This preview shows page 1. Sign up to view the full content.
Unformatted text preview: gression model?
Central question: How can we model separately diﬀerent factors
that are likely to impact the observed event, so that we can isolate
their individual eﬀects?
Heterogeneity: lives with very diﬀerent characteristics (eg
males and females, smokers and non-smokers) have diﬀerent
level of mortality. In other words, diﬀerent factors (covariates)
may have diﬀerent eﬀects on the risk.
One method is to construct a model including the eﬀects of
the covariates on survival directly: a regression model.
The most widely used regression model is the proportional
hazards model (the Cox model)
3/45 Actuarial Statistics – Module 3: Semi-parametric methods: Cox Regression Model
In many survival analysis problems, covariates are of the
Demographic / Societal (eg age, gender, education)
Behavioral (eg smoking, physical activity level, alcohol)
Physiological (eg blood pressure, heart rate) Covariates can be
continuous measurements (weight)
discrete measurements (age last birthday)
indicators (1 for smoking, 0 for non-smoking)
qualitative indicators (5 severe disease to 0 no symptoms) see K&M for a discussion of how to ‘code’ variables/factors
4/45 Actuarial Statistics – Module 3: Semi-parametric methods: Cox Regression Model
For the i th life we will denote the covariates (ris...
View Full Document
This document was uploaded on 04/03/2014.
- Three '14