Pith. sign in

REVIEW 2 major objections 5 minor 34 references

Conditional copula graphic estimator for semi-competing risks data

T0 review · 2 major / 5 minor · reviewed 2026-07-14 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Ignoring covariates in the margins of semi-competing risks data biases the non-terminal survival estimate; a working constant copula does not.

desk verdict Solid, usable extension of the copula-graphic estimator to conditional semi-competing risks; the finite-sample comparison of ignoring covariates is the real payoff, asymptotics and multi-covariate support are missing but not fatal for the claims made. read the letter →

arxiv 2607.09894 v1 pith:ACUYMHVQ submitted 2026-07-10 stat.ME

Shamsia Sobhan , Elif Fidan Acar This is my paper
classification stat.ME MSC 62N0162G0562H05
keywords Beran’sestimatorcopula-graphiccovariateadjustmentiterativealgorithmsemi-competingriskssemiparametricestimationArchimedeancopula
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Semi-competing risks data arise when a non-terminal event (for example relapse) can be censored by a terminal event (death). The paper develops a conditional copula-graphic estimator that lets a continuous covariate enter both the marginal survival functions and the dependence between the two event times. The estimator is semi-parametric: an Archimedean copula family is fixed while its parameter function and the margins are estimated non-parametrically by an alternating iterative algorithm that uses Beran’s estimator and local likelihood. Simulation and two transplant data sets show that omitting a covariate that truly affects the margins produces large bias in the non-terminal survival curve, whereas treating the copula parameter as constant still recovers nearly the same survival estimates even when dependence itself varies with the covariate. The practical message is that covariate adjustment for the margins is essential; full modelling of covariate-dependent association is needed only when the dependence function itself is of scientific interest.

What carries the argument

The sequential iterative algorithm (Algorithm 1) that alternates Beran’s nonparametric estimator for the terminal margin, the conditional copula-graphic formula for the non-terminal margin, and local-likelihood estimation of the Archimedean copula parameter function until the non-terminal survival estimate stabilizes.

What would settle it

Generate semi-competing risks data from a non-Archimedean or mixture copula whose dependence family itself changes with the covariate, then check whether the iterative estimator still recovers the true non-terminal survival curve and whether the constant-copula version remains nearly unbiased.

Watch

Extended reading notes

Core claim

A conditional copula-graphic estimator that adjusts for a continuous covariate in both margins and association, obtained by a sequential iterative algorithm, shows that failure to adjust the margins for covariates substantially biases the non-terminal survival function, while a simplified version that keeps the copula parameter constant yields essentially the same survival estimates as the fully covariate-dependent association model.

Load-bearing premise

The true conditional dependence must belong to one fixed one-parameter Archimedean family for every value of the continuous covariate; if the family is wrong or changes with the covariate, the estimator loses its justification.

Editorial extensions

If this is right

  • Analysts of clinical semi-competing risks data should routinely include continuous covariates in the marginal survival functions rather than relying on the classical unconditional copula-graphic estimator.
  • When only the non-terminal survival curve is required, the simpler constant-copula version (CCGE1) can be used without material loss of accuracy.
  • When the scientific question concerns how association itself varies with age or other covariates, the fully conditional estimator (CCGE2) must be retained.
  • A nonparametric bootstrap supplies practical pointwise confidence bands for both the conditional survival function and the dependence parameter.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same iterative scheme could be adapted to other parametric copula families once generators and their derivatives are supplied, potentially relaxing the Archimedean restriction.
  • Because bandwidth selection is performed by leave-one-out cross-validation, the method is already close to automatic and could be packaged for routine clinical use once multiple-covariate extensions exist.
  • The finding that margin misspecification is far more damaging than association misspecification suggests a practical two-stage workflow: first force covariate adjustment on the margins, then test whether the constant-copula working model is adequate.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper proposes a conditional copula-graphic estimator (CCGE) for the marginal survival function of a non-terminal event under semi-competing risks, allowing a continuous covariate to affect both the margins (via Beran's estimator) and the dependence (via a one-parameter Archimedean copula whose parameter is estimated by local likelihood). Two versions are developed: CCGE1 (working constant dependence) and CCGE2 (fully covariate-dependent dependence). Estimation proceeds by a sequential iterative algorithm that alternates updates of S1|X and the copula parameter until convergence. Finite-sample performance is examined under three data-generating models (no covariate effect, margin-only effect, margin-plus-dependence effect), three Archimedean families, three censoring rates and two sample sizes; the estimators are also illustrated on the Stanford heart-transplant and bone-marrow-transplant data sets. The central empirical claim is that ignoring covariate effects on the margins produces substantial bias in S1|X, whereas a working constant-copula assumption yields survival estimates nearly indistinguishable from the fully flexible estimator.

