Hi Shiyam, the DSMB probably considered that a statistically significant ARI of 3.1% in the placebo group was sufficient to reject your stated null hypothesis of

If I am reading this correctly, there is a 95% confidence that all such trials would have between a 0.3% and 5.7% 'risk difference,' where the point estimate here for failure (however that was defined) is less than half for the antibiotic, as opposed to the placebo.  A 95% confidence interval of this difference, where the lower end is above zero, and the top end is near 6%, sounds impressive here, to me.  

I think emphasis should be on confidence intervals, as opposed to hypothesis tests, as p-values are functions of sample size, and 'acceptance' levels need to be adjusted accordingly.  It is too easy to adjust the sample size to obtain the result someone wants to see.  An exception to this objection would be a sequential hypothesis test, where two 'simple' (point) hypotheses are compared, which is an efficient use of resources, 'on average,' but the sample size needed is not known ahead of time.  

In your work, I would imagine the occurrence of deaths would tend to shut things down sooner than expected, on occasion, but you can still get a confidence interval.  Just having the standard error of a point estimate can be very informative.  

Dear James,

I agree with you. By both the 95% CI and the test of significance p-value, the study risk difference can be deemed to be greater than 3 %. The only DSMB deal-breaker was the two deaths in the placebo group. I apologise for my earlier misinterpretation of the results. I hope I got it right this time.