Accelerated Failure Time Modeling in Censored Data Analysis

Accelerated Failure Time (AFT) modelling provides a parametric framework in which the logarithm of survival time is represented as a linear function of covariates. This contrasts with hazard-based methods by offering direct interpretation of covariate effects on the time scale—for instance, quantifying how a treatment multiplies expected survival time. AFT models accommodate various censoring mechanisms (right, left or interval) through techniques such as the Buckley-James estimator or weighted least-squares, which impute or weight censored observations without imposing strict proportional hazards assumptions. Choice of baseline distribution (Weibull, log-normal, generalised gamma) yields flexible shapes for underlying event-time distributions. Recent methodological innovations have extended classical AFT approaches into high-dimensional and network-structured settings via penalised likelihood and empirical-likelihood frameworks, enhancing variable selection and estimation accuracy. Such advancements strengthen the global applicability of AFT models in clinical trials, reliability engineering and omics research by combining interpretability with robust handling of incomplete event-time data.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent advances have tackled high-dimensional covariates, correlated predictors and complex censoring jointly. A seamless-L0 penalisation method for the AFT model enables simultaneous variable selection and estimation in ultra-high-dimensional contexts, achieving oracle properties and asymptotic normality while outperforming traditional penalties in simulation and real omics analyses. A network-constraint Weibull AFT model incorporates correlation patterns among predictors via a double-penalty approach that promotes both sparsity and grouping; an efficient proximal-gradient algorithm ensures scalable computation and theoretical consistency, facilitating discovery of biomarker networks. To address diverse censoring types within a unified semiparametric framework, weighted empirical likelihood has been extended to the accelerated life model, yielding weighted maximum-likelihood estimators and confidence intervals applicable to right, left, interval and partly interval-censored data, with simulations demonstrating robust performance across censoring scenarios.

Accelerated Failure Time Modeling in Censored Data Analysis publication trend

The graph below shows the total number of articles in accelerated failure time modeling in censored data analysis across all publications each year (not limited to Nature Index journals).

Technical terms

Accelerated Failure Time Model: A parametric regression model in which the logarithm of time to event is expressed as a linear combination of covariates, allowing direct interpretation of time-scale effects.

Censoring: A condition in time-to-event data where the event of interest is only partially observed—for example, right-censoring occurs when the event has not happened by study end.

Buckley-James Estimator: A semiparametric method for AFT models that replaces censored survival times with imputed values based on Kaplan–Meier estimates, enabling least-squares fitting.

Penalised Likelihood: A technique that adds a penalty term to the log-likelihood to enforce sparsity or structured selection among covariates, commonly used in high-dimensional inference.

Empirical Likelihood: A nonparametric method of inference that constructs likelihoods directly from data weights, extendable to weighted versions for handling censoring without strict distributional assumptions.

References

1. Accelerated failure-time model with weighted least-squares estimation: application on survival of HIV positives. Archives of Public Health (2021).
2. Variable selection and estimation for accelerated failure time model via seamless-$ L_0 $ penalty. AIMS Mathematics (2023).
3. A Network‐Constrain Weibull AFT Model for Biomarkers Discovery. Biometrical Journal (2024).
4. Weighted Empirical Likelihood for Accelerated Life Model with Various Types of Censored Data. Stats (2024).