Behning, Charlotte: Modeling longitudinal epidemiological data using novel methods for statistical learning and regression. - Bonn, 2024. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5-80166
@phdthesis{handle:20.500.11811/12634,
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5-80166,
author = {{Charlotte Behning}},
title = {Modeling longitudinal epidemiological data using novel methods for statistical learning and regression},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2024,
month = dec,

note = {The analysis of longitudinal data plays an important role in medical research. The data is typically collected during follow-up visits in epidemiological observational studies. These studies often investigate the natural history of (slowly) progressing diseases, with endpoints focusing either on changes in outcome variables over time (longitudinal change endpoints) or the time taken to reach a more severe disease stage (time-to-event endpoints).
This dissertation focuses mainly on the application of these methods in ophthalmology based on the experience gained evaluating the MACUSTAR study. The study aims to develop and validate new candidate endpoints for the early stages of age-related macular degeneration (AMD).
This cumulative dissertation consists of four scientific publications that cover several aspects of modeling longitudinal data using novel statistical learning methods and regression, looking into both longitudinal change and time-to-event endpoints.
The first project investigates the challenge of recruiting participants with low disease burden. To this end, a Poisson mixed-effects regression model was applied to identify factors associated with increased screening rates of participants with asymptomatic early AMD stages in the multi-center MACUSTAR study.
The second work deals with modeling the growth of geographic atrophy (GA) using a novel linear mixed-effects regression framework that directly incorporates the unknown disease age at baseline using random effects. To capture nonlinear GA enlargement, possible transformation parameters were systematically assessed using Box-Cox transformation.
The last two publications present approaches to evaluate time-to-event data in the presence of competing events in statistical learning algorithms. Here, an imputation approach was applied, transforming competing event data such that existing single-event methods could be trained. The methods were evaluated using extensive simulation studies and applied on real-world data sets.
All research articles have been accepted for publication in international peer-reviewed journals (see Publications A-D)},

url = {https://hdl.handle.net/20.500.11811/12634}
}

Die folgenden Nutzungsbestimmungen sind mit dieser Ressource verbunden:

InCopyright