Gene Profile Modeling and Integration for EWOC Phase I Clinical Trial Design While Fully Utilizing All Toxicity Information Open Access

Tian, Feng (Spring 2020)

Permanent URL:


Background: Personalized medicine incorporating genomic profile has become the frontier in modern medicine. It is especially profound in optimizing healthcare for cancer patients because many gene mutations significantly affect cancer progression and efficacy of treatments. Personalized Maximum Tolerated Dose (pMTD) estimation in Phase Ⅰ clinical trial is the initial key step to integrate genomic profile into personalized medicine.

Methods: Considering the limited sample size of phase Ⅰ trails, selecting a small number of representative gene mutation profiles is required to keep the pMTD estimation valid. The main aim of this study is to achieve above goal by performing variable selections and comprehensive index construction using four common methods: model selection with logistic regression, regularization, principle components analysis, and random forest.

Results: The results of four methods are compared in the consistency of selected genes, the simplicity and the variety in dose estimation using EWOC-NETS (escalation with overdose control using normalized equivalent toxicity score) framework. We found that different methods are fairly consistent in selecting the important genes. The Elastic Net method is the optimal one to generate a model with simplicity and precise dose estimation in predicting tumor response.

Conclusion: For future pMTD estimation, it is a good idea to use Elastic Nets as a main reference and the common elements recommended by other mentioned methods as additional support to decide the required representative gene information to be incorporated into EWOC-NETS. The extracting and incorporation of summary genomic data will have great potential to improve treatment precision and trial efficacy.

Table of Contents

1. Introduction. 1

2. Method. 3

2.1 Methods for gene profile modeling. 3

2.1.1 Logistic Model Selection. 4

2.1.2 Regularization Methods. 5

2.1.3 Principle Components Analysis 6

2.1.4 Random Forest Analysis. 7

2.2 EWOC-NETS model that utilize gene information. 8

3. Results. 12

3.1 Results for factor selection. 12

2.1.1 Logistic Model Selection. 12

2.1.2 Regularization Methods. 13

2.1.3 Principle Components Analysis. 14

2.1.4 Random Forest Analysis. 15

3.2 Incorporation of Gene Profile to EWOC-NETS. 16

4. Conclusion and Discussion. 16

5. References. 19

6. Tables and Figures. 20

7. Appendix. 24

About this Master's Thesis

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
Subfield / Discipline
  • English
Research Field
Committee Chair / Thesis Advisor
Committee Members
Last modified

Primary PDF

Supplemental Files