Cel Sci Corporation (CVM)

[przgwxl]

@Fosco: stopping the trial early

I found a paper that cited criteria for stopping early for trial success. It mentions O'Brien-Fleming stopping criteria of p<0.0138 and hazard ratio of < 0.739.

My computations, based on Fosco's "mixed" model, shows the p-value below 0.0138 in September, 2019 (162 events in the control arm). However, the hazard ratio is 0.84, so the early stopping criteria were not met. There is the same issue for 1 October 2019. So IMO the IDMC was correct not to stop the trial. The O'Brien-Fleming criteria were not met. Things were looking very good, but these criteria are deliberately very conservative.

Fast forward and assume the trial ended today. I get 168 events in the control arm, implying 130 events in the treatment arm. The p-value is <0.0006 and the hazard ratio is 0.774. The hazard ratio is still above the criterion.

Three points here:

1) the IDMC was probably right not to stop the trial. Although the p-value was looking pretty good(passed 5% sometime August, 2019(, the hazard ratio was not low enough to stop the trial early.

2) The hazard ratio criterion is pretty severe. A 10-20% treatment effect for OS is never going to meet this criterion.

3) If Fosco's survival numbers (very conservative/optimistic), and a 18.5% dropout rate are correct, the treatment effect is 20.5% with plenty of statistical power. Still not enough to stop the trial early, but a no-brainer in terms of satisfying the criterion for a positive trial.

Fosco's survival model is much more optimistic than the Cel-Sci SEER model, almost 25% higher at 3 years. Other boards have mentioned CVM inclusions/exclusions might make SEER pessimistic for survival, so this may be appropriate, but presumably the Cel-Sci consultants tailored the Cel-Sci model based on these factors. So Fosco's model is very optimistic about survival, even relative to the model presented by Geert.