Distributionally Robust Learning in Survival Analysis

Ioannis Ch. Paschalidis; Lauren Wise; Yeping Jin

REVIEW 5 major objections 5 minor 18 references

Distributionally Robust Learning in Survival Analysis

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A distributionally robust Cox model reduces to an exponential conic program and outperforms standard survival models under distributional shift and outliers.

desk verdict Plausible DRL-Cox extension, but the implemented constraint truncation breaks the theoretical guarantees, and the surrogate loss is not the Cox partial likelihood. read the letter →

arxiv 2506.01348 v2 pith:TTRU77YK submitted 2025-06-02 cs.LG math.OCstat.ML

Yeping Jin , Lauren Wise , Ioannis Ch. Paschalidis This is my paper · ORCID

classification cs.LGmath.OCstat.ML

keywords distributionallyrobustoptimizationCoxproportionalhazardsmodelsurvivalanalysisWassersteindistanceambiguitysetexponentialconicprogrammingfinite-sampleguaranteecensoreddata

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

This paper sets out to establish that Cox regression can be made robust to distributional shift and outliers by optimizing the worst-case loss over a Wasserstein ball of distributions around the training data. To make that min-max problem tractable, it replaces the per-subject Cox loss with a surrogate that adds an extra exponential term, so the loss becomes convex in the covariates and nonnegative for long-duration subjects. The main theorem converts the resulting robust problem into an exponential conic program that is an upper bound on the original min-max value and is exact when the feature space is all of $\mathbb{R}^d$. A finite-sample guarantee says that with an ambiguity radius $\epsilon$ chosen like $\sqrt{\log(1/\alpha)/N}$, the true expected loss stays below the robust training objective with probability at least $1-\alpha$. If these claims are correct, survival models could retain their accuracy when the data used for training and the data seen in practice come from different distributions.

What carries the argument

The central object is the Wasserstein-distance ambiguity set $\Omega_\epsilon = \{P : W_s(P, \hat{P}_N) \le \epsilon\}$ centered at the empirical distribution, coupled with the surrogate individual loss (3) that inserts an extra $e^{\beta'x}$ term into the log and treats all training subjects as fixed constants. That insertion makes the loss convex in the covariate direction and nonnegative for long-duration subjects, which is what allows Wasserstein duality, the dual norm, and the convex conjugate to convert the min-max problem into the exponential conic program (5). The proof also discretizes the duration coordinate, whose piecewise-constant, non-convex behavior blocks a direct duality argument.

What would settle it

On a small synthetic dataset with a bounded feature space, evaluate both sides of Theorem 2 by solving the min-max problem (4) with a fine grid search and solving the conic program (5); a strict gap between the two values would show the upper bound is loose in exactly the bounded-feature regime where the real-data experiments are run.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the distributionally robust Cox problem (4), built on the redefined individual loss (3), has a tractable convex reformulation: the exponential conic program (5), whose objective is $\epsilon\|(\beta,\alpha)\|_q$ plus an averaged weighted log-sum-exp loss over subjects, with $O(N^2)$ constraints relating the slack variables $s_i$ to risk-set sums. The proof passes through Wasserstein strong duality, discretizes the non-convex duration direction, and evaluates the convex conjugate of the surrogate loss, which pins down the penalty coefficient as $\|\beta\|_q$. The relaxation is tight when the covariate space is exactly $\mathbb{R}^d$. The paper further claims a finite-sample robustness guarantee, and reports that in simulations and on two healthcare datasets the model consistently beats standard Cox, penalized Cox, accelerated failure time, and random survival forest baselines in C-index and time-dependent AUC under contamination.

Load-bearing premise

The load-bearing premise is that the redefined individual loss in Eq. (3), which adds an extra exponential term and treats all training subjects as fixed constants, is an acceptable surrogate for the Cox partial likelihood; the paper itself acknowledges in Section 2.3 that this design deviates from the standard Cox loss, and if the surrogate is not faithful, the theoretical bound and the name 'Cox' do not transfer to the standard model.

