topic Re: Pairwise comparison of KM survival curve in Discussions

Pairwise comparison of KM survival curve

CorrelationDeer — Fri, 09 Jun 2023 01:01:12 GMT

Hello JMP community,

I am currently analyzing a dataset that includes the survival of subjects under eight different diets. I conducted a log-rank test and found that there is a significant difference between at least some of the groups. Now, I want to perform post-hoc pairwise comparisons to determine which specific groups differ from one another in terms of survival.

Attached is a file that contains an example of the data I'm trying to analyze.

can anyone help me, please?

Thanks

Re: Pairwise comparison of KM survival curve

Jed_Campbell — Wed, 26 Apr 2023 15:39:05 GMT

I'm not sure I'm understanding completely what you're looking for, but this sounds a bit like a Reliability problem. With your data table, I used the Fit Life by X platform, assumed a Lognormal distribution, then from the nonparametric overlay plot I could see that White 2 was the only different distribution.

This isn't exactly a pairwise comparison, so I made the Estimates table into a data table and used Graph Builder to show intervals for location and scale, which is more like the pairwise comparison.

Is this what you're looking for?

Re: Pairwise comparison of KM survival curve

CorrelationDeer — Fri, 28 Apr 2023 09:34:04 GMT

Hi,

thank you so much first of all, but this is not what I'm looking for.

I am interested in comparing the means of survival days between pairs of groups.

this is the plot I get from: Analyze -> Reliability and survival -> Survival -> "Y, Time to event=Survival days" -> "Grouping=Diet"

According to the "Test between groups" there is significant deference:

My goal is to test these differences between each pair of groups (Pairwise comparison?).

I hope I'm clear now and sorry for the confusion :)

Re: Pairwise comparison of KM survival curve

Jed_Campbell — Fri, 28 Apr 2023 15:06:42 GMT

Ah, you can do this in the Fit Y by X platform. To get the Wilcoxon (and other nonparametric) pairwise tests, select the red triangle...Nonparametric...Wilcoxon Each Pair. For more traditional (mean/stdev) pairwise comparisons, select the red triangle...Compare Means...then one of the options (usually student's t or Tukey). I've attached a script to run this to your sample table.