Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Targeted Data Fusion for Region-Specific Survival Effects in the AMP HIV Prevention Trials

T0 review · 3 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Two multi-source estimators for target-site survival curves—one efficient when data can be pooled, one privacy-preserving and adaptive—are shown to match or improve on target-only analysis.

desk verdict Useful federated survival estimator with a real theorem gap: the oracle set in Theorem 2.12 is defined by marginal equality, but the local estimators' consistency demands full CCOD, so the strict efficiency gain is not actually established. read the letter →

arxiv 2501.18798 v4 pith:S65266KG submitted 2025-01-30 stat.ME math.STstat.MLstat.TH

Yi Liu , Alexander W. Levis , Ke Zhu , Shu Yang , Peter B. Gilbert , Larry Han This is my paper · ORCID
classification stat.MEmath.STstat.MLstat.TH MSC 62D0562N0162G0562G20
keywords federatedlearningcausalsurvivalanalysisdatafusionefficientinfluencefunctiondistributionshiftprivacy-preservinginferenceAMPHIVpreventiontrialsdoublyrobustestimation
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 develops two estimators for a treatment-specific survival curve in a target population when data come from several sites with differing covariate, outcome, or censoring distributions. The first assumes a common conditional outcome distribution across sites and is shown to be semiparametrically efficient. The second is a federated estimator that, without sharing individual-level data, reweights site-specific influence-function estimates through an L1-regularized objective that downweights sources whose estimates disagree with the target. Simulations and an analysis of the AMP HIV prevention trials show the federated estimator maintains nominal coverage while achieving modest efficiency gains over target-only analysis. If correct, the methods make multi-site, privacy-preserving causal survival analysis practical without sacrificing statistical efficiency.

What carries the argument

The machinery is the efficient influence function (EIF) for the target-site survival estimand expressed with product-integral notation, so that continuous and discrete event times are handled uniformly. The federated estimator builds site-specific EIFs under a working 'partial CCOD' assumption, corrects for covariate shift with a density ratio $\omega_{k,0}(X)=P(X\mid R=0)/P(X\mid R=k)$, and then aggregates them with weights $\eta_{t,a}$ obtained from a convex L1-penalized objective that penalizes sites whose estimator differs from the target-only estimator (the discrepancy $\hat{\chi}^{k,0}_{n,t,a} = \hat{\theta}^{k,0}(t,a) - \hat{\theta}^0(t,a)$). Cross-fitting with ensemble machine learning supplies the nuisance functions, and isotonic regression enforces monotone survival curves.

What would settle it

Construct a two-site experiment with covariate shift in which the sites share only summary statistics that cannot identify the density ratio $\omega_{k,0}(X)$ (for example, sharing only marginal means while the true ratio depends on higher moments), run the federated algorithm, and check whether the estimator remains unbiased for $\theta_0(t,a)$; if it converges to a biased limit, the Theorem 2.12 guarantee fails in that regime.

Watch

Extended reading notes

Core claim

The paper's central claim is that target-site, treatment-specific survival functions can be estimated by borrowing information from other sites even when outcome distributions differ across sites, as long as the aggregation weights are chosen data-adaptively. The federated estimator $\hat{\theta}^{\mathrm{fed}}_n(t,a)$ is asymptotically normal around the target survival function $\theta_0(t,a)$ with asymptotic variance no greater than that of the target-only estimator, and strictly smaller when any source site provides a consistent estimate of $\theta_0(t,a)$ (Theorem 2.12). For the setting where data can be pooled and a common conditional outcome distribution holds, the paper's CCOD estimator is uniformly regular asymptotically linear and attains the semiparametric efficiency bound (Theorem 2.8). Both claims are supported by simulation and by re-analysis of the AMP trials, where region-specific weights track observed covariate and outcome shifts.

Load-bearing premise

The method's privacy-preserving validity depends on the covariate density ratio between each source and the target being consistently estimable from the coarse summary statistics that sites are allowed to share, a capability that is assumed from prior work rather than verified here.

Editorial extensions

If this is right

  • Multi-regional trials can report region-specific survival curves and causal contrasts (risk difference, survival ratio, restricted mean survival time difference) without pooling participant-level data.
  • When an informative source site is present, the federated estimator's asymptotic variance is strictly smaller than target-only, so the method converts source-site similarity into precision.
  • When the CCOD assumption holds and data sharing is allowed, the CCOD estimator reaches the semiparametric efficiency bound, making it a benchmark for pooled survival analysis under covariate shift.
  • The federated weights provide a time- and treatment-specific diagnostic of which sites align with the target, down-weighting sites that deviate.
  • The estimators inherit double robustness: consistency of the site-specific estimator holds if either the conditional survival model or the other nuisance functions (censoring, propensity, density ratio) are correctly specified.

