05-19-2016 05:15 AM
Usually when I perform Cox-regressions (PHREG - two treatment groups) the estimated hazard ratio is often quite similar to the ratio of incidence rates in each treatment groups.
I now run a PHREG (2 treatment groups) which resulteted in a HR of 1.5. When I compared it with the ratio of incidence rate, this ratio resultated in 2.0.
I have validated all number oand calculations and I now wonder why they can differ so much? Non-constant hazards?
Thanks in advance
05-19-2016 01:37 PM
Non-constant for 2 groups implies non-proportionality, in which case, PHREG might not be the best way to analyze your results. A key assumption of Cox regression is proportionality of the hazard rates. Another comparison to check this out is to compare the HR from PHREG to the Kaplan-Meier (nonparametric) estimate you can obtain from LIFETEST.