Significance. The work fills a genuine methodological gap: most existing semi-competing-risks methods either ignore covariates or restrict them to discrete/binary settings or fully parametric frailty models. Extending the copula-graphic estimator of Heuchenne et al. (2014) and the conditional version of Braekers & Veraverbeke (2005) to the semi-competing-risks setting with nonparametric margins and a flexible dependence parameter is a natural and useful contribution. The careful simulation design (three DGMs, three families, three censoring rates, two n) and the explicit comparison of CGE versus CCGE1 versus CCGE2 provide clear, actionable guidance for practitioners: margin adjustment is essential, while a constant-copula working assumption is often sufficient for S1|X. The two real-data illustrations reinforce the same message. Although asymptotic theory is absent and only a single continuous covariate is handled, the finite-sample evidence is solid and the practical recommendations are immediately usable.

major comments (2)
  1. Section 2.1, equation (2) and the subsequent identifiability claim: the model assumes that the conditional copula belongs to a single known one-parameter Archimedean family for every value of the continuous covariate. The paper correctly notes that the pointwise extension of Heuchenne et al. (2014) guarantees identifiability under this assumption, but supplies no diagnostic or robustness check when the family is misspecified or changes with x. Because both the iterative algorithm and the resulting S1|X estimator rest on this assumption, a modest sensitivity study (or at least a clear statement of the risk) would strengthen the central claim.
  2. Section 2.4 and the paragraph preceding Algorithm 1: the authors acknowledge that establishing consistency of the iterative estimator is challenging and therefore rely entirely on finite-sample evidence. While the simulations (Tables 2, S3–S7) are thorough and the algorithm converges in 3–4 iterations in practice, the absence of even a sketch of asymptotic justification for the sequential procedure leaves the theoretical status of CCGE1/CCGE2 incomplete. A brief discussion of the obstacles and possible routes (e.g., via the theory of Z-estimators or empirical processes for Beran-type estimators) would be valuable.
minor comments (5)
  1. Table 1 (and the corresponding table in the supplement) lists generator functions and inverse-link functions; the main-text reference to “Table S2” is inconsistent with the numbering used in the body. A single, self-contained table in the main text would improve readability.
  2. Figure 1 and the analogous figures in the supplement display Kendall’s tau estimates; the caption states “5th and 95th quantiles” while the text sometimes refers to “90 % Monte Carlo confidence intervals.” Clarifying the exact coverage would avoid confusion.
  3. Section 4.1: the bandwidth pilot range (15–56) and the selected values (h1=37, h2=27, hC=56) are reported, but no sensitivity plot or leave-one-out CV criterion values are shown. A short remark on robustness to bandwidth choice would be helpful.
  4. The discussion (Section 5) correctly flags the single-covariate limitation and the need for formal copula-family selection. Adding a sentence that the heuristic log-likelihood comparison already recovers the true family in >80 % of the simulated scenarios (Table S8) would give readers a concrete sense of the practical reliability of that heuristic.
  5. Minor typographical inconsistencies appear (e.g., “copula graphic” vs. “copula-graphic”, “Statiscal Methods” in the references). A careful proof-reading pass is recommended.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the iterative conditional copula-graphic estimator and its finite-sample comparisons are constructed from independent nonparametric ingredients and external simulation truth.