Reading between the lines

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

  • The core idea—downweight sources by their discrepancy in the estimand space rather than in covariate space—should transfer to other missing-data and generalization problems, such as external control borrowing in regulatory submissions.
  • The theoretical guarantees hinge on the summary statistics being rich enough to identify the density ratio; a natural test is to compare federated results against pooled results when data sharing is actually allowed, treating the gap as a measure of information loss from the privacy model.
  • The framework's time-varying weights suggest a dynamic interpretation of 'informative site' that could be exploited for adaptive trial designs that enroll sites based on early estimated weights.
  • Combining the FED weighting with the CCOD estimator might yield an estimator that is efficient under CCOD yet robust to its violation, a hybrid the paper does not explicitly construct.
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

3 major / 4 minor

Summary. The manuscript proposes two estimators for target-site treatment-specific survival curves in multi-site right-censored data. The CCOD estimator assumes a common conditional outcome distribution across sites, is based on a derived semiparametric efficient influence function, and is shown to be uniformly asymptotically linear. The federated estimator uses site-specific EIF-based estimators under a working partial CCOD assumption, aggregates them with L1-regularized adaptive weights, and claims asymptotic normality around the target survival function with variance no larger than target-only estimation, strictly smaller under certain informative-source conditions. The paper reports simulations across five heterogeneity scenarios and applies the methods to AMP trial data. Appendices contain the technical proofs and algorithm details.

Significance. The CCOD part is a solid contribution: the EIF derivation and the uniform RAL proof are detailed, the nuisance estimation strategy is sensible, and the simulations match the theoretical claims. The federated part addresses an important and timely problem, privacy-preserving borrowing across heterogeneous sites, and the empirical results show near-nominal coverage with moderate RMSE reductions. The paper is not circular: the target survival curve is the estimand, not an input to the derivations. However, the main advertised efficiency-gain theorem relies on unstated and imported oracle conditions, and one key consistency implication in Appendix D.3 conflates marginal survival equality with the stronger conditional outcome equality needed by Theorem D.7. These issues are localized to the federated theory but are load-bearing for the paper's central claim.

major comments (3)
  1. [Section 2.3.3 and Appendix D.3] The proof of Theorem 2.12 defines the oracle selection space as S*_{t,a} = {k : theta_k(t,a) = theta_0(t,a)} and asserts that for k in S* the site-specific estimator btheta^{k,0}_n(t,a) is consistent for theta_0(t,a). However, the only local consistency result, Theorem D.7, explicitly assumes 'the CCOD holds' in addition to Conditions D.6-D.8. Marginal equality of survival functions at a single time point is strictly weaker than S_k(·|a,x) = S_0(·|a,x), which is what Theorem D.7 requires. In the outcome-shift settings the method is designed to handle, a source site can have theta_k(t,a) = theta_0(t,a) while the conditional outcome distributions differ, in which case btheta^{k,0}_n(t,a) is generally biased for theta_0(t,a). Thus the oracle set in Appendix D.3 does not imply the claimed consistency, and the sufficient condition for strict variance reduction in Theorem 2.12 is not established. The authors should either define S* in terms of consistency of btheta^{k,0}_n itself, or add assumptions that link theta_k = theta_0 to the conditional equality required by Theorem D.7.
  2. [Section 2.3.1, Algorithm 2, Condition D.6(b)] The federated estimator's local influence functions require the density ratio omega_{k,0}(X) = P(X|R=0)/P(X|R=k), and the privacy-preserving implementation assumes this ratio is consistently estimable from coarse summary statistics shared under the privacy model (Algorithm 2, step 14). Condition D.6(b) only postulates consistency of the density ratio estimator; the paper does not state conditions under which the allowed coarsened summaries are sufficient. This is load-bearing because if omega_{k,0} cannot be evaluated consistently, the local influence functions cannot be computed and the federated estimator does not have the claimed asymptotic properties. The authors should either state density-ratio consistency as an explicit high-level assumption with conditions, or provide a theorem describing which summary statistics suffice and what convergence rates result.
  3. [Appendix D.3] Theorem 2.12 depends on Lemmata 4 and 5 from Han et al. [25] for recovery of the optimal federated weights and for negligibility of the weight-estimation uncertainty, but these lemmas are not stated in this manuscript. The proof also introduces a 'modified informative source condition' without verifying it beyond observing that eta^s_{t,a} > 0. Since the strict efficiency-gain claim rests on these imported oracle properties, the authors should state the lemmas' conditions explicitly, adapt them to the present survival setting, or prove the required oracle properties directly. As written, the theorem's conclusions are conditional on unstated results.
minor comments (4)
  1. [Figure 3] The RRMSE panel is difficult to read: the vertical axis is labeled 10-30 while the relevant values printed in the panel are below 1. Please reconcile the axis scale and the printed values, and make the legend legible.
  2. [Theorem 2.8] In the statement of Theorem 2.8, 'convergence in distribution' should be 'converges in distribution'.
  3. [Algorithm 1] The algorithm output uses bV^dagger_{t,a} while Theorem 2.12 and the text use bV_{t,a}; please standardize the notation.
  4. [Section 2.3.2] In Equation (2) and Algorithm 1, the objective uses centered influence functions bphi*, while the discrepancy bchi^{k,0}_{n,t,a} is computed from uncentered influence functions. The relation is correct, but a brief sentence explaining why this is the appropriate centering convention would improve clarity.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the EIF derivations and weighted estimator are self-contained; minor self-citation burden remains, with a correctness gap in the oracle-set argument that is not circular.

