Doctoral Dissertations

Keywords and Phrases

Accelerated Failure Time Models; Accelerated Gap Time Models; Deep Learning; Effective Age Process; Gated Recurrent Units (GRU); Recurrent Event Data

Abstract

Recurrent event data arise in many fields such as medicine, reliability, insurance, and economics, where the same event may occur repeatedly for a subject. Accelerated Failure Time (AFT) models provide an intuitive framework for relating covariates to event times and offer a useful alternative to proportional hazards models, allowing direct prediction of event timing under right censoring. However, existing AFT extensions for recurrent events, such as accelerated gap time (AGT) models, often fail to account for interventions between events and may not capture complex temporal patterns.

In this work, we first propose a class of semiparametric AGT models incorporating an effective age process to account for interventions following each event. To address estimation challenges arising from the infinite-dimensional baseline hazard and non-monotone score functions, we develop a weighted efficient score function using parametric submodels. Simulation studies demonstrate that the proposed estimators are consistent and asymptotically normal. An application to a biomedical recurrent event dataset illustrates the practical utility of the method.

Building on this framework, we introduce RNN-AGT, a deep learning extension of AGT models that captures nonlinear and history-dependent effects. The model employs Gated Recurrent Units (GRUs) to learn sequential dependence in gap times and a rank-based loss function to accommodate incomplete gap times, combined with a subsampling strategy to reduce computational cost. Simulation studies across various data-generating settings show stable convergence and strong predictive performance based on adjusted mean squared error (AMSE) and inverse probability of censoring weighted concordance index (IPCW C-index). Applications to biomedical datasets demonstrate that RNN-AGT provides strong discrimination while capturing complex nonlinear and temporal patterns.

Advisor(s)

Adekpedjou, Akim

Committee Member(s)

Olbricht, Gayla R.
Dauxois, Jean Yves
Bohner, Martin, 1966-
Wen, Xuerong Meggie

Department(s)

Mathematics and Statistics

Degree Name

Ph. D. in Mathematics

Publisher

Missouri University of Science and Technology

Publication Date

Spring 2026

Journal article titles appearing in thesis/dissertation

Paper I: Pages 87-172 are intended for submission to AIMS Mathematics.

Paper II: Pages 173-215 are intended for submission to the Journal of Statistics in Medicine.

Pagination

xv, 228 pages

Note about bibliography

Includes_bibliographical_references_(pages 220-227)

Rights

© 2026 Emmanuel Masavo Djegou , All Rights Reserved

Document Type

Dissertation - Open Access

File Type

text

Language

English

Thesis Number

T 12590

Download

Share

 
COinS