Editorial extensions

If this is right

The robust Cox objective is solvable with exponential conic programming, and its $O(N^2)$ constraint set can be truncated to $O(\gamma N)$ constraints; the paper uses $\gamma=3$ and reports about thirty seconds per trial, so the method is practical on moderate datasets.
Choosing the ambiguity radius $\epsilon$ at or above $B(\alpha)$, which scales as $\sqrt{\log(1/\alpha)/N}$, yields a finite-sample guarantee that the true expected loss is bounded by the empirical robust objective with probability at least $1-\alpha$.
If the surrogate loss is accepted as the objective, DRL-Cox should keep its ranking accuracy when covariates shift or outliers contaminate features, and the experiments consistently show C-index and iAUC gains over regular and penalized Cox, sample-splitting Cox, AFT, and random survival forests.
Because the derivation needs convexity in the covariate direction, the same route does not directly extend to deep survival models; the authors point to constraining weight matrices as a possible adaptation.

Reading between the lines

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

The paper's guarantee and the conic reformulation are stated for the surrogate loss (3), not for the standard Cox partial likelihood (1); if a reader cares about the classical Cox objective, the connection is asserted by analogy rather than proven.
Because the conic program is exact only when the feature space is all of $\mathbb{R}^d$, on bounded feature spaces the program is an upper bound and the reported gains may partly reflect that looseness.
A clean-data comparison of the surrogate loss against standard Cox would isolate the robustness contribution from the regularizing effect of the extra exponential term and the $\epsilon\|(\beta,\alpha)\|_q$ penalty.
The same duality-and-discretization template could be applied to other losses that are convex in features but non-convex in one coordinate, such as ranking or quantile objectives.

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.

Desk Editor's Note

read the letter

I'll be direct: the paper's main idea is worth a look—Wasserstein DRL applied to a Cox-type loss with an exponential conic reformulation and a finite-sample bound is a natural combination that, to my knowledge, hasn't been done in survival analysis. The authors are honest about modifying the individual loss (Eq. 3) and they provide code and data for reproducibility. The empirical comparison includes reasonable baselines.

The soft spots are real. Most serious is the constraint truncation in Section 4. Equation (5) has O(N^2) constraints; the experiments use only k < i + 3 (gamma=3). Dropping inequality constraints in a minimization relaxes the feasible set, so the tested program can have a lower optimal value than (5). Theorem 2's upper bound and Lemma 5's finite-sample guarantee apply to (5), not to the relaxed variant. The one-sentence assertion that truncation 'maintains the integrity' is unsupported; under distributional shift the active maximum could easily lie beyond i+2. Until the authors either solve the full program for moderate N or prove that the truncated constraint set preserves the bound, the theoretical claims don't cover the reported results.

Second, the surrogate loss deviates from the Cox partial likelihood. The authors disclose this, which helps, but it means the theory and the name 'Cox' apply to a modified objective. The practical impact may be small, but it should be stated more prominently.

Third, the proof of Theorem 2 has gaps: the minimax interchange in (16b) is invoked without checking the required compactness or convexity, the X=R^d equality is asserted rather than derived, and Lemma 5 is imported without verification. The result is likely true, but the paper doesn't fully demonstrate it.

Finally, the empirical claims are overstated. The reported gains over regular Cox are about 1–2% absolute C-index, not the 'more than 5%' in the introduction unless one compares against the weakest baseline. No error bars or standard deviations are given, and the hyperparameter selection for epsilon is described loosely.

Overall, the core idea has merit and the authors are transparent about their choices, but the gap between theory and implementation is load-bearing. I would send it to peer review because the combination is new and the authors can likely address the gamma issue, but I would not use it as-is as a basis for my own work. If the full program is intractable, they need a rigorous argument that the truncation is harmless, or the guarantees must be re-proven for the truncated set.

Referee Report

5 major / 5 minor

Summary. The paper proposes DRL-Cox, a distributionally robust version of Cox regression. It replaces the standard Cox partial-likelihood loss with a modified individual loss (Eq. 3), embeds this loss in a Wasserstein-ambiguity min-max problem (Eq. 4), and derives an exponential conic program (Eq. 5) claimed to be an upper bound on the robust loss, with equality claimed when the covariate space is R^d. The paper also states a finite-sample performance guarantee (Lemma 5). Experiments on the WHAS500 dataset and the PRESTO miscarriage cohort compare DRL-Cox with regular and penalized Cox models, AFT, and RSF under covariate shifts and outliers, reporting higher C-index and iAUC for DRL-Cox. Code and partial data are made available.

