Sequentially Doubly Robust Estimation of Conditional Survival Probability with Time-Varying Covariates
Pith reviewed 2026-05-18 08:28 UTC · model grok-4.3
The pith
A sequentially doubly robust estimator consistently estimates conditional survival probabilities when time-varying covariates explain censoring on a discrete schedule.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The central claim is that the proposed estimator of the conditional survival probability is sequentially doubly robust, meaning it is consistent if within each time window the censoring distribution is consistently estimated or both the time-to-event distribution and the conditional mean probability are consistently estimated. In the special case of marginal survival, it is asymptotically efficient.
What carries the argument
The sequentially doubly robust estimator, which combines inverse probability weighting for censoring with augmentation terms from outcome regression and an additional conditional mean, applied sequentially across discrete time windows.
If this is right
- If the censoring model is correct in every window the estimator converges even when the time-to-event and conditional-mean models are misspecified.
- If both the time-to-event distribution and the conditional mean are correctly estimated in every window the estimator converges even when the censoring model is misspecified.
- When targeting the marginal survival probability the estimator attains the semiparametric efficiency bound.
- The estimator can be applied directly to scheduled-visit data such as post-vaccination immune-response measurements that explain loss to follow-up.
Where Pith is reading between the lines
- The sequential structure may carry over to other longitudinal problems that mix continuous-time outcomes with discrete measurement grids.
- If visit intervals become dense the estimator could serve as an approximation to fully continuous-time covariate models.
- Empirical checks that compare the estimator against a fully parametric benchmark on the same data would quantify the gain from the robustness property.
Load-bearing premise
Censoring within each discrete time window is fully explained by the observed time-varying covariates, and at least one of the nuisance functions is consistently estimated in each window.
What would settle it
A simulation or real dataset where censoring depends on an unobserved factor or where all nuisance estimators are inconsistent in at least one window, producing clear bias in the estimated conditional survival curve.
read the original abstract
It is often of interest to study the association between covariates and the cumulative incidence of a right-censored time-to-event outcome. When time-varying covariates are measured on a fixed discrete time scale, it is desirable to account for these more up-to-date covariates when addressing censoring. For example, in vaccine trials, it is of interest to study the association between immune response levels after administering the vaccine and the cumulative incidence of the endpoint, while accounting for loss to follow-up explained by immune response levels measured at multiple post-vaccination visits. Existing methods rely on stringent parametric assumptions, do not account for informative censoring due to time-varying covariates when time is continuous, only estimate a marginal survival probability, or do not fully use the discrete-time structure of post-treatment covariates. We propose a nonparametric estimator of the continuous-time survival probability conditional on covariates, accounting for censoring due to time-varying covariates measured on a fixed discrete time scale. We show that the estimator is sequentially doubly robust: it is consistent if, within each time window between adjacent visits, the censoring distribution is consistently estimated, or both the time-to-event distribution and a conditional mean probability are consistently estimated. We also show that, in the special case of estimating the marginal survival probability, our estimator is asymptotically efficient. We demonstrate the superior performance of our estimator in a simulation experiment, and apply the method to a COVID-19 vaccine efficacy trial.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a nonparametric estimator for the continuous-time survival probability conditional on covariates, accounting for right-censoring due to time-varying covariates observed at fixed discrete visits. It establishes sequential double robustness: the estimator is consistent if, within each inter-visit time window, either the censoring distribution is consistently estimated or both the time-to-event distribution and a conditional mean probability are consistently estimated. The paper also shows asymptotic efficiency in the special case of marginal survival probability and demonstrates performance via simulations and an application to a COVID-19 vaccine efficacy trial.
Significance. If the sequential double-robustness property holds, the estimator provides a flexible, assumption-light approach to handling informative censoring with discrete-time-updated covariates, which is relevant for clinical-trial settings such as vaccine studies. The efficiency result for the marginal case and the explicit use of the discrete-visit structure are strengths; the method could improve upon existing parametric or marginal-only approaches when at least one of the per-window nuisance conditions is met.
major comments (1)
- [§3] §3 (Theoretical Results), consistency theorem: the sequential double-robustness argument relies on a recursive/product structure across windows. The manuscript must explicitly show that estimation errors from an earlier window (even when corrected by the 'or' condition in that window) do not distort the effective covariate or survival distribution passed to the next window in a way that invalidates the per-window condition later. A lemma or inductive step addressing propagation is needed; without it the overall consistency claim is not yet load-bearing.
minor comments (2)
- [§4] §4 (Simulation): report the specific estimators (e.g., parametric, kernel, or machine-learning) used for each nuisance function and the bandwidth or tuning choices; this would clarify how the reported finite-sample performance relates to the theoretical conditions.
