Abstract

Behavioral Changes Following Cancer Diagnosis: A Comprehensive Mutinomial Regression Based Study Using Automatic Model Selection on HINTS 4 Data

Author: Kyle Anderson

Behavioral changes following cancer diagnoses have not been studied in depth yet. Identifying the important predictors of behavioral change following cancer diagnoses can help fine tune public health messages and campaigns which can ultimately lead to improved survival odds, quality of life and increased life span for these patients. In this study we analyze the HINTS 4 data and investigate the behavioral changes of subjects following cancer diagnosis. We design and develop a novel comprehensive method for analysis of large complexly sampled survey data. HINTS 4 is a nation-wide, cluster sampled data with 50 jackknife replicates containing survey responses on 371 items of 3,677 subjects. We first perform a dimensionality reduction preliminary steps that use principal components analyses. We subsequently implement a multinomial logistic regression modeling combined with automatic variable selection algorithm to derive the best descriptive mode. Our approach provides the most sophisticated and precise model for behavioral changes following cancer diagnosis that improves our understanding of human behavior and allows for creation of more efficient public health messages.

Co Author/Co-Investigator Names/Professional Title: Cyril Rakovski-Associate Professor of Statistics