Significance. If the derivation and finite-sample guarantee were fully rigorous, the paper would offer a useful tractable formulation for robust survival analysis and a principled alternative to sample-splitting DRL approaches. The empirical setup is relevant, and the authors provide code and use both a public benchmark and a large cohort. However, the current proof has unverified minimax steps, the implemented model differs from the analyzed program by truncating constraints, and the loss is a surrogate for the Cox partial likelihood; these gaps currently limit the strength of the central claims.

major comments (5)

[Sec. 3.1, Eq. (16b)] The step from Eq. (16a) to Eq. (16b) exchanges sup over theta and inf over (r, alpha) before taking sup over x, and the later step immediately before Eq. (19) exchanges max over j and inf over (theta, alpha). Both steps are minimax interchanges, but the paper invokes the Minimax Theorem without checking the compactness, convexity, or saddle-point conditions for these specific functions, which are not shown to be jointly convex-concave in the relevant variables. Because these interchanges are the core of the derivation of Eq. (5), Theorem 2 is not fully established as written.
[Sec. 3.1, Theorem 2] The claim that Eq. (5) is an exact reformulation of Eq. (4) when X = R^d is asserted but not proved. The argument shows that one inequality (Eq. (16c)) is tight when X = R^d, but inequalities remain in Eqs. (18) and (19), and no argument demonstrates that those become equalities on R^d. The exactness claim should either be proved with the missing steps or removed and replaced by a statement of the conditions under which equality holds.
[Sec. 4, constraint truncation] The experiments solve a gamma = 3 truncated version of Eq. (5) that keeps only constraints with k < i + gamma. Removing inequality constraints from a minimization relaxes the feasible set, so the truncated program's optimal value can be strictly smaller than the optimal value of Eq. (5). The upper-bound statement in Theorem 2 and the finite-sample guarantee in Lemma 5 apply to Eq. (5), not to this truncated model, and the unsupported assertion that the truncation 'maintains the integrity of the optimization process' does not address the gap. The reported improvements could therefore be artifacts of the extra relaxation rather than of the Wasserstein DRL construction.
[Sec. 2.3, Eq. (3)] The individual loss in Eq. (3) is not the Cox partial-likelihood loss; it inserts an additional e^{beta'x} term inside the log. All theoretical results and experiments use this modified loss, so the finite-sample guarantee and robustness claims pertain to a surrogate objective rather than to the standard Cox model. The statement that this modification 'preserves theoretical consistency with the proportional hazards assumption' is not substantiated, and the manuscript should either justify the surrogate as a principled loss or soften the claims connecting it to standard Cox regression.
[Sec. 3.2, Lemma 5 and Sec. 4, choice of epsilon] Lemma 5 is quoted from Zhao and Guan without proof and assumes a bounded and 'finite' data space, which is ambiguous. The guarantee requires epsilon >= B(alpha), but in the experiments epsilon is selected by cross-validation and a concentration method, with no demonstration that the selected radii satisfy the lemma's threshold. Thus the reported empirical results are not covered by the stated finite-sample guarantee.

minor comments (5)

[Sec. 3.1, Eq. (5)] The constraint in Eq. (5) uses the index i both for the constraint and for the summation index inside the log; a distinct summation index, such as sum_{m=1}^k, would remove the ambiguity.
[Sec. 4, Tables 2 and 3] Tables 2 and 3 report averages over five iterations without confidence intervals or significance tests, making it difficult to assess whether the reported advantages of DRL-Cox are meaningful.
[Sec. 4.1, Figure 1] Figure 1 would benefit from error bars or confidence bands, especially since the text emphasizes non-monotonic behavior across shift levels.
[Sec. 3.2, Lemma 5] The phrase 'the data space X x Y is bounded and finite' is ambiguous: boundedness and finiteness are different conditions, and the lemma should specify which is intended.
[Sec. 3.2] The text refers to the 'training loss c_hat_N' but only J_hat_N is defined; the notation should be introduced consistently before it is used.