full rationale

The paper defines the conditional joint survival via an Archimedean copula (eqs. 1–2), estimates the terminal-event margin by Beran’s estimator, plugs the copula into the Rivest–Wells-style copula-graphic formula for the non-terminal margin, and recovers the copula parameter by local pseudo-likelihood; the sequential algorithm simply alternates these two consistent steps until a fixed point. Identifiability is imported from the external result of Heuchenne et al. (2014) applied pointwise, not from any self-citation. Simulation truth (DGMs 1–3) is generated independently of the fitting procedure, and the reported ISB/IMSE comparisons therefore constitute genuine finite-sample evaluation rather than tautological recovery of fitted quantities. Real-data analyses likewise report estimates without feeding conclusions back into model definition. No step reduces by construction to its own inputs.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claims rest on standard survival-analysis and copula assumptions plus a handful of modeling choices (Archimedean family, single continuous covariate, local-linear calibration, cross-validated bandwidths). No new physical entities are postulated; free parameters are the usual nonparametric tuning constants.

free parameters (3)
  • bandwidths h1, h2, hC
    Selected by leave-one-out cross-validation; different pilot grids are used for the two real-data examples and affect the smoothness of both the survival curves and the Kendall-tau function.
  • Archimedean family choice (Clayton/Frank/Gumbel)
    Selected by comparing pseudo-log-likelihoods under a constant-parameter model; the subsequent analysis conditions on the chosen family.
  • inverse-link functions g^{-1}
    Fixed by convention for each family (Table 1) to map the unrestricted calibration function onto the copula-parameter domain.
assumptions (3)
  • domain assumption Conditional independence of the censoring time Z from (Y1,Y2) given the continuous covariate X
    Stated in Section 2.1; required for Beran's estimator of the terminal-event margin and for the joint model (1).
  • domain assumption The conditional copula belongs to a fixed one-parameter Archimedean family for every x, with generator satisfying the monotonicity condition of Heuchenne et al. (2014)
    Equation (2) and the paragraph immediately following; guarantees pointwise identifiability.
  • ad hoc to paper Local linear approximation of the calibration function η(x) is adequate inside the kernel window
    Used to construct the local pseudo-likelihood (7); standard in local-likelihood literature but not verified for the semi-competing-risks likelihood.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Conditional copula graphic estimator for semi-competing risks data." pith.science (2026). https://pith.science/paper/ACUYMHVQ

@misc{pith2026260709894,
  author       = {Pith},
  title        = {Pith review of: Conditional copula graphic estimator for semi-competing risks data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ACUYMHVQ}},
  note         = {Machine review of arXiv:2607.09894}
}
read the original abstract

In semi-competing risks data, the interest lies in the estimation of the survival function of a non-terminal event time, which is subject to dependent censoring by a terminal event. This problem has been extensively studied in the literature, but mostly focusing on unconditional settings. However, in many clinical applications incorporating covariates is necessary to control for confounding and improve survival function estimation. In this paper, we propose a conditional copula-graphic estimator that allows for covariate adjustment in the marginal survival functions of the non-terminal and terminal event times as well as in their dependence structure. The proposed estimator is semiparametric in that the conditional copula is specified parametrically using an Archimedean copula, but its dependence parameter function and margins are estimated nonparametrically. The estimator is obtained via a sequential iterative algorithm with alternating updates of the survival function of the non-terminal event and the conditional copula. The performance of the conditional copula-graphic estimator is assessed using simulated and real data, and is compared to that of the unconditional copula-graphic estimator to investigate the consequences of failing to account for covariate effects.

Figures

Figures reproduced from arXiv: 2607.09894 by the authors.

Figure 1
Figure 1. Mean, 5th and 95th quantiles of Kendall’s tau estimates under Frank copula with sample size n = 100. Data is generated from DGM 1(top row), DGM 2 (middle row) and DGM 3 (bottom row) with no (left column), low (middle column) and moderate (right column) censoring rates of non-terminal event. Dashed lines represent the Kendall’s tau estimates from CGE (red), CCGE1 (green), CCGE2 (blue); Dotted and solid lines represen… view at source ↗
Figure 2
Figure 2. Survival function of the heart transplant waiting time and Kendall’s tau [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Survival function of the heart transplant waiting time and Kendall’s tau esti [PITH_FULL_IMAGE:figures/full_fig_p016_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

