0
$\begingroup$

Suppose I have two propensity matched groups whose covariates are balanced. How do I analyze clinical outcomes between the groups. I would be interested in overall survival, one year survival, and incidence of adverse events such as stroke.

  • Can I use standard generalized logistic regression models and kaplan meier curves with log rank test?
  • Do I need to treat the clinical subjects as paired and adjust my analysis appropriately (i.e stratified log rank and Mcnemar test)

Thank you!

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

This is debated. Peter Austin strongly recommends accounting for the paired nature of the matched groups. Other methodologies do not. I think it makes sense to do so. Generally, power will be increased if within-pair distances are low. The differences will not be stark, however.

Improve this answer
$\endgroup$
1
  • $\begingroup$ I have two matched sets with n = 50. I've computed standard kaplan meier curves and a logrank test with p ~ 0.7. After, I tried to account for matches in 2 ways. In the first, I utilized the survdiff() function in the survival package with a cluster(ID) variable to account for pairing. In the second method, I ran the same thing with a coxph function and a cluster(ID) variable. survdiff gives me a value of p <0.0001 whereas coxph gives me a p value close to the baseline 0.7 with a slightly different robust standard error. Which should I trust. Does cluster(ID) not work with survdiff? $\endgroup$
    – chill_hakawati
    Commented Mar 28, 2020 at 21:01

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.