full rationale

No circular step of the kind defined in the review schema is present. The target survival function theta_0(t,a) enters the derivations as the estimand, not as an input, so the CCOD EIF (Theorem 2.7) and the site-specific EIF (Theorem 2.10) are derived from first-order pathwise-differentiation arguments rather than assumed. The federated estimator is a weighted average of these estimators, and the variance comparison with target-only estimation is a direct consequence of the optimization over weights; the claim that the target-only estimator is a feasible weight vector is mathematically tautological but not circular. The main self-citation burden is real but modest: Appendix D.3 invokes Lemmata 4 and 5 and Proposition 1 from Han et al. [25], a prior paper by co-author Han, to justify recovery of the oracle weights and strict variance reduction. That citation is load-bearing, but it is a published, external result with its own simulations, so under the review rules it counts as independent support rather than circularity. Separately, there is a genuine correctness concern that is not circularity: Appendix D.3 defines the oracle set as S*_{t,a} = {k : theta_k(t,a) = theta_0(t,a)} and asserts that for k in S* the local estimator is consistent for theta_0(t,a), whereas Theorem D.7 requires the stronger CCOD assumption (T(a) independent of R given X) for consistency. Marginal survival equality does not imply the working partial CCOD, so the proof of the strict efficiency gain is incomplete in outcome-shift settings. This is a proof gap, not a reduction of the result to its own inputs. Simulations and the AMP application are external benchmark evaluations rather than fitted restatements of the theoretical claims.

Assumptions & free parameters 2 free parameters · 8 assumptions · 0 invented entities

No new physical or causal entities are introduced. The method's extra load-bearing items beyond standard causal assumptions are the rate conditions for nuisance estimation, the working partial CCOD used to derive local EIFs, and the assumption that coarse summary statistics suffice to estimate density ratios.

free parameters (2)
  • lambda (tuning parameter for L1 federated weights) = Chosen by cross-validation, value not reported
    Controls the sparsity of federated weights in objective (2), and results may be sensitive to its choice.
  • time grid spacing epsilon in Algorithm 1 = Not specified in the paper
    Federated weights and survival estimates are computed on a discrete grid, and the paper does not state the grid resolution used in simulations or the AMP analysis.
assumptions (8)
  • domain assumption SUTVA and consistency: observed outcomes equal potential outcomes under assigned treatment, as stated in Section 2.1.
    Standard causal inference foundation for linking observed data to potential outcomes.
  • domain assumption Unconfoundedness and treatment-specific non-informative censoring (Assumptions 2.1 and 2.2).
    Needed for identifying treatment-specific survival curves from observed data.
  • domain assumption Positivity of treatment, censoring, and site probabilities, bounded away from zero (Assumptions 2.3 and 2.6).
    Required for inverse-weighting and for density ratio terms.
  • domain assumption CCOD for the pooled estimator: T(a) is independent of R given X (Assumption 2.5).
    Defines the setting in which pooled data are exchangeable for outcome distribution.
  • domain assumption Working partial CCOD for federated local estimators: Sk = S0 is used to derive site-specific EIFs (Theorem 2.10, Section 2.3.1).
    Used only to motivate the EIF form; aggregation weights are designed to tolerate violations.
  • standard math Nuisance rate conditions: product-error terms are o_p(n^{-1/2}) uniformly (Conditions D.1-D.3 and D.6-D.8).
    Standard double/debiased machine learning conditions for asymptotic linearity of cross-fitted estimators.
  • ad hoc to paper Oracle weight-selection and informative-source conditions in Appendix D.3, imported from Han et al. [25].
    Load-bearing for Theorem 2.12's claim that the estimated weights recover the variance-minimizing oracle and that a strict efficiency gain occurs.
  • ad hoc to paper Density ratio omega_{k,0}(X) is consistently estimable from coarse summary statistics shared under privacy constraints (Section 2.3.1).
    Not demonstrated in this paper; delegated to Han et al. [25].