34 extracted references

  1. [1]

    The analysis of semi‐competing risks data using Archimedean copula models , volume =

    Wang, Antai and Guo, Ziyan and Zhang, Yilong and Wu, Jihua , journal =. The analysis of semi‐competing risks data using Archimedean copula models , volume =

  2. [2]

    Penalised semi-parametric copula method for semi-competing risks data: application to hip fracture in elderly , volume =

    Sun, Tao and Liang, Weijie and Zhang, Gongzi and Yi, Danhui and Ding, Ying and Zhang, Lihai , journal =. Penalised semi-parametric copula method for semi-competing risks data: application to hip fracture in elderly , volume =

  3. [3]

    Bivariate copula regression models for semi-competing risks , volume =

    Wei, Y and Wojtys M and Sorrell, L and Rowe, P , journal =. Bivariate copula regression models for semi-competing risks , volume =

  4. [4]

    and Azimaee, Parisa and Hoque, E

    Acar, Elif F. and Azimaee, Parisa and Hoque, E. , date-added =. Predictive assessment of copula models , volume =. Canadian Journal of Statistics , number =

  5. [5]

    A reanalysis of the Stanford heart transplant data , volume =

    Aitkin, Murray and Laird, Nan and Francis, Brian , date-added =. A reanalysis of the Stanford heart transplant data , volume =. Journal of the American Statistical Association , number =

  6. [6]

    Survival analysis: techniques for censored and truncated data , year =

    Klein, John P and Moeschberger, Melvin L , date-added =. Survival analysis: techniques for censored and truncated data , year =

  7. [7]

    Cardiac transplantation in man: VI

    Clark, David A and Stinson, Edward B and Griepp, Randall B and Schroeder, John S and Shumway, Norman E and Harrison, Donald C , date-added =. Cardiac transplantation in man: VI. Prognosis of patients selected for cardiac transplantation , volume =. Annals of Internal Medicine , number =

  8. [8]

    Covariance analysis of heart transplant survival data , volume =

    Crowley, John and Hu, Marie , date-added =. Covariance analysis of heart transplant survival data , volume =. Journal of the American Statistical Association , number =

Show all 34 references
  1. [9]

    and Chimitova, E

    Demin, V. and Chimitova, E. , booktitle =. A method for selection of the optimal bandwidth parameter for Beran's nonparametric estimator , year =

  2. [10]

    and Bastert, G

    Schumacher, M. and Bastert, G. and Bojar, H. and Hiibner, K. and Olschewski, M. and Sauerbrei, W. and Schmoor, C. and Beyerle, C. and Neumann, R.L.A. and Rauschecker, H.F. for the German Breast Cancer Study Group (GBSG) , date-added =. A randomized 2 x 2 trial evaluating hormo...

  3. [11]

    Semiparametric estimation of conditional copulas , volume =

    Abegaz, Fentaw and Gijbels, Ir. Semiparametric estimation of conditional copulas , volume =. Journal of Multivariate Analysis , pages =

  4. [12]

    and Craiu, Radu V

    Acar, Elif F. and Craiu, Radu V. and Yao, Fang , date-added =. Dependence calibration in conditional copulas: A nonparametric approach , volume =. Biometrics , number =

  5. [13]

    Nonparametric regression with randomly censored survival data , year =

    Beran, Rudolf , institution =. Nonparametric regression with randomly censored survival data , year =

  6. [14]

    A copula-graphic estimator for the conditional survival function under dependent censoring , volume =

    Braekers, Roel and Veraverbeke, No. A copula-graphic estimator for the conditional survival function under dependent censoring , volume =. Canadian Journal of Statistics , number =

  7. [15]

    Maximum likelihood analysis of semicompeting risks data with semiparametric regression models , volume =

    Chen, Yi-Hau , journal =. Maximum likelihood analysis of semicompeting risks data with semiparametric regression models , volume =

  8. [16]

    Conditional copula models for right-censored clustered event time data , volume =

    Geerdens, Candida and Acar, Elif Fidan and Janssen, Paul , date-modified =. Conditional copula models for right-censored clustered event time data , volume =. Biostatistics , number =

