True, but find a statistician who works in that clinical area, and it's a bonus all round. A decent clinical trial benefits from both clinical and statistical expertise.

From a simplistic viewpoint, for Deepak's consideration, you need to have a pretty good idea of what the control/standard rate of events is (ie. without intervention, using standard care -> this is effectively your control arm). Once you have a good idea of the expected event rate in the control arm, you then consider what plausible difference should be expected once the intervention is admininstered, keeping in mind are you talking about an absolute difference or a relative difference? What is the smallest plausible difference that would be clinically worthwhile and would it be possible with the intervention?

Example, control/standard care has an event rate of 15%. Prior research (phase I and II studies for example) might suggest that the intervention could reduce the event rate to 10% (absolute difference of 5%, relative reduction of 33%). That being the case, you might look at a few sample size calculations looking at intervention rates at 9%, 10&, 11% and 12% and then also consider the impact of withdrawals/non-compliance rates ranging from 1% to 15%. Assume that your intervention will not work as brilliantly as you expect (they rarely do) and go for the smallest difference possible - although keep in mind this will also require a larger sample size.

But initially, Deepak also needs to figure out what the primary hypothesis is, what is the outcome (proportion, event time, continuous measurement, etc) and what analysis would be required to address/answer the primary hypothesis. Once this is decided (in consultion with other clinical colleagues and a statistician), then they can look at various sample size scenarios and pick the one which will answer an important clinical question with an efficient and logistically achievable sample size (taking into account how long it will take to set up sites, recruit patients, non-compliance, etc).

Having worked with a number of trial statisticians, some of them have very good knowledge in their clinical areas of research and can provide insightful feedback for the design of the trial, not just a sample size number.

If you dont know or even cant estimate the expected "effect" you cant calculate the sample size...If you not head for a pilot study, you should have evidence for the (assumed) comparibilty with the standard drug. So you should also think about the possible cohort (prevalence/ incidence of disease to treat with your drug) and think about the desirable effect - Then you can calculate a "first guess" of your number.

thank you Fabin ............ how can i find desirable effect?. how can i calculate sample size using prevalence....

I just wanted to remember you, that the prevalence or incidence may (naturally) limit your possible sample size. If you are not able to recruit i.e. 100 patients to show an possible effect of your drug, your power is limited. The "desirable effect" is the one, which may be an improve compared to gold standard. Example: NNT=100 (standard) --> NNT=90 (new) ; improvement 10%. The efficancy might be higher or even lower - so take a guess what would be possible for your trial, because ressources and maybe patients to treat are limited.

Helpful might be: http://www.openepi.com/v37/SampleSize/SSMean.htm

I would seriously suggest you get a statistician involved who has clinical trial experience. They can help with sample size and design issues and a good trial statistician should be able to give you some input with regard to adherence to GCP guidelines.

Dear Peta, statisticians might be able to calculate quite correct sample sizes, but they can not answer the questions of clinical relevance, expected "side"-effects of a drug a.s.o. (which are necessary to calculate correct sample sizes) ... These are the facts you have to deliver to the statistician.

True, but find a statistician who works in that clinical area, and it's a bonus all round. A decent clinical trial benefits from both clinical and statistical expertise.

From a simplistic viewpoint, for Deepak's consideration, you need to have a pretty good idea of what the control/standard rate of events is (ie. without intervention, using standard care -> this is effectively your control arm). Once you have a good idea of the expected event rate in the control arm, you then consider what plausible difference should be expected once the intervention is admininstered, keeping in mind are you talking about an absolute difference or a relative difference? What is the smallest plausible difference that would be clinically worthwhile and would it be possible with the intervention?

Example, control/standard care has an event rate of 15%. Prior research (phase I and II studies for example) might suggest that the intervention could reduce the event rate to 10% (absolute difference of 5%, relative reduction of 33%). That being the case, you might look at a few sample size calculations looking at intervention rates at 9%, 10&, 11% and 12% and then also consider the impact of withdrawals/non-compliance rates ranging from 1% to 15%. Assume that your intervention will not work as brilliantly as you expect (they rarely do) and go for the smallest difference possible - although keep in mind this will also require a larger sample size.

But initially, Deepak also needs to figure out what the primary hypothesis is, what is the outcome (proportion, event time, continuous measurement, etc) and what analysis would be required to address/answer the primary hypothesis. Once this is decided (in consultion with other clinical colleagues and a statistician), then they can look at various sample size scenarios and pick the one which will answer an important clinical question with an efficient and logistically achievable sample size (taking into account how long it will take to set up sites, recruit patients, non-compliance, etc).

Having worked with a number of trial statisticians, some of them have very good knowledge in their clinical areas of research and can provide insightful feedback for the design of the trial, not just a sample size number.

I am sure the following site will do the trick for you as long as you know the effectiveness of the control therapy. Then you add the expected difference between treatments (normally you assume they are the same or the experimental substance is slightly worse when it comes to non-inferiority studies). The significance level (alpha) is usually 2.5% (equals 5% in two-tailed analysis) and the power (1-beta) is usually 80%. Last but not least, you need to define the non-inferiority limit (d), higher than which a difference between the two treatments in terms of effectiveness is not accepted and the tested drugs is considered inferior. The rest is calculated by the program. It worked for me...

With the above parameters, the [sample size] found means that: "If there is truly no difference between the standard and experimental treatment, then [sample size] patients are required to be 80% sure that the upper limit of a one-sided 97.5% confidence interval (or equivalently a 95% two-sided confidence interval) will exclude a difference in favour of the standard group of more than 10%"

Here is the link:

http://www.sealedenvelope.com/power/?calc=3

@Dimitrios, I think the OP wants a superiority trial, not a non-inferiority trial, as you last paragraph suggests.

Anyway, because of the presence of the innocent word "multicentre" in the original question I strongly agree with Peta and also recommend the consultation of an experienced statistician.

The link I gave allows the calculation for all types of trials. However, I doubt he wants a superiority trial. Actually, I would be suprised, if he did. To my knowledge the only trial that proved superiority of the experimental drug was the one conducted decades ago and examined the role of streptomycin in tuberculosis.

If someone has very little idea of the effect size they expect to see, I would not recommend handing them a sample size calculator/website to work it out themselves. They need more guidance than that.

@Dimitrios, maybe we have a different understanding for "experimental drug" but in my field most RCTs are in fact superiority trials. For instance you normally need to show efficacy of a new drug (against standard care) to get FDA approval, it doesn't suffice to show that it is not worse than standard.

@Eik, I would be happy if you could share some of them with us. It would be very useful. In ID, non-inferiority studies are the rule. And statistically safer, if I may add.

Thanks for all valuable suggestions and my study is non- inferiority study