ). This omission makes it hard to verify the scope without the full proof.
  3. [Simulation section] Simulation study: The data-generating mechanisms for the multi-component composite endpoints, the specific censoring rates, and the censoring model fitted in the IPCW step are described at too high a level to reproduce the reported finite-sample bias, coverage, and efficiency results across links.
minor comments (3)
  1. [Abstract] The abstract states that the unweighted equations numerically coincide with the proportional win fraction estimator 'regardless of ties,' but the precise conditions (identity link vs. logit, handling of ties as failures) should be clarified in the main text for immediate readability.
  2. [§2] Notation for the prioritized win indicator, tie probability, and the link function g(·) is introduced late; defining these symbols in §2 before the estimating equations would improve flow.
  3. [Application section] In the HF-ACTION reanalysis, the exact prioritization order of the composite components and the fitted censoring model should be stated explicitly to allow readers to assess the practical effect sizes.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the positive assessment and constructive feedback on our manuscript. We address each major comment below and outline the planned revisions.

read point-by-point responses

  1. Referee: [§3, Theorem 1] §3 (IPCW estimating equations U_n(β)): Consistency and the validity of the sandwich variance in Theorem 1 require correct specification of the censoring hazard model (e.g., Cox or parametric) conditional on covariates. Misspecification would make the weighted equations target a different quantity than the population win fraction, yet the manuscript provides neither robustness results nor simulations under misspecification.

    Authors: We agree that correct specification of the censoring model is necessary for the IPCW estimator to consistently target the population win fraction. This assumption is implicit in our derivation but will be stated explicitly in the revised manuscript. We will also add a short discussion noting that misspecification could lead to bias and that robustness checks may be warranted in practice. Given the minor revision recommendation, we believe adding this clarification suffices without new simulations, though we are open to including a brief sensitivity analysis if requested. revision: partial

  2. Referee: [Theorem 1] Theorem 1: The large-sample properties are derived, but the statement does not explicitly list the regularity conditions (e.g., positivity of censoring weights, smoothness of the link function, boundedness of covariates, and convergence rates for Ĝ). This omission makes it hard to verify the scope without the full proof.

    Authors: Thank you for this observation. In the revised manuscript, we will augment the statement of Theorem 1 with a remark that explicitly lists the key regularity conditions, including positivity of the censoring weights (inf G(t|X) > 0), smoothness of the link function, boundedness of covariates, and the required convergence rate for the estimator Ĝ of the censoring survival function. This will make the assumptions transparent without altering the proof in the appendix. revision: yes

  3. Referee: [Simulation section] Simulation study: The data-generating mechanisms for the multi-component composite endpoints, the specific censoring rates, and the censoring model fitted in the IPCW step are described at too high a level to reproduce the reported finite-sample bias, coverage, and efficiency results across links.

    Authors: We appreciate the feedback on reproducibility. In the revised version, we will expand the simulation section to include more precise descriptions: the specific hazard rates and distributions for each component of the composite endpoint, the exact censoring rates (e.g., 20%, 40%), and the covariates included in the censoring model (e.g., Cox model with treatment and other baseline variables). We will also clarify the implementation details to allow exact reproduction of the results. revision: yes

Circularity Check

0 steps flagged

No circularity; new link-function framework and IPCW equations are independently defined

full rationale

The derivation introduces a generalized regression model for the conditional win fraction via arbitrary link functions and defines IPCW estimating equations that directly target the uncensored population win fraction. Large-sample consistency and sandwich variance follow from standard M-estimation theory under the explicit assumption of correct censoring model specification. The noted numerical coincidence with prior proportional win fraction estimators is derived as a consequence of the new equations rather than presupposed as an input; no step equates a claimed result to a fitted parameter or self-citation by construction. The framework remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The method rests on standard survival-analysis assumptions for right-censored data and on the existence of a correctly specified model for the censoring distribution; no free parameters or new entities are introduced.

axioms (1)
  • domain assumption Censoring is independent of the potential event times conditional on observed covariates
    Required for the inverse-probability weights to produce unbiased estimating equations for the uncensored win fraction.

pith-pipeline@v0.9.0 · 5502 in / 1497 out tokens · 43536 ms · 2026-05-10T20:24:19.347960+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

  1. [1]

    Bebu, I. and J. M. Lachin (2016). Large sample inference for a win ratio analysis of a composite outcome based on prioritized components. Biostatistics\/ 17\/ (1), 178--187

    work page 2016

  2. [2]

    Buyse, M. (2010). Generalized pairwise comparisons of prioritized outcomes in the two-sample problem. Statistics in Medicine\/ 29\/ (30), 3245--3257

    work page 2010

  3. [3]

    Cox, D. R. (1972). Regression models and life-tables. Journal of the Royal Statistical Society: Series B (Methodological)\/ 34\/ (2), 187--202

    work page 1972

  4. [4]

