Rank estimation for the accelerated failure time model with partially interval-censored data
Pith reviewed 2026-05-23 00:32 UTC · model grok-4.3
The pith
A Gehan-type rank estimating function directly yields regression coefficients for the accelerated failure time model under partial interval censoring via linear programming.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The paper constructs a Gehan-type monotone estimating function for the accelerated failure time model based on the weighted log-rank test idea, applicable to partially interval-censored data. This function permits solving for the regression coefficients via linear programming, and standard empirical process theory establishes consistency and asymptotic normality. The approach extends to general rank-based functions and to multivariate clustered data, with variance estimation provided.
What carries the argument
The Gehan-type monotone estimating function constructed from weighted log-rank scores, which remains monotone for partially interval-censored observations and is minimized by linear programming.
If this is right
- Regression coefficients are obtained directly without separate nonparametric estimation of the residual distribution function.
- Consistency and asymptotic normality follow from standard empirical process theory once monotonicity is established.
- The same linear-programming procedure extends to general weighted rank estimating functions.
- An efficient variance estimator is available for the resulting coefficient estimates.
- The procedure applies without modification to multivariate clustered partially interval-censored observations.
Where Pith is reading between the lines
- Analyses of progression-free survival or time-to-event data in oncology trials could become routine rather than requiring specialized maximum-likelihood software.
- The linear-programming formulation makes bootstrap or resampling variance estimates computationally inexpensive.
- The monotonicity property may extend to other censoring patterns or to models with time-dependent covariates if analogous score functions can be shown to preserve order.
Load-bearing premise
The constructed Gehan-type estimating function remains monotone when some failure times are known only to lie inside intervals.
What would settle it
A data configuration or simulation in which the estimating function ceases to be monotone, so that linear programming no longer yields a unique or consistent solution for the regression coefficients.
read the original abstract
This paper presents a unified rank-based inferential procedure for fitting the accelerated failure time model to partially interval-censored data. A Gehan-type monotone estimating function is constructed based on the idea of the familiar weighted log-rank test, and an extension to a general class of rank-based estimating functions is suggested. The proposed estimators can be obtained via linear programming and are shown to be consistent and asymptotically normal via standard empirical process theory. Unlike common maximum likelihood-based estimators for partially interval-censored regression models, our approach can directly provide a regression coefficient estimator without involving a complex nonparametric estimation of the underlying residual distribution function. An efficient variance estimation procedure for the regression coefficient estimator is considered. Moreover, we extend the proposed rank-based procedure to the linear regression analysis of multivariate clustered partially interval-censored data. The finite-sample operating characteristics of our approach are examined via simulation studies. Data example from a colorectal cancer study illustrates the practical usefulness of the method.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript develops a rank-based procedure for the accelerated failure time (AFT) model with partially interval-censored data. It constructs a Gehan-type estimating function (extensible to a broader class of weighted log-rank functions) that is asserted to remain monotone, permitting solution via linear programming; consistency and asymptotic normality are claimed via standard empirical process theory. The approach is positioned as avoiding nonparametric estimation of the residual distribution (unlike MLE methods), includes an efficient variance estimator, extends to multivariate clustered data, and is evaluated via simulations and a colorectal cancer data example.
Significance. If the monotonicity property is rigorously established, the method supplies a computationally direct regression-coefficient estimator for AFT models under partial interval censoring, sidestepping the complex nonparametric components typical of likelihood-based alternatives. The linear-programming formulation and empirical-process arguments would constitute a clear practical and theoretical contribution if the key technical step is verified.
major comments (2)
- [estimator construction (§2)] Section on estimator construction (abstract and §2): The central claim that the Gehan-type estimating function remains monotone for partially interval-censored observations (thereby permitting linear programming and direct application of empirical process theory for consistency and asymptotic normality) is load-bearing, yet the manuscript supplies no explicit verification, derivation, or counter-example check showing that the weighted indicators with interval constraints on the residuals preserve monotonicity in β.
- [variance estimation (§3)] Variance estimation procedure (mentioned in abstract and §3): The efficient variance estimator is asserted but the explicit formula or computational steps under the partial interval-censoring structure (e.g., how the influence function or sandwich form accounts for the interval constraints) are not derived or displayed, leaving the asymptotic normality claim without a fully operational variance estimator.
minor comments (2)
- [simulation studies] Simulation section: The finite-sample results are summarized but the precise data-generation mechanism for partial interval censoring (e.g., how the inspection times and censoring intervals are generated) should be stated explicitly to allow reproducibility.
- [general rank-based extension] Notation: The definition of the weighted estimating function for the general rank class should include an explicit display of the weight function and the interval-adjusted comparison rule to clarify the extension beyond the Gehan case.
Simulated Author's Rebuttal
We thank the referee for the constructive and insightful comments on our manuscript. We address each major comment below and will revise the paper to incorporate the requested clarifications and derivations.
read point-by-point responses
-
Referee: Section on estimator construction (abstract and §2): The central claim that the Gehan-type estimating function remains monotone for partially interval-censored observations (thereby permitting linear programming and direct application of empirical process theory for consistency and asymptotic normality) is load-bearing, yet the manuscript supplies no explicit verification, derivation, or counter-example check showing that the weighted indicators with interval constraints on the residuals preserve monotonicity in β.
Authors: We agree that an explicit verification of monotonicity is a key technical detail that should be provided. In the revised manuscript we will add a dedicated derivation in Section 2. The argument proceeds by showing that each summand in the Gehan-type estimating function is non-increasing in β because the indicator comparisons are formed from residuals whose ordering is preserved under the interval constraints; the weights are positive and do not depend on β. This establishes the required monotonicity, justifies the linear-programming formulation, and supports direct application of the empirical-process arguments already outlined in the paper. revision: yes
-
Referee: Variance estimation procedure (mentioned in abstract and §3): The efficient variance estimator is asserted but the explicit formula or computational steps under the partial interval-censoring structure (e.g., how the influence function or sandwich form accounts for the interval constraints) are not derived or displayed, leaving the asymptotic normality claim without a fully operational variance estimator.
Authors: We acknowledge that the explicit expression for the variance estimator under partial interval censoring is not displayed. In the revision we will derive and present the influence-function representation in Section 3, showing how the interval constraints enter the estimating equation and are propagated through the sandwich formula. The resulting estimator will be stated in a form that can be computed directly from the observed data and the fitted residuals, thereby rendering the asymptotic normality result fully operational. revision: yes
Circularity Check
No circularity detected; derivation is self-contained
full rationale
The paper constructs a new Gehan-type estimating function for the AFT model under partial interval censoring by extending the weighted log-rank idea, then invokes standard empirical process theory for consistency and asymptotic normality after establishing monotonicity for LP solvability. No equations or text reduce the central estimator or its properties to a fitted input, self-citation chain, or definitional equivalence. The construction is presented as independent, with no load-bearing self-referential steps or renaming of known results as new derivations.
Axiom & Free-Parameter Ledger
axioms (1)
- domain assumption The data follow the accelerated failure time model and the estimating function is monotone.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.