  9. [17]

    On semi-competing risks data , volume =

    Fine, Jason P and Jiang, Hongyu and Chappell, Rick , journal =. On semi-competing risks data , volume =

  10. [18]

    Semiparametric inferences for association with semi-competing risks data , volume =

    Ghosh, Debashis , journal =. Semiparametric inferences for association with semi-competing risks data , volume =

  11. [19]

    Semi-competing risks data analysis: accounting for death as a competing risk when the outcome of interest is nonterminal , volume =

    Haneuse, Sebastien and Lee, Kyu Ha , journal =. Semi-competing risks data analysis: accounting for death as a competing risk when the outcome of interest is nonterminal , volume =

  12. [20]

    Likelihood-Based Inference for Semi-Competing Risks , volume =

    Heuchenne, C. Likelihood-Based Inference for Semi-Competing Risks , volume =. Communications in Statistics-Simulation and Computation , number =

  13. [21]

    Regression analysis based on conditional likelihood approach under semi-competing risks data , volume =

    Hsieh, Jin-Jian and Huang, Yu-Ting , journal =. Regression analysis based on conditional likelihood approach under semi-competing risks data , volume =

  14. [22]

    Regression analysis based on semicompeting risks data , volume =

    Hsieh, Jin-Jian and Wang, Weijing and Adam Ding, A , journal =. Regression analysis based on semicompeting risks data , volume =

  15. [23]

    Estimating survival and association in a semicompeting risks model , volume =

    Lakhal, Lajmi and Rivest, Louis-Paul and Abdous, Belkacem , journal =. Estimating survival and association in a semicompeting risks model , volume =

  16. [24]

    Estimating the survival functions in a censored semi-competing risks model , volume =

    Laurent, St. Estimating the survival functions in a censored semi-competing risks model , volume =. Sankhya A , number =

  17. [25]

    Quantile regression adjusting for dependent censoring from semicompeting risks , volume =

    Li, Ruosha and Peng, Limin , journal =. Quantile regression adjusting for dependent censoring from semicompeting risks , volume =

  18. [26]

    Regression modeling of semicompeting risks data , volume =

    Peng, Limin and Fine, Jason P , journal =. Regression modeling of semicompeting risks data , volume =

  19. [27]

    A martingale approach to the copula-graphic estimator for the survival function under dependent censoring , volume =

    Rivest, Louis-Paul and Wells, Martin T , journal =. A martingale approach to the copula-graphic estimator for the survival function under dependent censoring , volume =

  20. [28]

    Estimating the association parameter for copula models under dependent censoring , volume =

    Wang, Weijing , journal =. Estimating the association parameter for copula models under dependent censoring , volume =

  21. [29]

    Statistical analysis of illness--death processes and semicompeting risks data , volume =

    Xu, Jinfeng and Kalbfleisch, John D and Tai, Beechoo , journal =. Statistical analysis of illness--death processes and semicompeting risks data , volume =

  22. [30]

    A new flexible dependence measure for semi-competing risks , volume =

    Yang, Jing and Peng, Limin , journal =. A new flexible dependence measure for semi-competing risks , volume =

  23. [31]

    Combining parametric, semi-parametric, and non-parametric survival models with stacked survival models , volume =

    Wey, Andrew and Connett, John and Rudser, Kyle , journal =. Combining parametric, semi-parametric, and non-parametric survival models with stacked survival models , volume =

  24. [32]

    Estimates of marginal survival for dependent competing risks based on an assumed copula , volume =

    Zheng, Ming and Klein, John P , journal =. Estimates of marginal survival for dependent competing risks based on an assumed copula , volume =

  25. [33]

    Simplifying a prognostic model: a simulation study based on clinical data , volume =

    Ambler, Gareth and Brady, Anthony R and Royston, Patrick , journal =. Simplifying a prognostic model: a simulation study based on clinical data , volume =

  26. [34]

    Getting more out of survival data by using the hazard function , year =

    Hess, Kenneth R and Levin, Victor A , journal =. Getting more out of survival data by using the hazard function , year =

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.