1

I wonder if I can be advised on design for a sequential clinical trial with interim analyses designed to stop the trial early for efficacy or futility ? What factors decide on number and frequency of interim analyses? I can see the first interim timing might depend on getting enough predicted patients (given predictions on accrual and endpoint rates) to meet power for efficacy and conditional power for futility. Or are there other factors ?? But what factors do I take into account when deciding whether to have 1 interim, 2 interims or 5 ? And should the interims be evenly spaced or not ? Many thanks ..

user3156942
  • 360

1 Answers1

2

It's mostly whether the interim analyses (IA) make sense for their intended purpose and logistical considerations (e.g. in a large long term trial over years, it can be very stressful/intense to do one every 3 months and having them closer together than that may be logistically impossible). The planned trial duration, recruitment speed, treatment duration and three time it takes to make IA decisions are key factors in that. This tends to differ a lot between study types and IA purposes (e.g. large cardiovascular outcome RCT with >10,000 patients ruining for years vs. Oncology dose escalation trial with dose escalation IAs after every 3 to 6 patients).

E.g. 25% of planned data is often not a lot for stopping very often when an intervention does not even work (futility), but can occasionally result in futility stopping. But stopping for demonstrated efficacy could occur, if it's possible that the effect size is much larger than expected. On the other hand, if recruitment into the trial is reasonably fast and patients are only followed for a short time (let's say<1 year), then having an IA much later than that may be somewhat pointless (all patients may already be recruited and only have some weeks of treatment to go). This is especially the case, if it takes a longish time to make a decision (e.g. 100s of patients, so weeks of data cleaning, some weeks for data transfer & running reports for lots of data types being looked at, send to a data monitoring committee 2 weeks before meeting etc.) and can be very different in a small single centre early stage study (data cleaning for just a few patients and 1 to 2 key data points is much easier and faster, an external DMC may not be involved etc.).

Spacing IAs too close also adds little, if not much new information becomes available. To some extent, you can just simulate this (e.g. using a convenient normal approximation) to get a feeling for it. For whether a fixed spacing makes sense, statically this of course does not really matter (thanks flexible alpha spending functions!), but if you need to schedule meetings with busy DMC members, then having something like a fixed every x-months schedule can help.

Björn
  • 32,022
  • many thanks for your answer Bjorn, there is loads to think about. FOr some strange reason I had not thought of trial length as an input. I wonder if prior belief in equipoise or lack of should dictate number of interims eg if we strongly suspect equipoise we have several interims particularly powered for futility. If we really believe in the new drug multiple efficacy interims with high alpha. For intermediate cases would we have 1 interim after say 25% of patients and if we don't stop let it run to the end ? – user3156942 Jul 12 '20 at 07:54
  • I suppose number of interims should also be decided alongside the alpha we assign early interims - a big alpha cutoff is equivaleent to 2 interims maybe in some sense ....many thanks again ... – user3156942 Jul 12 '20 at 07:55
  • 2
    To some extent the # of interims is also a function of how many IAs you can realistically do. In a 8 y cardiovascular study, you can fit in several (& it may make sense). In a study that recruits in 18 months & each patient is treated for 1 y, you can do by far fewer. What you want to do at each IA may also depend on factors outside of your influence (e.g. do ethics committees/regulators require safety IAs at a certain frequency? If you stop for demonstrated efficacy with that amount of data, does the drug get regulatory approval [e.g. is there enough safety data]/would physicians use it?). – Björn Jul 12 '20 at 13:10