how reviews work

0 comments
Cite this review

Pith. "Pith review of Targeted Data Fusion for Region-Specific Survival Effects in the AMP HIV Prevention Trials." pith.science (2026). https://pith.science/paper/S65266KG

@misc{pith2026250118798,
  author       = {Pith},
  title        = {Pith review of: Targeted Data Fusion for Region-Specific Survival Effects in the AMP HIV Prevention Trials},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/S65266KG}},
  note         = {Machine review of arXiv:2501.18798}
}
read the original abstract

The Antibody Mediated Prevention (AMP) trials opened a new scientific frontier by showing that passively administered monoclonal broadly neutralizing antibodies (bnAbs) could prevent HIV-1 acquisition. Conducted across multiple geographic regions, including the United States, Brazil, Peru, Switzerland, and sub-Saharan Africa, the AMP trials revealed substantial regional heterogeneity in treatment efficacy. These differences, together with privacy and regulatory limits on central data pooling, call for methods that borrow strength across regions without sharing individual-level data. To estimate region- and treatment-specific survival curves under distributional heterogeneity, we develop a federated learning approach that combines site-specific estimators via an L1-regularized criterion that downweights data sources not aligned with the target. We further extend the framework to a general class of causal contrasts, including the risk difference (RD), survival ratio (SR), and restricted mean survival time (RMST) difference. Through extensive simulations and an analysis of the AMP trials under different target populations, we show that the proposed approach provides privacy-preserving, region-adaptive inference with improved precision.

Figures

Figures reproduced from arXiv: 2501.18798 by the authors.

Figure 1
Figure 1. DAG for data structures under different CCOD assumptions. When CCOD holds (Panel [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the Federated Algorithm. Each site has its underlying survival functions, [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Simulation results evaluated at day 90, under [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Data analysis results (South Africa as the target site). [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: True treatment-specific survival curves across different sites. Each curve is derived from [PITH_FULL_IMAGE:figures/full_fig_p022_5.png]
Figure 6
Figure 6. Figure 6: Upper panel: Average federated weights of each site at different time point by site heterogeneity cases. Lower panel: catter plots of site-specific federated weights vs. discrepancy measure (χb k n,t,a) 2 values, under 5 scenarios of site heterogeneity and 3 selected t…
Figure 7
Figure 7. Figure 7: Estimation bias (boxplots), relative root mean square error (RRMSE) compared to TGT, [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]
Figure 8
Figure 8. Figure 8: Estimation bias (boxplots), relative root mean square error (RRMSE) compared to TGT, [PITH_FULL_IMAGE:figures/full_fig_p025_8.png]
Figure 9
Figure 9. Figure 9: Estimation bias (boxplots), relative root mean square error (RRMSE) compared to TGT, [PITH_FULL_IMAGE:figures/full_fig_p026_9.png]
Figure 10
Figure 10. Figure 10: Estimated region-specific survival curves of the HVTN 704/HPTN 085 and HVTN [PITH_FULL_IMAGE:figures/full_fig_p028_10.png]
Figure 11
Figure 11. Figure 11: Additional data analysis results when treating the other three regions (OA, BP and US) [PITH_FULL_IMAGE:figures/full_fig_p029_11.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Efficient Inference for Time-to-Event Outcomes by Integrating Right-Censored and Current Status Data

    stat.ME 2025-08 conditional novelty 7.0 of 10

    A new semiparametric data fusion framework derives efficient and doubly robust estimators of survival probabilities by combining right-censored and current status data.

Reference graph

Works this paper leans on

59 extracted references · 52 canonical work pages · cited by 1 Pith paper

  1. [25]

    Federated adaptive causal estimation (face) of target treatment effects.Journal of the American Statistical Association, (just-accepted):1–25, 2025

    Larry Han, Jue Hou, Kelly Cho, Rui Duan, and Tianxi Cai. Federated adaptive causal estimation (face) of target treatment effects.Journal of the American Statistical Association, (just-accepted):1–25, 2025

    work page 2025

  2. [1]

    Time-dependent mediators in survival analysis: modeling direct and indirect effects with the additive hazards model.Biometrical Journal, 62(3):532–549, 2020

