Is differential follow up periods in survival analysis a problem?

[ahmed farhan]

Hi

I'm looking to compare mortality post-operatively following two different surgical techniques. due to technological advances, one of these surgical techniques was only performed around five years after the other one. however both techniques are now commonly practiced. as a result, i have a group of patients whose follow up period lags around five years behind the other. is this a problem when doing kaplan meier survival analysis or computing a hazard ratio using cox regression? the kaplan meier plot appears five years longer for one intervention than the other.

thanks

[Clyde Schechter]

When the follow-up period is shortened, the consequence for survival analysis is that the observation is censored earlier than it otherwise would be, and perhaps a precise failure (death) time is missed as a result. This decreases the precision of the estimates in the group with shorter survival, but it does not introduce bias in the hazard ratio estimate.

Nevertheless, I think that it is unwise to do this comparison for a different reason. It is entirely possible, in fact, I would say likely, that mortality is different during that first five year period just because the passage of those 5 years may well be accompanied by changes in health and health care that are unrelated to the procedures you are studying. Indeed, mortality rates are notoriously known to have important secular trends that would confound any comparison of effects that occurred in two different "eras." If we were talking about a shorter period of time, say a few months, rather than 5 years, this could probably be ignored. But five years is certainly long enough for a change in mortality rates having nothing to do with your study procedures to have occurred.

So what I would do is compare only cases from the era when both procedures were in use. That's a more apples-to-apples comparison.