Model selection curves for survival analysis with accelerated failure time models

J. H. Karami, K. Luo, T. Fung

    Research output: Chapter in Book/Report/Conference proceedingConference proceeding contributionpeer-review

    Abstract

    Many model selection processes involve minimizing a loss function of the data over a set of models. A recent introduced approach is model selection curves, in which the selection criterion is expressed as a function of penalty multiplier in a linear (mixed) model or generalized linear model. In this article, we have adopted the model selection curves for accelerated failure time (AFT) models of survival data. In our simulation study, it was found that for data with small sample size and high proportion of censoring, the model selection curves approach outperformed the traditional model selection criteria, Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). In other situations with respect to sample size and proportion of censoring, the model selection curves correctly identify the true model whether it is a full or reduced model. Moreover, through bootstrapping, it was shown that the model selection curves can be used to enhance the model selection process in AFT models.
    Original languageEnglish
    Title of host publication60th ISI World Statistics Congress
    Subtitle of host publicationproceedings
    Place of PublicationThe Hague, The Netherlands
    PublisherInternational Statistical Institute
    Pages2642-2647
    Number of pages6
    ISBN (Print)9789073592353
    Publication statusPublished - 2015
    EventWorld Statistics Congress of the International Statistical Institute (60th : 2015) - Rio de Janeiro
    Duration: 26 Jul 201531 Jul 2015

    Conference

    ConferenceWorld Statistics Congress of the International Statistical Institute (60th : 2015)
    CityRio de Janeiro
    Period26/07/1531/07/15

    Keywords

    • log-likelihood function
    • penalty function
    • penalty multiplier
    • survival data

    Fingerprint Dive into the research topics of 'Model selection curves for survival analysis with accelerated failure time models'. Together they form a unique fingerprint.

    Cite this