- [§2] Notation: define the time-window indexing (t_k, t_{k+1}) and the associated conditional distributions more explicitly in the main text before the estimator definition to aid readability.
Simulated Author's Rebuttal
We thank the referee for their constructive and detailed review. The comment on the consistency theorem raises a valid point about the need for greater explicitness in the inductive argument. We address it below and will strengthen the manuscript accordingly.
read point-by-point responses
-
Referee: [§3] §3 (Theoretical Results), consistency theorem: the sequential double-robustness argument relies on a recursive/product structure across windows. The manuscript must explicitly show that estimation errors from an earlier window (even when corrected by the 'or' condition in that window) do not distort the effective covariate or survival distribution passed to the next window in a way that invalidates the per-window condition later. A lemma or inductive step addressing propagation is needed; without it the overall consistency claim is not yet load-bearing.
Authors: We agree that an explicit inductive lemma would improve the clarity and rigor of the presentation. The current proof proceeds recursively: within each inter-visit window the estimator is shown to be consistent under the per-window 'or' condition (consistent censoring or consistent time-to-event plus conditional mean), and the resulting conditional survival factor is multiplied into the cumulative product. Because the time-varying covariates are observed only at the discrete visit times, the filtration up to the start of window m+1 is a measurable function of the data up to the end of window m; consistency of the cumulative estimator up to m therefore implies that the effective covariate distribution entering window m+1 satisfies the same regularity conditions used for window m. To make this propagation fully transparent we will add a new supporting lemma (Lemma 3.1) that proceeds by induction on the number of windows and bounds the total variation distance between the estimated and true cumulative survival functions at each step. The lemma will be placed immediately before the main consistency theorem in §3. This addition does not alter any of the stated results but directly addresses the referee's concern that the recursive structure must be shown to be load-bearing. revision: yes
Circularity Check
No circularity: sequential double robustness derived from standard semiparametric arguments
full rationale
The paper establishes sequential double robustness for its nonparametric estimator of conditional continuous-time survival probability by showing consistency holds when, in each discrete inter-visit window, either the censoring distribution is consistent or both the time-to-event distribution and conditional mean are consistent. This follows from recursive application of influence-function or martingale-based arguments across windows, with nuisance consistency treated as an external assumption rather than a fitted quantity renamed as a prediction. No self-definitional equations, load-bearing self-citations, or ansatz smuggling appear in the derivation chain; the result is not equivalent to its inputs by construction and remains falsifiable via external nuisance estimation performance.
Axiom & Free-Parameter Ledger
axioms (1)
- domain assumption Censoring within each interval between discrete measurement times is explained by the observed time-varying covariates.
Lean theorems connected to this paper
-
IndisputableMonolith/Foundation/ArithmeticFromLogic.leanreality_from_one_distinction unclear?
unclearRelation between the paper passage and the cited Recognition theorem.
We show that the estimator is sequentially doubly robust: it is consistent if, within each time window between adjacent visits, the censoring distribution is consistently estimated, or both the time-to-event distribution and a conditional mean probability are consistently estimated.
-
IndisputableMonolith/Cost/FunctionalEquation.leanwashburn_uniqueness_aczel unclear?
unclearRelation between the paper passage and the cited Recognition theorem.
Theorem 2 (Multiply robust formula). Under Conditions 1 and 2, for every k=0,...,K, E*[T_k,τ({U_j,S_j,G_j})|X>t_k,H_k]=Q^*_{k,τ}(H_k) if ... (i) S_j=S^*_j and U_j=U^*_j,τ or (ii) G_j=G^*_j
What do these tags mean?
- matches
- The paper's claim is directly supported by a theorem in the formal canon.
- supports
- The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
- extends
- The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
- uses
- The paper appears to rely on the theorem as machinery.
- contradicts
- The paper's claim conflicts with a theorem or certificate in the canon.
- unclear
- Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.