It is widely accepted that analyses of clinical trials have to be adjusted if interim analyses are planned in order to decide if early termination is necessary due to evidence for superiority of one of the studied interventions. There are a number of prospective statistical strategies for stopping clinical trials early because of overwhelming evidence of efficacy [1]. However, the focus seems to be on multiple (sequential) analysis of one outcome of interest.

My questions:

  • Is it necessary to adjust the analysis of the primary efficacy outcome, if interim analysis is planned on another outcome, say on a safety outcome?

  • If yes, what are possible strategies to follow?

[1] Sankoh AJ. Drug Information Journal, Vol. 33, pp. 165-176, 1999.


It depends on the purpose of the interim analysis and type of trial. In general, adjustments for interim analyses are potentially needed, if it is desired to control the type I error rate. If the interim analysis is for futility or safety stopping only (i.e. the trial may be abandoned without rejecting any null hypotheses, either because it seems unlikely the trial goals will be achieved, or because interventions do not appear to be safe), then technically an adjustment is not necessary.

However, some (e.g. Pocock) have argued that some alpha should always be spent to avoid situations, where those reviewing the data suggest stopping for demonstrated efficacy (even though it was not planned that this was possible). If some alpha is spent (even if just a very tiny amount), there is then some criterion, against which the evidence could be evaluated.

There is a theoretical possibility to "claim back" some alpha, if futility stopping according to strict objective criteria (that must be strictly adhered to) is foreseen, but this is not very popular due to the need for strict adherence to precisely specified crtieria.

If the analysis is for a primary or secondary (=some type I error control is desired for the endpoint) endpoint with a possibility to reject null hypotheses, then some adjustment will be needed.

Things that fall into other categories than those I have mentioned are rarer and often raise questions about trial integrity (e.g. suggestions to let a large group of people see interim results in order to decide about starting other trials, to publish data, to show data to investors etc.) rather than about statistical type I error control. I mean this in the sense that it is very hard to know exactly what problems are being introduced and how one would adjust for them in analyses.

  • $\begingroup$ (+1) Excellent answer. The alpha spending approach is not the one generally advocated for multiple outcomes in non-sequential trials. That would be a Bonferroni correction, right? Assigning a 0.025 (or hopefully lower) significance level to each of 2 comparisons achieves a 0.05 family wise error rate. It makes one wonder how the two approaches might be hybridized. $\endgroup$ – AdamO Jul 18 '16 at 0:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.