    Odd O Aalen, Mats J Stensrud, Vanessa Didelez, Rhian Daniel, Kjetil Røysland, and Susanne Strohmaier. Time-dependent mediators in survival analysis: modeling direct and indirect effects with the additive hazards model.Biometrical Journal, 62(3):532–549, 2020

    work page 2020

  3. [2]

    Generating survival times to simulate cox proportional hazards models with time- varying covariates.Statistics in medicine, 31(29):3946–3958, 2012

    Peter C Austin. Generating survival times to simulate cox proportional hazards models with time- varying covariates.Statistics in medicine, 31(29):3946–3958, 2012

    work page 2012

  4. [3]

    Doubly-robust estimators of treatment-specific survival distributions in observational studies with stratified sampling.Biometrics, 69(4):830–839, 2013

    Xiaofei Bai, Anastasios A Tsiatis, and Sean M O’Brien. Doubly-robust estimators of treatment-specific survival distributions in observational studies with stratified sampling.Biometrics, 69(4):830–839, 2013

    work page 2013

  5. [4]

    Springer, 1993

    Peter J Bickel, Chris AJ Klaassen, Peter J Bickel, Ya’acov Ritov, J Klaassen, Jon A Wellner, and YA’Acov Ritov.Efficient and adaptive estimation for semiparametric models, volume 4. Springer, 1993

    work page 1993

  6. [5]

    Methods for integrating trials and non-experimental data to examine treatment effect heterogeneity.Statistical Science, 38(4):640–654, 2023

    Carly Lupton Brantner, Ting-Hsuan Chang, Trang Quynh Nguyen, Hwanhee Hong, Leon Di Stefano, and Elizabeth A Stuart. Methods for integrating trials and non-experimental data to examine treatment effect heterogeneity.Statistical Science, 38(4):640–654, 2023

    work page 2023

  7. [6]

    Tutorial in biostatistics survival analysis in observational studies

    Kate Bull and David J Spiegelhalter. Tutorial in biostatistics survival analysis in observational studies. Statistics in medicine, 16(9):1041–1074, 1997. 16

    work page 1997

  8. [7]

    Transporting randomized trial results to estimate counter- factual survival functions in target populations.Pharmaceutical Statistics, 2024

    Zhiqiang Cao, Youngjoo Cho, and Fan Li. Transporting randomized trial results to estimate counter- factual survival functions in target populations.Pharmaceutical Statistics, 2024

    work page 2024

Show all 59 references
  1. [8]

    Double/debiased machine learning for treatment and structural parameters

    Victor Chernozhukov, Denis Chetverikov, Mert Demirer, Esther Duflo, Christian Hansen, Whitney Newey, and James Robins. Double/debiased machine learning for treatment and structural parameters. The Econometrics Journal, 2018

    2018

  2. [9]

    Targeted optimal treatment regime learning using summary statistics

    Jianing Chu, Wenbin Lu, and Shu Yang. Targeted optimal treatment regime learning using summary statistics. Biometrika, 110(4):913–931, 2023

    2023

  3. [10]

    Locallyweighted regression: anapproach toregressionanalysis by local fitting.Journal of the American statistical association, 83(403):596–610, 1988

    WilliamS Clevelandand Susan J Devlin. Locallyweighted regression: anapproach toregressionanalysis by local fitting.Journal of the American statistical association, 83(403):596–610, 1988

    1988

  4. [11]

    Causal inference methods for combining randomized trials and obser- vational studies: a review.Statistical science, 39(1):165–191, 2024

    Bénédicte Colnet, Imke Mayer, Guanhua Chen, Awa Dieng, Ruohong Li, Gaël Varoquaux, Jean-Philippe Vert, Julie Josse, and Shu Yang. Causal inference methods for combining randomized trials and obser- vational studies: a review.Statistical science, 39(1):165–191, 2024

    2024

  5. [12]

    Two randomized trials of neutralizing antibodies to prevent hiv-1 acquisition.New England Journal of Medicine, 384(11):1003– 1014, 2021

    Lawrence Corey, Peter B Gilbert, Michal Juraska, David C Montefiori, Lynn Morris, Shelly T Karuna, Srilatha Edupuganti, Nyaradzo M Mgodi, Allan C Decamp, Erika Rudnicki, et al. Two randomized trials of neutralizing antibodies to prevent hiv-1 acquisition.New England Journal of...

