## Due Wednesday, 4 December 1996

In Homework 8, problem 3, you used logistic regression methods to analyze data on the occurrence of heart attacks among 35 individuals with high cholesterol levels (Table 11.17 in Selvin). The aim was to find a predictive model for the risk of heart attack. Specifically, you were to assess the risks associated with weight, cholesterol level, and cigarette usage. For Homework 9, reanalyze the data using appropriate survival analysis methods, as follows:
1. Construct a new variable heavy which indicates whether an individual smokes more than 20 cigarettes per day. Assess whether heart attacks are more likely for heavy smokers than for light smokers by constructing the Kaplan-Meier (product-limit) estimates for the survival curves in the two groups.
2. Construct a proportional-hazards regression model for survival. Is heavy or cigs (the number of cigarettes smoked per day) a better predictor?
3. Compare the results from the logistic analysis and the Cox analysis. Describe how the models and the conclusions differ from one another.
4. One patient appears to have had a particularly high cholesterol level. Does this affect either the logistic or the survival analysis?

Hints and comments:

• kapmeier T C, by(G)
calculates the product-limit estimator of survival-time distributions for each value of G, and plots all of the curves on the same graph. In the command above, T is the survival-time variable, and C indicates whether that time corresponds to a death (C=1) or censoring (C=0).
• kapmeier has been replaced in version 5.0 of Stata. If you are using that version, you can temporarily revert to old-style Stata commands by giving the command version 4.0, after which commands like kapmeier will once again be recognized.
• The proportional hazards model is often referred to as "Cox regression," after Sir David Cox who first introduced it. A Stata command to do this calculation looks like this:
cox Y X1 X2 ... Xp, d(C)
where Y is the outcome (survival time), X1 through Xp are the predictors, and C is an indicator that is 1 if an event was observed (death, for instance) and a 0 if the observation was censored.
• The cox command also has been replaced in Stata 5.0 and later versions.

Last updated 26-Nov-96
For questions, send mail to r-thisted@uchicago.edu or mcpeek@galton.uchicago.edu