What inspired this wish list request?
We routinely perform survival analysis on multiple treatment groups (strata). The Kaplan-Meier analysis provides Log-rank and Wilcoxon statistics that allow evaluation of differences among all groups. When the model shows significance, we want to follow with posthoc analysis of pairwise comparisons of all groups with an appropriate alpha correction. This would be analogous to running an ANOVA followed by Tukey's HSD with Bonferroni correction. In SAS, we can run logrank tests over strata with Sidak correction.
What is the improvement you would like to see?
Having pairwise comparisons would make the survival analysis more powerful with finer resolution of which treatment groups are different from other treatment groups.
In SAS: STRATA 'grouping variable'/ test=logrank adjust=sidak;
Output: (xxxx and yyyy represent various groups (strata))
| Test of Equality over Strata |
| Test |
Chi-Square |
DF |
Pr >Chi-Square |
| Log-Rank |
259.7179 |
4 |
<.0001 |
| Adjustment for Multiple Comparisons for the Logrank Test |
| Strata Comparison |
|
p-Values |
| Group |
Group |
Chi-Square |
Raw |
Sidak |
| xxxxx |
yyyyyy |
234.6 |
<.0001 |
<.0001 |
| xxxxx |
yyyyyy |
17.1415 |
<.0001 |
0.0003 |
| xxxxx |
yyyyyy |
3.5276 |
0.0604 |
0.4634 |
| xxxxx |
yyyyyy |
1.2451 |
0.2645 |
0.9537 |
| xxxxx |
yyyyyy |
82.0587 |
<.0001 |
<.0001 |
| xxxxx |
yyyyyy |
155.3 |
<.0001 |
<.0001 |
| xxxxx |
yyyyyy |
169.1 |
<.0001 |
<.0001 |
| xxxxx |
yyyyyy |
4.8642 |
0.0274 |
0.2427 |
| xxxxx |
yyyyyy |
8.5731 |
0.0034 |
0.0336 |
| xxxxx |
yyyyyy |
0.5678 |
0.4511 |
0.9975 |
A separate analysis in SAS can assign letters to each group to denote level of significance.
Why is this idea important?
Every experiment with multiple treatment and control groups ultimately requires analysis of differences between groups. With a handful of groups this can be done manually. But with many groups this is tedious.