Efficient estimation for deep generalized accelerated hazards models with interval-censored data

Area of Science:

• Biostatistics
• Machine Learning
• Survival Analysis

Background:

• Analyzing interval-censored data is crucial in various fields, including medicine and reliability engineering.
• Existing models may struggle to capture complex, nonlinear relationships between risk factors and failure times.

Purpose of the Study:

• To propose a flexible and adaptive deep generalized accelerated hazards model for interval-censored data.
• To investigate the relationship between risk factors and the hazard rate in failure time analysis.

Main Methods:

• Development of a sieve maximum likelihood estimation procedure.
• Integration of deep neural networks for nonparametric effect capture.
• Utilization of monotonic splines for flexible modeling.

Main Results:

• Derivation of a nonasymptotic error bound for the estimator.
• Demonstration of asymptotic normality and semiparametric efficiency of the estimator.
• Validation of the method's finite-sample performance through simulation studies.

Conclusions:

• The proposed deep generalized accelerated hazards model offers a powerful tool for analyzing complex interval-censored data.
• The method demonstrates strong theoretical properties and practical applicability, as shown in the Atherosclerosis Risk in Communities study.