數學系演講公告

主講 人： 陳珍信 教授
中央研究院統計科學研究所
台灣大學公共衛生學院流行病學與預防醫學研究所
         

講    題：Location-Scale Regression Models with Non-Susceptibility in Event History Analysis
日    期：99年11月09日（星期二）
時    間：下午3:30-4:20

地    點：數學系(科學館S433室)

Abstract
In population-based studies and genomic medicine research, it is important to 
delineate genetic, environmental effects and their interactions on specific 
complex diseases/disorders.  Statistical methods for analyzing this kind of data 
have been developing. In most of the cases, outcomes of the disease/disorder 
under study are treated as either a dichotomous status or a continuous 
measurement of age at onset.  The former approaches do not consider the 
probability of later onset for observed non-cases, while the latter approaches 
ignore the disease non-susceptibility in the non-cases, who did not inherit the 
related genes or were never exposed to deleterious environmental factors. 
These two kinds of approaches may render misleading interpretations.

In this presentation we briefly review the conventional survival analysis which 
implicitly assumes all the study subjects are susceptible to the event.  To 
analyze the non-susceptibility, recent studies proposed mixture regression 
models to investigate the respective risk factors for the probability of disease 
susceptibility and the age-at-onset distribution simultaneously for right censored 
data.  In epidemiological studies with longitudinal data, we often encounter left, 
interval and right censored data, as well as possible left truncated data due to 
different entry ages of recruited healthy subjects.  The resultant survival curves 
may emerge plateaus on right tails and multiple crossings over covariate strata.  
To tackle these issues of incomplete data, we present the mixture regression 
model combining the logistic model with the accelerated failure time 
location-scale model.  We also apply the model to analyze age-at-onset studies 
in a few epidemiological research projects in Taiwan. A three-component 
mixture regression model is extended to study the willingness-to-pay for 
non-market health services using double-bound dichotomous choice contingent 
valuation surveys.

 

 

歡 迎 參 加           敬 請 張 貼

淡江大學數學系 敬啟