Date of Award

January 2016

Document Type

Open Access Thesis

Degree Name

Master of Public Health (MPH)

Department

School of Public Health

First Advisor

Zuoheng Wang

Second Advisor

Anthony Kim

Abstract

Current TNM staging system derived empirically from human papillomavirus (HPV) unrelated oropharyngeal cancer (OPC) has been shown inadequate to predict survival for HPV-related OPC. This study used three recursive partitioning algorithms, Classification Trees (CART), Conditional Inference Trees (CTree) and Model-based Recursive Partitioning (MOB) to derive a new staging scheme based on data from the National Cancer Data Base (NCDB). The derived staging systems were compared to the current system using the criteria such as hazard consistency within staging groups, hazard discrimination between groups, predictive ability and balance of distribution across groups. A total of 5,712 patients were included in the analysis. The staging system derived using the model-based recursive partitioning (MOB) has the best predictive ability and overall performance. It separates patients into four stages: Stage I (T1-2N0-2a), Stage II (T1-2N2b-3), Stage III (T3), and Stage IV (T4). Stage V is reserved for metastatic patients (M1). The theoretical advantages for the MOB algorithm of fitting the local parametric model in each node and adjusting for covariates affecting survival were confirmed with empirical analysis. Thus MOB algorithm is recommended for future TNM cancer staging studies.

Comments

This is an Open Access Thesis.

Share

COinS