Hello everybody,
This is more or less a statistical question. My apologies if it seems too basic.
One of the means whereby non-proportionality can be assessed in Cox regression is to enter an interaction term (time*covariate) in the model. I was wondering if it possible to interpret the 'main effect' term in the presence of the interaction term? For eg. the following results are from a randomized clinical trial assessing headache relief among two groups randomized to 'placebo' or 'drug' (follow-up is over 5 time intervals: 0, 12, 24, 48, 72 and 168 hours, thus data is interval censored): After st setting etc., following are the preliminary results:
1.
stcox drug
variable hazard ratio p-value
drug 1.09 0.20
2. With the interaction term (ln(_t)*drug), I get the following results using:
stcox drug, tvc(drug) texp(ln(_t))
drug 13.01 0.05
ln(_t)*drug 0.4 0.05
indicating that the hazards are not proportional. The solution would then be to stratify on time intervals (in the simplest scenario i.e. absence of other covariates). Results showed that hazard ratios were high (about 2.4)only for time interval 1 and progressively decreased.
However, my question is whether one can interpret the 'hazard ratio' associated with 'drug' in model 2 above? If yes, what would it be?
Many thanks in advance.
Devendra Amre
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/