I have two groups of patients who underwent a surgery using method A or method B. The first group are patients who were operated in 1980's and 1990's only with method A. The second group are patients operated recently with mostly method B, but also in some cases A. In addition to that, I have various variables about patients (gender, age, medical indicators, etc.) that capture the pre-operation medical history and types of symptoms that patient developed.

The goal of the study is to compare the "effecitveness" of methods A and B in terms of patients' survival times after operation.

Somewhat different patients (in terms of age, gender, etc.) are operated with A and B. For example, quite some more older people were operated with B than with A. I want to used propensity score matching to balance the data.

My question is:

Does it make sense to estimate propensity scores (method ~ age + gender + ...) and use them for creating a matched dataset for further analysis (e.g. Cox regression)?

In particular, is it a problem that for patients in the first group method B was not yet available, so none of them could potentially received the alternative treatment?

  • $\begingroup$ First you need to carefully define what the "effectiveness" metric will be. $\endgroup$ – Alecos Papadopoulos Oct 17 '13 at 10:40
  • $\begingroup$ What is 80' and 90'? Does this stand for minutes? $\endgroup$ – Momo Oct 17 '13 at 10:40
  • $\begingroup$ @AlecosPapadopoulos, patients were followed for several years. I want to compare survival times. $\endgroup$ – Michał Oct 17 '13 at 11:01
  • 1
    $\begingroup$ No it is not a problem that method B was not available. This fact does not affect any inference related to survival times. You are not trying to explain "why they chose A over B", but given the choice (for whatever reason), what was the survival time. $\endgroup$ – Alecos Papadopoulos Oct 17 '13 at 12:37
  • 1
    $\begingroup$ The fact most of the A surgeries occurred decades earlier than the B surgeries could imply that you can't really compare the two. The fact the time period is so radically different can cause significant confounding. Other seemingly unrelated treatments and life in general has changed a lot over such a long period, which will also reflect on your survival results. Don't underestimate this hurdle. $\endgroup$ – Marc Claesen Oct 18 '13 at 15:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.