%0 Journal Article
%T Modified Logistic Regression Models Using Gene Coexpression and Clinical Features to Predict Prostate Cancer Progression
%A Hongya Zhao
%A Christopher J. Logothetis
%A Ivan P. Gorlov
%A Jia Zeng
%A Jianguo Dai
%J Computational and Mathematical Methods in Medicine
%D 2013
%I Hindawi Publishing Corporation
%R 10.1155/2013/917502
%X Predicting disease progression is one of the most challenging problems in prostate cancer research. Adding gene expression data to prediction models that are based on clinical features has been proposed to improve accuracy. In the current study, we applied a logistic regression (LR) model combining clinical features and gene co-expression data to improve the accuracy of the prediction of prostate cancer progression. The top-scoring pair (TSP) method was used to select genes for the model. The proposed models not only preserved the basic properties of the TSP algorithm but also incorporated the clinical features into the prognostic models. Based on the statistical inference with the iterative cross validation, we demonstrated that prediction LR models that included genes selected by the TSP method provided better predictions of prostate cancer progression than those using clinical variables only and/or those that included genes selected by the one-gene-at-a-time approach. Thus, we conclude that TSP selection is a useful tool for feature (and/or gene) selection to use in prognostic models and our model also provides an alternative for predicting prostate cancer progression. 1. Introduction Prostate cancer (PCa) is the second leading cause of cancer-related deaths among men in the USA [1, 2]. Screening using serum prostate-specific antigen (PSA) has improved the early detection of PCa and has resulted in an increase in the proportion of patients with disease that is curable by prostatectomy [3, 4]. However, 20% to 30% of treated patients will develop a local or metastatic recurrence which reflects the most adverse clinical outcome [4]. Thus, from the clinical perspective, it is important to be able to predict which patients will experience a relapse. Traditional PCa prognosis models are based on some clinical features, such as pretreatment PSA levels, biopsy Gleason score (GS), and clinical stage, but in practice, they are inadequate to accurately predict disease progression [5]. With the development of microarray technology in recent years, a number of studies have been conducted to characterize the dynamics of gene expression in PCa progression by using DNA microarrays. In some studies, tumor expression signatures associated with clinical parameters and outcomes have been identified [6¨C9]. As a result, it is possible to develop the clinical models with the variables of gene signatures identified from microarray data and some clinical features to predict which men would experience progression to the metastatic form of PCa. However, it has been found that
%U http://www.hindawi.com/journals/cmmm/2013/917502/