I would like to ask whether it is possible to have a retrospective survival analysis.

We have a population of patients who had an examination twice (one now and one in past with varying time intervals). We measure at the time of the latter examination a specific gene, which of course is a constant factor through time and we are interested in a finding of the examination.

Is it correct to run a survival analysis for that finding between the gene groups? The concept is that the population basically existed in the first examination although formed by our measurement in the second.

  • $\begingroup$ What hypothesis do you want to test? What would be the event that you're interested in? $\endgroup$
    – JonB
    Commented Nov 10, 2015 at 13:14
  • $\begingroup$ The event is a very well defined state of a disease, binary in nature. Because we are interested in progression in time, the event is defined by differences between two investigations. $\endgroup$ Commented Nov 11, 2015 at 21:19

1 Answer 1


So you have a number of patients of which some have developed the disease during the time between the assessments. You have measured the specific gene of interest, and as you say, this is of course a static factor and may be used as a baseline variable.

Now, there are basically two possible types of information about the disease. The first one is that the patient is re-examined as planned and you discover the disease. You then know only that the disease has developed during the time interval, not the approximate time. The second type of information about the disease is that the patient develops symptoms of the disease and seek medical care for this, and then get their diagnosis. In this case, if you have access to the medical records, you have a better approximation of when the diseased developed.

If there are none or few of this second type (called interval-detected recurrences), you can apply standard methods for survival analysis according to Collett, 2014. If not, you need to apply other methods that I'm not currently familiar with. The answers to this question give some ideas, provided you use R.

  • $\begingroup$ Very interesting! I read it and it describes very well the issue, "the time interval will be a mixture of recurrencies detected at scheduled screening", but it is not the case for us. All patients had acute symptoms in the second investigation and medical record review was done to define an old benign lesion that progressed to a clinical syndrome. So we know then the exact time, or nearly for those patients that may delayed some hours or 1-2 days in the case of mild symptoms (but we arbitrary ignore it). But the "interval-detected recurrence" adhere in many other situations i think! Thank you! $\endgroup$ Commented Nov 12, 2015 at 20:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.