Circularity Check

0 steps flagged · score 1.0 of 10

No material circularity: the DRL-Cox reformulation and finite-sample bound rely on external duality and concentration results; the modified Cox loss and the gamma=3 truncation are explicit modeling and implementation choices, not inputs disguised as predictions.

full rationale

The central derivation is not circular. Theorem 2's exponential conic reformulation is obtained from the Wasserstein dual of Gao and Kleywegt (Cor. 3) and a convex-analysis argument; the only self-cited ingredient is the ambiguity-set framing from Chen and Paschalidis (2020), which is not the result being proved and is corroborated by independent DRL literature (Esfahani-Kuhn, Gao-Kleywegt). Lemma 5's finite-sample guarantee imports its concentration bound from Zhao and Guan (Prop. 3); once the true distribution P* lies in the Wasserstein ball, the inequality E_{P*}[l] <= \hat{J}_N is definitional from (2), so the statistical content is external, not fitted. The paper explicitly concedes that Eq. (3) 'deviates from the standard Cox loss' and that the testable model truncates the O(N^2) constraints in (5) to k < i+gamma with gamma=3; both are transparent validity and implementation gaps rather than circular steps. Self-citations to the same group's DRL framework and PRESTO data are not load-bearing for the main duality and concentration claims. Score 1 reflects minor self-citation presence, not circularity.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The model relies on standard Wasserstein DRL machinery plus an ad hoc loss modification and a constraint-window truncation. The main free parameters are the ambiguity radius epsilon and the window size gamma. No new physical or probabilistic entities are introduced.

free parameters (2)

Ambiguity radius epsilon = Not reported; selected by cross-validation and concentration bounds
Controls the size of the Wasserstein ball around the empirical distribution; the paper tunes it on validation sets, so the reported robustness depends on this choice.
Constraint window gamma = 3
Truncates the O(N^2) constraints in Eq. (5) to O(gamma N); this changes the optimization problem and no sensitivity analysis is provided.

assumptions (5)

standard math Wasserstein strong duality for empirical-measure-centered ambiguity sets (Gao and Kleywegt 2023, Cor. 2).
Used in Section 3.1 to convert the inner supremum in Eq. (6) into a finite-dimensional dual problem.
ad hoc to paper The individual loss (3) is a valid surrogate for the Cox partial likelihood.
Introduced in Section 2.3 to make the loss depend on single data points; the paper acknowledges it deviates from the standard Cox loss.
domain assumption Data points are sorted by decreasing duration and the loss is monotone decreasing in y, so the supremum over y can be restricted to observed durations.
Used around Eq. (10) to discretize the non-convex y direction in the proof of Theorem 2.
ad hoc to paper The minimax interchange in Eq. (16b) is valid, including compactness and convexity conditions.
The paper invokes von Neumann's minimax theorem without verifying the required conditions; for the relaxation an inequality would suffice, but the equality claim for X equal to R^d depends on this step.
domain assumption The data space X times Y is bounded and finite for the finite-sample guarantee.
Lemma 5 in Section 3.2 is quoted from Zhao and Guan and requires bounded or finite data space; the paper does not verify this for WHAS500 or PRESTO.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Distributionally Robust Learning in Survival Analysis." pith.science (2026). https://pith.science/paper/TTRU77YK

@misc{pith2026250601348,
  author       = {Pith},
  title        = {Pith review of: Distributionally Robust Learning in Survival Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TTRU77YK}},
  note         = {Machine review of arXiv:2506.01348}
}

read the original abstract

We introduce an innovative approach that incorporates a Distributionally Robust Learning (DRL) approach into Cox regression to enhance the robustness and accuracy of survival predictions. By formulating a DRL framework with a Wasserstein distance-based ambiguity set, we develop a variant Cox model that is less sensitive to assumptions about the underlying data distribution and more resilient to model misspecification and data perturbations. By leveraging Wasserstein duality, we reformulate the original min-max DRL problem into a tractable regularized empirical risk minimization problem, which can be computed by exponential conic programming. We provide guarantees on the finite sample behavior of our DRL-Cox model. Moreover, through extensive simulations and real world case studies, we demonstrate that our regression model achieves superior performance in terms of prediction accuracy and robustness compared with traditional methods.

