Bayesian information criterion approximations for model selection in multivariate logistic regression with application to electronic medical records

Selig, Katharina

Katharina Selig

If you experience problems opening the document, please try this link.

Original title:: Bayesian information criterion approximations for model selection in multivariate logistic regression with application to electronic medical records
Translated title:: Schwarz-Bayes-Informationskriterium Approximationen für die Modellselektion der multivariate logistische Regression mit Anwendung auf elektronische Patientendaten
Author:: Selig, Katharina
Year:: 2020
Document type:: Dissertation
Faculty/School:: Fakultät für Mathematik
Advisor:: Ankerst, Donna P. (Prof., Ph.D.)
Referee:: Ankerst, Donna P. (Prof., Ph.D.); Shaw, Pamela A. (Prof., Ph.D.)
Language:: en
Subject group:: MAT Mathematik
TUM classification:: BIO 110d; MAT 620d
Abstract:: We prove that the Bayesian information criterion approximates the Bayes factor for the multivariate logistic regression model for arbitrary and the unit information prior distributions and give the order of approximation. Further, we provide a framework in R for the Bayesian multivariate logistic regression. Using univariate and multivariate logistic regression models we develop a risk prediction model for five adverse pathological outcomes after prostatectomy based on electronic medical records. «
We prove that the Bayesian information criterion approximates the Bayes factor for the multivariate logistic regression model for arbitrary and the unit information prior distributions and give the order of approximation. Further, we provide a framework in R for the Bayesian multivariate logistic regression. Using univariate and multivariate logistic regression models we develop a risk prediction model for five adverse pathological outcomes after prostatectomy based on electronic medical records... »
Translated abstract:: Wir beweisen für die multivariate logistische Regression, dass für beliebige und für die unit information A-priori-Verteilungen das Schwarz-Bayes-Informationskriterium den Bayes Faktor approximiert. Eine Implementierung der Bayes'schen multivariaten logistischen Regression wird für R zur Verfügung gestellt. Basierend auf elektronischen Patientendaten, erstellen wir mit univariaten und multivariaten logistischen Regressionmodellen ein Risikoprognosemodell für fünf negative pathologische Befunde.
WWW:: https://mediatum.ub.tum.de/?id=1531001
Date of submission:: 17.12.2019
Oral examination:: 02.04.2020
File size:: 9299618 bytes
Pages:: 181
Urn (citeable URL):: https://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:91-diss-20200402-1531001-1-8
Last change:: 29.05.2020
BibTeX