    Huang, G

    Cui, Y., B. Huang, G. Dong, R. Uozumi, and L. Tian (2025). An ipcw adjusted win statistics approach in clinical trials incorporating equivalence margins to define ties. Statistics in Medicine\/ 44\/ (15-17), e70180

    work page 2025

  5. [5]

    Dong, G., D. C. Hoaglin, J. Qiu, R. A. Matsouaka, Y.-W. Chang, J. Wang, and M. Vandemeulebroecke (2020). The win ratio: on interpretation and handling of ties. Statistics in Biopharmaceutical Research\/

    work page 2020

  6. [6]

    Huang, D

    Dong, G., B. Huang, D. Wang, J. Verbeeck, J. Wang, and D. C. Hoaglin (2021). Adjusting win statistics for dependent censoring. Pharmaceutical Statistics\/ 20\/ (3), 440--450

    work page 2021

  7. [7]

    Uno, and F

    Fang, X., H. Uno, and F. Li (2025). While-alive regression analysis of composite survival endpoints. Biostatistics\/ 26\/ (1), kxaf047

    work page 2025

  8. [8]

    Flynn, K. E., I. L. Piña, D. J. Whellan, L. Lin, J. A. Blumenthal, S. J. Ellis, L. J. Fine, J. G. Howlett, S. J. Keteyian, D. W. Kitzman, W. E. Kraus, N. H. Miller, K. A. Schulman, J. A. Spertus, C. M. O’Connor, and K. P. Weinfurt (2009, 04). Effects of exercise training on health status in patients with chronic heart failure: Hf-action randomized control...

    work page 2009

  9. [9]

    Foutz, R. V. (1977). On the unique consistent solution to the likelihood equations. Technical Report 357

    work page 1977

  10. [10]

    Kowalski, J. and X. M. Tu (2008). Modern applied U-statistics . John Wiley & Sons

    work page 2008

  11. [11]

    Lumley, T. and N. Hamblett (2003). Asymptotics for marginal generalized linear models with sparse correlations. Technical Report 207, University of Washington, Seattle

    work page 2003

  12. [12]

    Luo, X., H. Tian, S. Mohanty, and W. Y. Tsai (2015). An alternative approach to confidence interval estimation for the win ratio statistic. Biometrics\/ 71\/ (1), 139--145

    work page 2015

  13. [13]

    Mao, L. (2024). Defining estimand for the win ratio: Separate the true effect from censoring. Clinical Trials\/ 21\/ (5), 584--594

    work page 2024

  14. [14]

    Mao, L. and T. Wang (2020, 09). A class of proportional win-fractions regression models for composite outcomes. Biometrics\/ 77\/ (4), 1265--1275

    work page 2020

  15. [15]

    Neve, J. D., O. Thas, and J.-P. Ottoy (2013). Goodness-of-fit methods for probabilistic index models. Communications in Statistics - Theory and Methods\/ 42\/ (7), 1193--1207

    work page 2013

  16. [16]

    Oakes, D. (2016). On the win-ratio statistic in clinical trials with multiple types of event. Biometrika\/ 103\/ (3), 742--745

    work page 2016

  17. [17]

    Budtz-Jørgensen, and J

    Ozenne, B., E. Budtz-Jørgensen, and J. Péron (2021). The asymptotic distribution of the net benefit estimator in presence of right-censoring. Statistical Methods in Medical Research\/ 30\/ (11), 2399--2412

    work page 2021

  18. [18]

    Pocock, S. J., C. A. Ariti, T. J. Collier, and D. Wang (2012). The win ratio: a new approach to the analysis of composite endpoints in clinical trials based on clinical priorities. European Heart Journal\/ 33\/ (2), 176--182

    work page 2012

  19. [19]

    Idlhaj, D

    Péron, J., M. Idlhaj, D. Maucort-Boulch, J. Giai, P. Roy, L. Collette, M. Buyse, and B. Ozenne (2021). Correcting the bias of the net benefit estimator due to right-censored observations. Biometrical Journal\/ 63\/ (4), 893--906

    work page 2021

  20. [20]

    Verbeeck, B

    Song, J., J. Verbeeck, B. Huang, D. C. Hoaglin, M. Gamalo-Siebers, Y. Seifu, D. Wang, F. Cooner, and G. Dong (2023). The win odds: statistical inference and regression. Journal of Biopharmaceutical Statistics\/ 33\/ (2), 140--150

    work page 2023

  21. [21]

    Thas, O., J. D. Neve, L. Clement, and J.-P. Ottoy (2012, 07). Probabilistic index models. Journal of the Royal Statistical Society Series B: Statistical Methodology\/ 74\/ (4), 623--671

    work page 2012

  22. [22]

    Wang, and Y

    Wang, B., Z. Wang, and Y. Cheng (2026). Generalized win-odds regression models for composite endpoints. Lifetime Data Analysis\/ 32\/ (1), 13

    work page 2026

  23. [23]

    Zhong, Y. and D. E. Schaubel (2022). Restricted mean survival time as a function of restriction time. Biometrics\/ 78\/ (1), 192--201

    work page 2022