Figures

Figures reproduced from arXiv: 2506.01348 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

18 extracted references · 8 canonical work pages

[7]

10 Distributionally Robust Learning in Survival Analysis Robert J

doi: 10.1287/opre.1090.0795. 10 Distributionally Robust Learning in Survival Analysis Robert J. Goldberg, Jorge Yarzebski, Darleen Lessard, and Joel M. Gore. Decade-long trends and fac- tors associated with time to hospital presentation in patients with acute myocardial infarction: The Worcester heart attack study. Archives of Inter- nal Medicine , 160(21...

arXiv
[13]

doi: 10.1080/01621459.2020.1796359. Noah R. Simon, Jerome H. Friedman, Trevor J. Hastie, and Robert Tibshirani. Regularization paths for Cox’s proportional hazards model via coordinate descent. Journal of statistical software , 39 5:1–13,

arXiv 2020
[14]

doi: 10.1371/journal.pone. 0283574. Robert Tibshirani. The LASSO method for variable selection in the Cox model. Statistics in medicine , 16 4:385–95,

doi:10.1371/journal.pone
[1972]

doi: 10.1111/j.2517-6161.1972.tb00899

ISSN 0035-9246. doi: 10.1111/j.2517-6161.1972.tb00899. x. Erick Delage and Yinyu Ye. Distributionally robust op- timization under moment uncertainty with applica- tion to data-driven problems. Operations Research, 58:595–612, 06

arXiv 1972
[1982]

doi: 10.1001/jama.1982

ISSN 0098-7484. doi: 10.1001/jama.1982. 03320430047030. Patrick J. Heagerty and Yingye Zheng. Survival model predictive accuracy and ROC curves. Biometrics, 61(1):92–105,

work page doi:10.1001/jama.1982 1982
[2000]

doi: 10.1001/archinte.160.21.3217

ISSN 0003-9926. doi: 10.1001/archinte.160.21.3217. Boran Hao, Guoyao Shen, Ruidi Chen, Chad W. Far- ris, Stephan W. Anderson, Xin Zhang, and Ioan- nis Ch. Paschalidis. Distributionally robust image classifiers for stroke diagnosis in accelerated MRI. In Medical Image Computing and Computer As- sisted Intervention – MICCAI 2023 , pages 768–777, Cham,

work page doi:10.1001/archinte.160.21.3217
[2005]

030814.x

doi: 10.1111/j.0006-341X.2005. 030814.x. Shu Hu and George H. Chen. Distributionally ro- bust survival analysis: A novel fairness loss without demographics. In Machine Learning for Health, ML4H 2022 , volume 193, pages 62–87,

work page doi:10.1111/j.0006-341x.2005 2005
[2008]

doi: 10.1214/08-AOAS169. J. D. Kalbfleisch and Ross L. Prentice. The statistical analysis of failure time data . Wiley series in proba- bility and statistics. J. Wiley, Hoboken, N.J, 2nd ed edition,

doi:10.1214/08-aoas169

Show all 18 references

[2009]

doi: 10.1287/opre.1080.0684. 12
[2010]

David Donoho

doi: 10.1287/opre.1090.0741. David Donoho. High-dimensional data analysis: The curses and blessings of dimensionality. AMS Math Challenges Lecture, pages 1–32, 01
[2012]

Joel Goh and Melvyn Sim

doi: 10.2139/ssrn.2167765. Joel Goh and Melvyn Sim. Distributionally ro- bust optimization and its tractable approximations. Operations Research, 58:902–917, 08

doi
[2015]

doi: 10.1111/ppe.12201. J. Yland, Wang, Taiyao, Zahra Zad, Sydney K Willis, Tanran R Wang, Amelia K Wesselink, Tammy Jiang, Elizabeth E Hatch, Lauren A Wise, and Ioan- nis Ch Paschalidis. Predictive models of pregnancy based on data from a preconception cohort study. Human Rep...

doi
[2018]

v32i1.11842

doi: 10.1609/aaai. v32i1.11842. Mo, Weibin, Zhengling Qi, and Yufeng Liu. Learn- ing optimal distributionally robust individualized treatment rules. Journal of the American Sta- tistical Association, 116(534):659–674,

doi
[2020]

Ruidi Chen, Boran Hao, and Ioannis Ch

doi: 10.1561/2400000026. Ruidi Chen, Boran Hao, and Ioannis Ch. Paschalidis. Distributionally robust multiclass classification and applications in deep image classifiers. IEEE Interna- tional Conference on Acoustics, Speech, and Signal Processing (ICASSP). doi: 10.1109/ICASSP4...
[2021]

