Clinical Trial - sample size for statistically comparable data?

#1
Hi Everyone

In a nutshell - I am looking to undertake an 8 week clinical trial of an equine veterinary health product. I am wanting to analyse corticosteroid concentrations deposited in the keratin of mane hair at Day 1 and then Day 56. I was hoping that 20 horses would be a sufficient number. Of these 20 horses, 10 would have an active treatment, whilst the remaining 10 would have a placebo treatment.


A couple of questions.
1). Is 20 horses a sufficient number?

2). Is the ratio of active to placebo treatments suitable?


I could at a push increase the number of horses involved within the trial to 25 or 30 - if that ensures more comparable data. IF I had an extra 5 or 10 horses, would it be best to put equal numbers on the active and placebo treatments? OR is it best to put these extra horses on the active treatment only and keep just the original 10 horses on the placebo?

I very much hope that makes sense! I would be grateful for any advice :tup:

Jack
 
#2
This very much depends on how big of an effect you are hypothesizing that your treatment will have. If you believe that it will have a fairly large (maybe 1 standard deviation), you can probably get by comparing two groups of ten. IF, on the other hand, you feel that it might be >.3, you will need a much more substantial sample to be able to detect that effect size.

Good luck
 

rogojel

TS Contributor
#3
hi,
why not treat all 20 and use a paired t-test? That would surely have more power then wasting 10 horses on the control. Unless of course you think that the treatment can somehow influence the the keratin content of the mane. Still, you would need to do a proper sample size calculation along the lines the42up sugested.

regatds
 
#4
This very much depends on how big of an effect you are hypothesizing that your treatment will have. If you believe that it will have a fairly large (maybe 1 standard deviation), you can probably get by comparing two groups of ten. IF, on the other hand, you feel that it might be >.3, you will need a much more substantial sample to be able to detect that effect size.

Good luck
Thank you for the quick reply! In all honesty I have no idea how big the effect is likely to be. What I am limited to in this trial is obviously the number of horses participating. With this being said I believe my best bet is to proceed with a paired t-test as suggested by rogojel

Thank you again for your contribution! have a nice day
 
#5
hi,
why not treat all 20 and use a paired t-test? That would surely have more power then wasting 10 horses on the control. Unless of course you think that the treatment can somehow influence the the keratin content of the mane. Still, you would need to do a proper sample size calculation along the lines the42up sugested.

regatds
Morning rogojel...again thank you for your reply, it is greatly appreciated! There is no reason why I couldn't put all 20/25 horses on the active treatment and do away completely with the horses on the control, thank you for suggesting this. I will also look into the sample size calculation!

...I previously mentioned two samples on Day 1 and Day 56. If I decide to include a third sample at Day 28, would I then instead use a one-way ANOVA for equal replicates?

Kind regards
 
#7
paired t-tests are expensive in terms of working with animals. To cover threats to validity, you have to wait to administer the second treatment (at least in work with cattle on a dietary treatment it was about 6 months). This also opens you up to other issues as well.

To be honest, it really depends on your knowledge of the literature surrounding your experiment. If you have the time/money for doing a paired t-test and your limiting factor is the number of subjects you can get, then that is the better statistical method since it will allow for the most power.

If you are constrained by time/money more so by subjects, then I recommend the two sample t-test.
 
#8
Morning rogojel...again thank you for your reply, it is greatly appreciated! There is no reason why I couldn't put all 20/25 horses on the active treatment and do away completely with the horses on the control, thank you for suggesting this. I will also look into the sample size calculation!

...I previously mentioned two samples on Day 1 and Day 56. If I decide to include a third sample at Day 28, would I then instead use a one-way ANOVA for equal replicates?

Kind regards
ah yes, I see, I reread your post.

You are using a placebo and not comparing two treatments. In which case, a paired-t is probably the way to go as rogojel suggested. You are just working for a baseline effect which the horses already should be before they begin treatment.

Though, two samples (control and treatment) are useful in experimental design due the various threats to validity (history, instrumentation, etc) that reviewers/committee chairs love to bring up.
 
#9
ah yes, I see, I reread your post.

You are using a placebo and not comparing two treatments. In which case, a paired-t is probably the way to go as rogojel suggested. You are just working for a baseline effect which the horses already should be before they begin treatment.

Though, two samples (control and treatment) are useful in experimental design due the various threats to validity (history, instrumentation, etc) that reviewers/committee chairs love to bring up.
Just to clarify, I had originally thought it was best to have 10 horses on the active treatment and 10 horses on a control (placebo) treatment. However, it was suggested by rogojel to instead have all 20 horses on the active treatment. If I collected a before and after sample (Day1 and Day 56) then it is best to use a paired t-test. If I decide to collect an additional sample (Day 28) then it is best to use a one-way ANOVA for equal replicates. By doing it this way I would as you mentioned, "be working for a baseline effect".

Thank you again the42up for your help and feedback! :)