Causal inference for censored data with continuous marks
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-05-16 18:27 UTCgrok-4.3open to challenge →
The pith
A mark-specific causal effect on failure times can be identified and estimated from censored data with continuous marks.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The paper defines a mark-specific treatment effect within the potential outcomes framework for censored data with continuous marks, characterizes its identifying conditions, proposes a local smoothing estimator for the causal effects, establishes its asymptotic properties, and develops testing methods for effects at specific mark points or intervals using Gaussian approximation.
What carries the argument
The mark-specific treatment effect, defined as the causal contrast in failure times conditional on a fixed value of the continuous mark, identified under independent censoring given treatment and covariates and estimated by local smoothing to address sparsity.
If this is right
- Treatment effects can be estimated and tested at chosen mark values or over intervals while respecting censoring.
- Asymptotic normality of the estimator supplies pointwise confidence intervals and valid p-values for the tests.
- The framework applies directly to marked survival data such as biomarker-tagged failure times in clinical trials.
Where Pith is reading between the lines
- The same local-smoothing idea might extend to settings with multiple continuous marks if the smoothness and independence conditions can be maintained.
- If the method proves stable, it could support mark-stratified treatment rules in precision medicine by highlighting mark values where the causal benefit is largest.
Load-bearing premise
Censoring is independent of the mark given treatment and covariates, and the underlying regression functions are smooth enough for local estimation to work.
What would settle it
Finding that the estimator's bias or coverage error increases sharply in data where censoring depends on the mark after conditioning on treatment and covariates would show the identification conditions fail.
read the original abstract
This paper presents a framework for causal inference in the presence of censored data,where the failure time is marked by a continuous variable referred to as a mark.The mark is observed after treatment and is not meaningful when the failure time is censored. In addition, due to the continuous nature of the marks, observations at each given mark are sparse. These facts make the identification and estimation of causality a challenging task. To address these issues, we define a new mark-specific treatment effect within the potential outcomes framework and characterize its identifying conditions. We then propose a local smoothing estimator for the causal effects and establish its asymptotic properties. We further develop testing methods to evaluate whether the treatment has an effect on the failure time when controlling the values of the mark at certain points or within a defined interval, and develop a Gaussian approximation method to obtain the critical values. We evaluate our method using simulation studies as well as a real dataset from the Antibody Mediated Prevention trials.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper develops a framework for causal inference with right-censored failure times that are marked by a continuous variable observed only on uncensored events. It defines a mark-specific treatment effect within the potential outcomes model, states identifying conditions that include conditional independence of censoring and the mark given treatment and covariates, proposes a local smoothing estimator, derives its asymptotic normality and rates, constructs tests for pointwise and interval effects via Gaussian approximation, and evaluates the procedure in simulations and on Antibody Mediated Prevention trial data.
Significance. If the central identifying assumptions hold, the work supplies a new estimand and a practical local-smoothing procedure for marked censored data, together with asymptotic theory and testing tools. These elements address a recurring setting in survival analysis where continuous marks (e.g., biomarker values) are observed only at events and are sparse, and the simulation and real-data illustrations indicate feasible implementation.
major comments (2)
- [§2.3] §2.3 (Identification): The mark-specific treatment effect is identified only under the assumption that censoring is independent of the continuous mark conditional on treatment and covariates (Assumption 3). Because marks are observed solely on uncensored failures, any unmeasured factor jointly affecting censoring time and mark value renders the local smoothing estimator inconsistent; the paper should supply either a sensitivity analysis or explicit diagnostics for this conditional independence.
- [§4] §4 (Asymptotics): Theorem 1 establishes asymptotic normality of the local estimator under bandwidth conditions h_n → 0, nh_n → ∞, and nh_n^5 → 0, yet the simulation section reports coverage probabilities without verifying that the chosen bandwidths satisfy these rates for the sample sizes examined; the reported coverage may therefore be optimistic.
minor comments (2)
- [§2.1] The notation for the mark-specific potential outcomes (e.g., Y(t,m)) is introduced without an explicit contrast to the standard cumulative incidence function; a short clarifying sentence would aid readers.
- [§6] In the real-data application, the range of marks examined and the bandwidth selection procedure are not stated explicitly; adding these details would improve reproducibility.
Simulated Author's Rebuttal
We thank the referee for the careful reading and constructive comments on our manuscript. We address each major comment below and indicate the changes we will make in the revision.
read point-by-point responses
-
Referee: [§2.3] §2.3 (Identification): The mark-specific treatment effect is identified only under the assumption that censoring is independent of the continuous mark conditional on treatment and covariates (Assumption 3). Because marks are observed solely on uncensored failures, any unmeasured factor jointly affecting censoring time and mark value renders the local smoothing estimator inconsistent; the paper should supply either a sensitivity analysis or explicit diagnostics for this conditional independence.
Authors: We agree that Assumption 3 is a strong identifying condition whose violation would render the estimator inconsistent. In the revised manuscript we will expand Section 2.3 with a more detailed discussion of the assumption’s plausibility in the AMP trial setting, where censoring is primarily administrative. We will also add a new sensitivity-analysis subsection that introduces a controlled dependence parameter between the censoring time and the mark and reports how the mark-specific effect estimates change under mild violations. revision: yes
-
Referee: [§4] §4 (Asymptotics): Theorem 1 establishes asymptotic normality of the local estimator under bandwidth conditions h_n → 0, nh_n → ∞, and nh_n^5 → 0, yet the simulation section reports coverage probabilities without verifying that the chosen bandwidths satisfy these rates for the sample sizes examined; the reported coverage may therefore be optimistic.
Authors: We acknowledge that the simulation section does not explicitly verify the bandwidth rates. In the revision we will add a short table (or paragraph) in the simulation study that reports the cross-validated bandwidths for each n and confirms that they satisfy h_n → 0, n h_n → ∞ and n h_n^5 → 0 for the sample sizes considered. This will clarify that the reported coverage probabilities are consistent with the asymptotic regime of Theorem 1. revision: yes
Circularity Check
No significant circularity; derivation is self-contained under explicit assumptions
full rationale
The paper defines a mark-specific treatment effect inside the standard potential outcomes framework, states identifying conditions (including conditional independence of censoring and mark given treatment/covariates), proposes a local smoothing estimator, and derives its asymptotic normality and rates directly from those assumptions plus standard kernel smoothing arguments. No equation reduces the target estimand or estimator to a fitted input by construction, no uniqueness theorem is imported from self-citation, and no ansatz is smuggled via prior work. The central claims remain falsifiable once the stated assumptions are granted; the derivation chain does not collapse into its own inputs.
Axiom & Free-Parameter Ledger
axioms (1)
- domain assumption Identifying conditions for mark-specific treatment effect hold, including censoring independent of mark given treatment and covariates
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.