Appears pretty linear - one more death in control allows 3 more total deaths to maintain significance, Again, the issue is that the 87 cited deaths aren't final - pending last week (+) and deaths post 28 days (-).

Control Treatment Total P(two sided)
35/131 . . 46/262 . . 81 . . .0429
36/131 . . 48/262 . . 84 . . .0452
37/131 . . 50/262 . . 87 . . .0475
38/131 . . 52/262 . . 90 . . .0497

Interpreting this: in this range, sig holds even if added deaths are pro-rata - that is, not showing LL benefit.

This range of control deaths (26.7% to 29.0%) is well toward the low end of what has been on the board lately for s/c populations.

Help from other geeks, please. This board has been using the totally-valid difference in proportions test. However, everything I'm seeing reported has used "relative risk." Can someone do the table on that basis? I tried a couple of cases once and differences were in fractions of a death.

(For real geeks, this is just the difference test on the logs of the proportions.)