    2021

  6. [13]

    Competing risks and time-dependent covariates.Biometrical journal, 52(1):138–158, 2010

    Giuliana Cortese and Per K Andersen. Competing risks and time-dependent covariates.Biometrical journal, 52(1):138–158, 2010

    2010

  7. [14]

    Regression models and life-tables

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

    1972

  8. [15]

    Yifan Cui, Michael R Kosorok, Erik Sverdrup, Stefan Wager, and Ruoqing Zhu. Estimating hetero- geneous treatment effects with right-censored data via causal survival forests.Journal of the Royal Statistical Society Series B: Statistical Methodology, 85(2):179–211, 2023

    2023

  9. [16]

    A review of generalizability and transportability

    Irina Degtiar and Sherri Rose. A review of generalizability and transportability. Annual Review of Statistics and Its Application, 10:501–524, 2023

    2023

  10. [17]

    R package version 1.0.0

    Iván Díaz.survtmlerct: Efficiency guarantees for covariate adjustment in RCTs with survival outcomes. R package version 1.0.0

  11. [18]

    A fast trans-lasso algorithm with penalized weighted score function.Computational Statistics & Data Analysis, 192:107899, 2024

    Xianqiu Fan, Jun Cheng, Hailing Wang, Bin Zhang, and Zhenzhen Chen. A fast trans-lasso algorithm with penalized weighted score function.Computational Statistics & Data Analysis, 192:107899, 2024

    2024

  12. [19]

    Time-dependent covariates in the cox proportional-hazards regression model

    Lloyd D Fisher and Danyu Y Lin. Time-dependent covariates in the cox proportional-hazards regression model. Annual review of public health, 20(1):145–157, 1999. 17

    1999

  13. [20]

    On the role of surrogates in conformal inference of individual causal effects.arXiv preprint arXiv:2412.12365, 2024

    Chenyin Gao, Peter B Gilbert, and Larry Han. On the role of surrogates in conformal inference of individual causal effects.arXiv preprint arXiv:2412.12365, 2024

    2024 arXiv

  14. [21]

    Improving randomized controlled trial analysis via data-adaptive borrowing.Biometrika, page asae069, 2024

    Chenyin Gao, Shu Yang, Mingyang Shan, Wenyu YE, Ilya Lipkovich, and Douglas Faries. Improving randomized controlled trial analysis via data-adaptive borrowing.Biometrika, page asae069, 2024

    2024

  15. [22]

    A survey of product-integration with a view toward application in survival analysis.The annals of statistics, 18(4):1501–1555, 1990

    Richard D Gill and Soren Johansen. A survey of product-integration with a view toward application in survival analysis.The annals of statistics, 18(4):1501–1555, 1990

    1990

  16. [23]

    Breaking free from the hazard ratio: Embracing the restricted mean survival time in clinical trials, 2023

    Larry Han. Breaking free from the hazard ratio: Embracing the restricted mean survival time in clinical trials, 2023

    2023

  17. [24]

    Truncated, not forgotten—handling left truncation in time-to-event studies, 2024

    Larry Han. Truncated, not forgotten—handling left truncation in time-to-event studies, 2024

    2024

  18. [26]

    Privacy-preserving, communication-efficient, and target-flexible hospital quality measurement.The Annals of Applied Statistics, 18(2):1337–1359, 2024

    LarryHan, YigeLi, BijanNiknam, andJoséRZubizarreta. Privacy-preserving, communication-efficient, and target-flexible hospital quality measurement.The Annals of Applied Statistics, 18(2):1337–1359, 2024

    2024

  19. [27]

    Multiply robust federated estimation of targeted average treatment effects

    Larry Han, Zhu Shen, and Jose Zubizarreta. Multiply robust federated estimation of targeted average treatment effects. Advances in Neural Information Processing Systems, 36:70453–70482, 2023

    2023

  20. [28]

    Identifying surrogate markers in real-world comparative effectiveness research.Statistics in Medicine, 41(26):5290–5304, 2022

    Larry Han, Xuan Wang, and Tianxi Cai. Identifying surrogate markers in real-world comparative effectiveness research.Statistics in Medicine, 41(26):5290–5304, 2022

    2022

  21. [29]

    The hazards of hazard ratios.Epidemiology, 21(1):13–15, 2010

    Miguel A Hernán. The hazards of hazard ratios.Epidemiology, 21(1):13–15, 2010

    2010

  22. [30]

    Inference of treatment effect and its regional modifiers using restricted mean survival time in multi-regional clinical trials.arXiv preprint arXiv:2404.08128, 2024

    Kaiyuan Hua, Hwanhee Hong, and Xiaofei Wang. Inference of treatment effect and its regional modifiers using restricted mean survival time in multi-regional clinical trials.arXiv preprint arXiv:2404.08128, 2024

    2024 arXiv

  23. [31]

