University of South Florida Scholar Commons Graduate Teses and Dissertations Graduate School 10-30-2014 A Decision Support Model for Personalized Cancer Treatment Florentino Antonio Rico-Fontalvo University of South Florida, A Decision Support Model for Personalized Cancer Treatment by Florentino Antonio Rico-Fontalvo A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Industrial Engineering Department of Industrial and Management Systems Engineering College of Engineering University of South Florida Major Professor: Grisselle Centeno, Ph.D. Steven Eschrich, Ph.D. Javier Torres-Roca, M.D. Ali Yalcin, Ph.D. Jose Zayas-Castro, Ph.D. Date of Approval: October 30, 2014 Keywords: Supervised Learning, Fuzzy Logic, Systems Biology, Gene Expression, Rectal Cancer, Random Forest Copyright © 2014, Florentino Antonio Rico-Fontalvo DEDICATION I want to dedicate this dissertation to my parents: Martha Cecilia Fontalvo-Rivera and Florentino Antonio Rico-Calvano: you are the source of my drive, inspiration and motivation. Thank you Mom and Dad, none of this could not have been possible without your support, love and encouragement. This is a tribute to both of you. ACKNOWLEDGMENTS I would like to thank my mentor and major advisor, Dr. Grisselle Centeno, who has been with me through this journey. You believed in me from the very beginning, and it has been an honor working with you. Your role as an advisor goes beyond this dissertation, I have no words that can express my gratitude. I sincerely thank you for giving the opportunity of achieving my greatest academic accomplishment. I cannot express enough thanks to my committee for their continued support and encouragement. Dr. Zayas-Castro, thank you for being an integral part of my academic journey. I will always be grateful for your mentoring, academic, professional and financial support I received from you during the last years, I will always carry your advice in the journey I have ahead. Dr. Yalcin, thank you for your friendship and the learning opportunities I had with you. Dr. Eschrich and Dr. Torres-Roca, I offer my sincere appreciation for sharing and contributing so much to my research, thank you for opening a new world of knowledge and believing in my potential, my heartfelt thanks. The completion of this work could not have been possible without the support of Gloria Latter, Liz Conrad and Catherine Burton. Thank you for your help and having patience with me during this dissertation journey. To my brother and sisters: Jorge, Heidy and Ximena, you are an inspiration and my role models. My nephews and niece: Camilo, Alejandro and Juliana, I always carry you in my mind. Finally, to Santiago: my deepest gratitude. It was your encouragement when the times got rough that kept me going. Thank you for your patience and unconditional support, know that it is much appreciated. TABLE OF CONTENTS LIST OF TABLES iii LIST OF FIGURES iv ABSTRACT vi CHAPTER 1: INTRODUCTION 1 1.1 Background 1 1.1.1 Rectal Cancer Diagnosis 3 1.1.2 Staging 4 1.1.3 Treatment Options 5 1.1.4 Adverse Effects of Radiation Treatment 6 1.2 Personalized Medicine 7 1.3 Patient-Centered Decision Making 8 1.4 Review of Literature 9 1.5 Problem Statement 11 1.6 Global Research Objectives 13 1.7 Document Organization 13 CHAPTER 2: PREDICTION OF RADIOSENSITIVITY OF CANCER TUMOR CELLS IN RESPONSE TO RADIATION THERAPY USING GENE EXPRESSION PROFILES 15 2.1 Introduction 15 2.2 Review of Prediction Models in Computational Biology 17 2.3 Objectives 20 2.4 Methods and Materials 21 2.4.1 Output 21 2.5 Feature Selection 23 2.6 Predictive Model Development 24 2.6.1 Multivariate Regression with 2-way Interactions 26 2.6.2 Classification and Regression Trees 28 2.6.3 Random Forest 30 2.7 Validation 32 2.7.1 Rectal Cancer Dataset 33 2.7.2 Esophageal Cancer Dataset 34 2.8 Discussion 35 CHAPTER 3: A FUZZY APPROACH FOR TREATMENT SELECTION IN CANCER TREATMENT 36 3.1 Concepts in Fuzzy Logic 37 3.1.1 Fuzzy Inputs and Outputs 38 i 3.1.2 The Fuzzy State Space 39 3.2 Review of Related Literature 40 3.3 Objectives 42 3.4 Hypotheses 42 3.5 Fuzzy Inference System Approach 42 3.5.1 State Transitions Matrices 44 3.5.2 Membership Functions 46 3.5.3 Input Data 47 3.5.4 Measure of Preference 54 3.5.5 Sensitivity Analysis for Radiosensitivity 58 3.6 Discussion 59 CHAPTER 4: CONCLUSIONS AND FUTURE RESEARCH 60 4.1 Conclusions 61 4.2 Future Research 61 REFERENCES 63 APPENDICES 72 Appendix A Rectal Cancer Detection and Staging 73 Appendix B Figure Permission 74 Appendix C SEER Data Use Agreement 75 