Peyman Mohajerin Esfahani and Daniel Kuhn

doi: 10.1214/20-AOS2004. Peyman Mohajerin Esfahani and Daniel Kuhn. Data- driven distributionally robust optimization us- ing the Wasserstein metric: performance guar- antees and tractable reformulations. Math. Pro- gram., 171(1-2):115–166,

doi
[2022]

doi: 10.1093/humrep/deab280

ISSN 0268-1161. doi: 10.1093/humrep/deab280. J. Yland, Zahra Zad, Tanran R. Wang, Amelia K. Wesselink, Tammy Jiang, Elizabeth E. Hatch, Ioan- nis Ch. Paschalidis, and Lauren A. Wise. Predictive models of miscarriage on the basis of data from a preconception cohort study. Ferti...

doi
[2023]

Paul Glasserman and Xingbo Xu

doi: 10.1287/MOOR.2022.1275. Paul Glasserman and Xingbo Xu. Robust risk mea- surement and model risk. Available at SSRN 2167765, 14, 1

2022
[2024]

doi: 10.1016/j.fertnstert.2024.04.007

ISSN 0015-0282. doi: 10.1016/j.fertnstert.2024.04.007. 11 Distributionally Robust Learning in Survival Analysis Hao Helen Zhang and Wenbin Lu. Adaptive Lasso for Cox’s proportional hazards model. Biometrika, 94(3):691–703, 05

2024 doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [7]

10 Distributionally Robust Learning in Survival Analysis Robert J

doi: 10.1287/opre.1090.0795. 10 Distributionally Robust Learning in Survival Analysis Robert J. Goldberg, Jorge Yarzebski, Darleen Lessard, and Joel M. Gore. Decade-long trends and fac- tors associated with time to hospital presentation in patients with acute myocardial infarction: The Worcester heart attack study. Archives of Inter- nal Medicine , 160(21...

arXiv

[2] [13]

doi: 10.1080/01621459.2020.1796359. Noah R. Simon, Jerome H. Friedman, Trevor J. Hastie, and Robert Tibshirani. Regularization paths for Cox’s proportional hazards model via coordinate descent. Journal of statistical software , 39 5:1–13,

arXiv 2020

[3] [14]

doi: 10.1371/journal.pone. 0283574. Robert Tibshirani. The LASSO method for variable selection in the Cox model. Statistics in medicine , 16 4:385–95,

doi:10.1371/journal.pone

[4] [1972]

doi: 10.1111/j.2517-6161.1972.tb00899

ISSN 0035-9246. doi: 10.1111/j.2517-6161.1972.tb00899. x. Erick Delage and Yinyu Ye. Distributionally robust op- timization under moment uncertainty with applica- tion to data-driven problems. Operations Research, 58:595–612, 06

arXiv 1972

[5] [1982]

doi: 10.1001/jama.1982

ISSN 0098-7484. doi: 10.1001/jama.1982. 03320430047030. Patrick J. Heagerty and Yingye Zheng. Survival model predictive accuracy and ROC curves. Biometrics, 61(1):92–105,

work page doi:10.1001/jama.1982 1982

[6] [2000]

doi: 10.1001/archinte.160.21.3217

ISSN 0003-9926. doi: 10.1001/archinte.160.21.3217. Boran Hao, Guoyao Shen, Ruidi Chen, Chad W. Far- ris, Stephan W. Anderson, Xin Zhang, and Ioan- nis Ch. Paschalidis. Distributionally robust image classifiers for stroke diagnosis in accelerated MRI. In Medical Image Computing and Computer As- sisted Intervention – MICCAI 2023 , pages 768–777, Cham,

work page doi:10.1001/archinte.160.21.3217

[7] [2005]

030814.x

doi: 10.1111/j.0006-341X.2005. 030814.x. Shu Hu and George H. Chen. Distributionally ro- bust survival analysis: A novel fairness loss without demographics. In Machine Learning for Health, ML4H 2022 , volume 193, pages 62–87,

work page doi:10.1111/j.0006-341x.2005 2005

[8] [2008]

doi: 10.1214/08-AOAS169. J. D. Kalbfleisch and Ross L. Prentice. The statistical analysis of failure time data . Wiley series in proba- bility and statistics. J. Wiley, Hoboken, N.J, 2nd ed edition,

doi:10.1214/08-aoas169