@MASTERSTHESIS\{IMM2014-06779,
    author       = "S. S{\o}nderby",
    title        = "Non-parametric survival analysis in breast cancer using clinical and genomic markers",
    year         = "2014",
    school       = "Technical University of Denmark, Department of Applied Mathematics and Computer Science",
    address      = "Richard Petersens Plads, Building 324, {DK-}2800 Kgs. Lyngby, Denmark, compute@compute.dtu.dk",
    type         = "",
    note         = "{DTU} supervisor: Ole Winther, olwi@dtu.dk, {DTU} Compute",
    url          = "http://www.compute.dtu.dk/English.aspx",
    abstract     = "Background: New survival models based on Gaussian Processes (GP) and Random Forests (RF) have been developed, and have shown good performance in large cancer cohorts.
Purpose: To investigate if these new survival models can improve prediction of 10 year recurrence in a pooled dataset of breast cancer patients. 
Data Sources: Breast cancer patients collected by (Haibe-Kains et al. 2012)
Data Extraction: Patient clinical data and gene expression data from several platforms were extracted. Clinical data, including receptor status, was incomplete. Methods for inference of {ER,} HER2 and PgR receptor status from gene expression data was developed. These methods work independenty of the gene expression platform. Recurrence predictors where extracted from expression data.
Results: A pilot study showed that {RF} survival had worse performance than {GP} based models. {RF} survival was not investigated further. Area under curve (AUC) scores for recurrence prediction in breast cancer patients was calculated for the models Cox {GP} model (CoxGP) and Cox proportional hazard (CoxPH). When appropriate, models were evaluated on dataset with dierent number of covariates.
Limitations: The included data is a pooled dataset and may be skewed.
Conclusion: CoxGP models show better performance than CoxPH. It is shown that addition of features extracted from gene expression data improve prediction of 10 year recurrence in both CoxGP and CoxPH models.

Published code availabe:

https://bitbucket.org/skaae/thesis\_sweave
https://bitbucket.org/skaae/datathesis
https://bitbucket.org/skaae/ktsp"
}