Graduation Year


Document Type




Degree Granting Department

Chemical Engineering

Major Professor

John J. Heine, Ph.D.

Committee Member

William E. Lee, Ph.D.

Committee Member

Maria Kallergi, Ph.D.

Committee Member

Steven Eschrich, Ph.D.

Committee Member

Srinivas Katkoori, Ph.D.


Survival Analysis, Differential Evolution, TMA, DNA Repair, Statistical Learning


Some early stage NSCLC patients have a better survival prospects than others. In any event, the long-term prognosis for NSCLC patients is poor. Various measures were investigated to gain a better understanding of those patient characteristics that confer better survival or predict disease recurrence. A dataset comprised of stage 1 NSCLC patients (n=162) that underwent resection was investigated. Clinical variables (CVs) and tissue microarray (TMA) images with DNA repair protein and standard H&E expressions were investigated. Patients were dichotomized into two groups by survival characteristics and logistic regression (LR) modeling was used to predict favorable survival outcome. Various patient strata were investigated with Cox regression and Kaplan Meier survival analysis (i.e. accepted survival analysis methods). A statistical learning (SL) method comprised of a kernel mapping and Differential Evolution optimization was developed to integrate SL techniques with LR and accepted survival analysis methods by first combining various patient measures to form a hybrid variable. Younger age, female gender, and adenocarcinoma subtype confer better survival prospects, whereas recurrence confers poor survivability. The SL hybrid modeling produced greater favorable outcome associations and survival hazard relationships than the accepted approaches. Automated texture measures from the HE stained TMA images were significantly related to survival, tumor-type, and tumor-grade. DNA repair measures in isolation or in combination with CVs were not related to survival, favorable outcome or recurrence, and none of the CVs were related to recurrence.

A platform was established to incorporate automated TMA analysis and SL techniques into standard epidemiologic practice, and baseline predictive models were constructed. Future work will investigate novel biomarkers and larger datasets using this established framework to construct prognostic models for clinical applications for lung cancer patients in general and to better understand disease recurrence.