Propensity Score matching

[#1]

Is it possible to generate Kaplan–Meier survival curves for the two matched groups after propensity score matching using JMP Student Edition 19?

According to information on the JMP Community website, propensity score matching is available starting from JMP Student Edition 19, and I was able to successfully complete the propensity score matching. However, I could not determine how to compare survival between the two matched groups (i.e., how to create Kaplan–Meier curves based on the matched pairs).

I would greatly appreciate your guidance on this matter. In addition, if this type of survival analysis is possible in JMP Pro but not in the Student Edition, I would be grateful if you could also clarify that point.

JMP Student Edition19では傾向スコアマッチング後に、マッチングさせた2群でのカプランマイヤー曲線を作成することは可能でしょうか？

JMP Communityで調べますと、JMP Student Edition 19から傾向スコアマッチング可能とあり、傾向スコアマッチングは完了したのですが、マッチさせた２群での生存率の比較(カプランマイヤー曲線)の仕方がわかりませんでした。

ご教授頂けますと幸いです。もし、Student Editionでなく、JMP Proであれば解析可能などもわかりましたら、教えてください。

[#2]

hi, and welcome to the community!

I just released this from the spam trap, but you probably want to re-post your question in https://community.jmp.com/t5/Discussions/bd-p/discussions, "Discussions" , at the top left of the menu bar. I think this group is not monitored much and is for discussions about the community rather than discussions about the software.

 (No idea what made it go into the spam trap...)

[#3]

Thank you for replying.

I try to send the comment in "Discussions" , at the top left of the menu bar.

[#4]

Thanks for sharing this question. I was also curious about performing survival analysis after propensity score matching in JMP Student Edition. It would be really helpful if someone could explain whether Kaplan–Meier curves can be generated directly from the matched data or if there is a specific workflow to follow. Looking forward to insights from experienced users.

[#5]

Hi.
not sure if you've been able to figure out a solution here. But there isn't a 'matched pairs' option in the Kaplan Meier analysis. But... depending upon your data, it could be possible to create a new variable that is the 'difference' between the two values for each subject (or whatever is the group in your data). you'd just need to make sure properly defining what is censoring if you have any of that in your data. This is analogous to how a one-sample t-test on the mean difference is the same as doing a paired t-test.
Alternatively, you could use the "parametric survival" platform in "Fit Model". there you'd be specifying your grouping variable as if it's a block. This platform however will be doing Kaplan Meier but you'll be choosing an appropriate distribution (e.g., Weibull).
Hope this helps

[#6]

Hello,

I think that the short answer must be "No, JMP Pro or JMP Student Edition cannot do propensity score matching  for survival analysis".

The features in JMP Pro and JMP Student Edition are same. So, JMP Student Edition can do what JMP Pro can do, and JMP Student Edition cannot do what JMP Pro cannot do.

I think the long answer must become very complicated.

(1) If you think it is enough to do "usual" survival analysis (like "usual" log-rank test, and "usual" Cox hazard for estimating hazard ratios), you can do it on JMP Pro and JMP Student Edition. But I personally think that it is not enough.

(2) You can invoke R from JMP. But it needs some programming.

As regards to (1), the story is very complicated. At first, there is some critics for using propensity score mathing (even for usual linear models) itself. For example, See a book by Shinozaki, Hagiwara, Taguri and Matsuyama (ISBN 978-4-254-12312-8),  p.97 (Japanese).  In mathematical statistics, AIPW (augmented inverse probability weighting. a.k.a. doubly robust estimation) for cross-sectional data (like linear models) is explained and summarized very well with relatively simple theories. For example, see Tsiatis' book (ISBN  978-0-387-32448-7) if you have some advanced knowledge about mathematical statistics. But the propensity score matching is very and very difficult for checking the validness.

As far as I know, there is no consensus for the "best" propensity score matching. Someone think 1:1 matching without replacement is the best, someone think 1:1 matching with replacement is the best, the other ones think 1:m matching with/without replacement is the best.

Even if you think the propensity score matching is valid as your causal inference, there are still other problems. As far as I know, there is no consensus for the best method for estimating standard errors (even in cross sectional analysis like linear models). Some references are written on pp.96-97 in Shinozaki et al.  For example, if you do 1:1 matching without replacement, someone think that you do not need to include the matching variable, but the others don't.

So, I personally think you need to decide the main propensity score mathing analysis in advance, and you need to perform some different propensity score matching settings as sensitivity analyses. But JMP can do only "usual" log-rank test or "usual" Cox's proportional analysis for estimating hazard ratios. so you do not do  them.

If you think that you perform “usual” logrank test for 1:1 matching data, I would like to explain the procedure. But I personally think the method is not enough for estimating causal effects.

JMP Pro or JMP Student Edition cannot do REGADJ, IPWR or AIPW for survival analysis, either.