    Nonparametric estimation from incomplete observations.Journal of the American statistical association, 53(282):457–481, 1958

    Edward L Kaplan and Paul Meier. Nonparametric estimation from incomplete observations.Journal of the American statistical association, 53(282):457–481, 1958

    1958

  24. [32]

    Dasom Lee, Shu Yang, and Xiaofei Wang. Doubly robust estimators for generalizing treatment effects on survival outcomes from randomized controlled trials to a target population.Journal of Causal Inference, 10(1):415–440, 2022

    2022

  25. [33]

    Targeting underrepresented populations in precision medicine: A federated transfer learning approach.The Annals of Applied Statistics, 17(4):2970–2992, 2023

    Sai Li, Tianxi Cai, and Rui Duan. Targeting underrepresented populations in precision medicine: A federated transfer learning approach.The Annals of Applied Statistics, 17(4):2970–2992, 2023

    2023

  26. [34]

    Efficient estimation under data fusion.Biometrika, 110(4):1041–1054, 2023

    Sijia Li and Alex Luedtke. Efficient estimation under data fusion.Biometrika, 110(4):1041–1054, 2023. 18

    2023

  27. [35]

    Multi-source conformal inference under distribution shift

    Yi Liu, Alexander Levis, Sharon-Lise Normand, and Larry Han. Multi-source conformal inference under distribution shift. InForty-first International Conference on Machine Learning

  28. [36]

    Estimation of the cumulative incidence function under multiple dependent and independent censoring mechanisms

    Judith J Lok, Shu Yang, Brian Sharkey, and Michael D Hughes. Estimation of the cumulative incidence function under multiple dependent and independent censoring mechanisms. Lifetime data analysis, 24:201–223, 2018

    2018

  29. [37]

    Debiased machine learning for counterfactual survival functionals based on left-truncated right-censored data.arXiv preprint arXiv:2411.09017, 2024

    Eric R Morenz, Charles J Wolock, and Marco Carone. Debiased machine learning for counterfactual survival functionals based on left-truncated right-censored data.arXiv preprint arXiv:2411.09017, 2024

    2024 arXiv

  30. [38]

    On the application of probability theory to agricultural experiments: Essay on principles

    J Neyman. On the application of probability theory to agricultural experiments: Essay on principles. Statistical Science, 5:465–480, 1923

    1923

  31. [39]

    A semiparametric cox–aalen transformation model with censored data.Biometrics, 79(4):3111–3125, 2023

    Xi Ning, Yinghao Pan, Yanqing Sun, and Peter B Gilbert. A semiparametric cox–aalen transformation model with censored data.Biometrics, 79(4):3111–3125, 2023

    2023

  32. [40]

    Analysis of semiparametric regression models for repeated outcomes in the presence of missing data.Journal of the american statistical association, 90(429):106–121, 1995

    James M Robins, Andrea Rotnitzky, and Lue Ping Zhao. Analysis of semiparametric regression models for repeated outcomes in the presence of missing data.Journal of the american statistical association, 90(429):106–121, 1995

    1995

  33. [41]

    The central role of the propensity score in observational studies for causal effects.Biometrika, 70(1):41–55, 1983

    Paul R Rosenbaum and Donald B Rubin. The central role of the propensity score in observational studies for causal effects.Biometrika, 70(1):41–55, 1983

    1983

  34. [42]

    Estimating causal effects of treatments in randomized and nonrandomized studies.Journal of Educational Psychology, 66:688–701, 1974

    DB Rubin. Estimating causal effects of treatments in randomized and nonrandomized studies.Journal of Educational Psychology, 66:688–701, 1974

    1974

  35. [43]

    Empirical processes

    AW van der Vaart and Jon A Wellner. Empirical processes. In Weak Convergence and Empirical Processes: With Applications to Statistics, pages 127–384. Springer, 2023

    2023

  36. [44]

    Statistical inference for variable importance

    Mark J Van der Laan. Statistical inference for variable importance. The international journal of biostatistics, 2(1), 2006

    2006

  37. [45]

    Super learner.Statistical applications in genetics and molecular biology, 6(1):1–21, 2007

    Mark J van der Laan, Eric C Polley, and Alan E Hubbard. Super learner.Statistical applications in genetics and molecular biology, 6(1):1–21, 2007

    2007

  38. [46]

    AW Van der Vaart and JA Wellner.Weak Convergence and Empirical Processes.Springer & Verlag New York, 1996

    1996

  39. [47]

    Doubly robust estimation under covariate-induced de- pendent left truncation.Biometrika, page asae005, 2024