Appendix D SEER Database Variables Used 76 Appendix E Parameter Estimates for the Logistic Regression 77 Appendix F Transition Probabilities for Adverse Effects and Efficacy 78 ii LIST OF TABLES Table 1 Survival rates for rectal and colon cancer by stage 2 Table 2 Summary of cancer treatment selection models in the literature 12 Table 3 Summary of prediction models in computational biology 18 Table 4 SF2 measured values for 48 cell lines in the database 22 Table 5 Multivariate regression model selection 27 Table 6 Decision model elements and membership functions 45 Table 7 Patient cohort descriptive statistics 48 Table 8 Cancer and tumor stage statistics 49 Table 9 Treatment options 49 Table 10 Logistic regression chi-square values for selected variables 50 Table 11 Odds ratio estimates for logistic regression 51 Table 12 Criteria used to grade toxicity from radiation therapy 52 Table 13 Example of predicted patient clinical parameters 53 Table 14 Survival transition matrices 53 Table 15 State vectors for all treatment options and clinical parameters 54 Table 16 Simulation of various preference scenarios 55 Table D.1 SEER database variables used 76 Table E.1 Parameter estimates for the logistic regression 77 Table F.1 Transition probabilities for adverse effects and efficacy 78 iii LIST OF FIGURES Figure 1 Diagram of colon and rectum 1 Figure 2 Rectal cancer detection and staging process 3 Figure 3 Dissertation organization 14 Figure 4 SF2 and transformed SF2 21 Figure 5 Experimental design 25 Figure 6 Model performance in terms of adjusted R-square 28 Figure 7 Decision tree prediction model 30 Figure 8 Variable importance based on entropy reduction 31 Figure 9 Random forest algorithm 32 Figure 10 Multivariate regression prediction results on the rectal cancer dataset 33 Figure 11 Random forest prediction results on the rectal cancer dataset 33 Figure 12 Multivariate regression prediction results on the esophageal cancer dataset 34 Figure 13 Random forest prediction results on the esophageal cancer dataset 34 Figure 14 The characteristic function of a crisp set (a) and the membership function of a fuzzy set (b) 38 Figure 15 Degree of membership of the crisp value to the fuzzy value of the fuzzy state variable 39 Figure 16 Fuzzy inference system approach 44 Figure 17 Membership functions in terms of survival, adverse events and efficacy 46 Figure 18 Pre-modeling and knowledge extraction data processing steps 47 Figure 19 Results of simulation of various preference profiles 56 Figure 20 Sensitivity analysis based for survival 57 iv Figure 21 Sensitivity analysis based for efficacy 57 Figure 22 Sensitivity analysis for various treatment efficacy levels 58 Figure A.1 Rectal cancer detection and staging 73 v ABSTRACT This work is motivated by the need of providing patients with a decision support system that facilitates the selection of the most appropriate treatment strategy in cancer treatment. Treatment options are currently subject to predetermined clinical pathways and medical expertise, but generally, do not consider the individual patient characteristics or preferences. Although genomic patient data are available, this information is rarely used in the clinical setting for real-life patient care. In the area of personalized medicine, the advancement in the fundamental understanding of cancer biology and clinical oncology can promote the prevention, detection, and treatment of cancer diseases. The objectives of this research are twofold. 1) To develop a patient-centered decision support model that can determine the most appropriate cancer treatment strategy based on subjective medical decision criteria, and patient’s characteristics concerning the treatment options available and desired clinical outcomes; and 2) to develop a methodology to organize and analyze gene expression data and validate its accuracy as a predictive model for patient’s response to radiation therapy (tumor radiosensitivity). The complexity and dimensionality of the data generated from gene expression microarrays requires advanced computational approaches. The microarray gene expression data processing and prediction model is built in four steps: response variable transformation to emphasize the lower and upper extremes (related to Radiosensitive and Radioresistant cell lines); dimensionality reduction to select candidate gene expression probesets; model development using a Random Forest algorithm; and validation of the model in two clinical cohorts for colorectal and esophagus cancer patients. vi