Dear Colleagues,
I have got data for a large cohort of patients who received treatment A or treatment B. I would like to compare the impact of both treatments on patients’ outcomes.
There is a strong bias in patient selection for each treatment and therefore propensity scores are quite different between groups. When doing cox regression analysis adjusting for the propensity score, I did not find any difference between treatment A and treatment B, as I was expecting.
After doing propensity score matching, the results were similar. However, the resulting matched sample is not large enough according to a power analysis I have done. My questions are:
- Is there any role for bootstrapping in this case?
- Should I not use propensity score matching in this study and just rely on cox regression analysis with adjustment on the propensity score?
Many thanks
- Is there any role for bootstrapping in this case?
Bootstrapping will not solve a low power problem.
- Should I not use propensity score matching in this study and just rely on cox regression analysis with adjustment on the propensity score?
The highest power would be attained using Cox regression with adjustment on the propensity score because the complete sample is used, so you could stick with that. It is however always a good idea to estimate the effect using different methods to evaluate the sensitivity of your result to the method used.
- Badges
- Science topic