    Yuyao Wang, Andrew Ying, and Ronghui Xu. Doubly robust estimation under covariate-induced de- pendent left truncation.Biometrika, page asae005, 2024

    2024

  40. [48]

    Inference for treatment-specific survival curves using machine learning.Journal of the American Statistical Association, pages 1–13, 2023

    Ted Westling, Alex Luedtke, Peter B Gilbert, and Marco Carone. Inference for treatment-specific survival curves using machine learning.Journal of the American Statistical Association, pages 1–13, 2023. 19

    2023

  41. [49]

    Correcting an estimator of a multivariate monotone function with isotonic regression.Electronic journal of statistics, 14(2):3032, 2020

    Ted Westling, Mark J van der Laan, and Marco Carone. Correcting an estimator of a multivariate monotone function with isotonic regression.Electronic journal of statistics, 14(2):3032, 2020

    2020

  42. [50]

    A framework for leveraging machine learning tools to estimate personalized survival curves.Journal of Computational and Graphical Statistics, pages 1–11, 2024

    Charles J Wolock, Peter B Gilbert, Noah Simon, and Marco Carone. A framework for leveraging machine learning tools to estimate personalized survival curves.Journal of Computational and Graphical Statistics, pages 1–11, 2024

    2024

  43. [51]

    Adjusted kaplan–meier estimator and log-rank test with inverse probability of treatment weighting for survival data.Statistics in medicine, 24(20):3089–3110, 2005

    Jun Xie and Chaofeng Liu. Adjusted kaplan–meier estimator and log-rank test with inverse probability of treatment weighting for survival data.Statistics in medicine, 24(20):3089–3110, 2005

    2005

  44. [52]

    Combining multiple observational data sources to estimate causal effects

    Shu Yang and Peng Ding. Combining multiple observational data sources to estimate causal effects. Journal of the American Statistical Association, 2019

    2019

  45. [53]

    Semiparametric estimation of structural failure time models in continuous-time processes.Biometrika, 107(1):123–136, 2020

    Shu Yang, Karen Pieper, and Frank Cools. Semiparametric estimation of structural failure time models in continuous-time processes.Biometrika, 107(1):123–136, 2020

    2020

  46. [54]

    Modeling survival distribution as a function of time to treatment discontinuation: A dynamic treatment regime approach.Biometrics, 74(3):900–909, 2018

    Shu Yang, Anastasios A Tsiatis, and Michael Blazing. Modeling survival distribution as a function of time to treatment discontinuation: A dynamic treatment regime approach.Biometrics, 74(3):900–909, 2018

    2018

  47. [55]

    Robust estimation of optimal dynamic treatment regimes for sequential treatment decisions.Biometrika, 100(3):681–694, 2013

    Baqun Zhang, Anastasios A Tsiatis, Eric B Laber, and Marie Davidian. Robust estimation of optimal dynamic treatment regimes for sequential treatment decisions.Biometrika, 100(3):681–694, 2013

    2013

  48. [56]

    Enhancing statistical validity and power in hybrid con- trolled trials: A randomization inference approach with conformal selective borrowing.arXiv preprint arXiv:2410.11713, 2024

    Ke Zhu, Shu Yang, and Xiaofei Wang. Enhancing statistical validity and power in hybrid con- trolled trials: A randomization inference approach with conformal selective borrowing.arXiv preprint arXiv:2410.11713, 2024

    2024 arXiv

  49. [57]

    The adaptive lasso and its oracle properties.Journal of the American statistical association, 101(476):1418–1429, 2006

    Hui Zou. The adaptive lasso and its oracle properties.Journal of the American statistical association, 101(476):1418–1429, 2006. 20 A Additional Details and Results for Numerical Experiments A.1 Data generating process Three covariates X1, X2, and X3 are sampled as transformat...

    2006

  50. [58]

    sup u∈[0,t] 1 b¯Gm(u | a, X) − 1 ¯G∞(u | a, X) #2 , ¯A5,n,m,t,a = P

    However, for the BP and US regions, the federated weights are nearly 1 for the target site and 0 for all other sites. This pattern suggests that when targeting the survival curves of BP or US, other sites contribute substantial biases—an observation that corroborates our findi...

  51. [59]

    sup u∈[0,t] Gm n bφCCOD n,m,u,a − φCCOD ∞,u,a Tm # = E

    for more theory on empirical processes and their applications. Lemma D.2 (Lemma 4 in Westling et al. [48]). Let F = {x 7→ ft(x) : t ∈ [0, τ]} be a class of functions on a sample spaceX such thatfs(x) ≤ ft(x) for all0 ≤ s ≤ t ≤ τ and x ∈ X, and such that an envelope F for F sat...

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.