I am just becoming familiar with the sample size calculation for clinical trials
The questions below are conceptual (not on actual calculations)
1. If an intervention is expected to influence several parameters (ex, blood pressure, sugar levels, weight), are the sample sizes calculated for each parameter separately and the biggest # is the number of patients to be included in the trial?
2. If a trial was conducted without sample size calculation and the results were shown to be statistically significant (vs control), does the fact that sample size was not calculated have bearing of the results? For example, if done prior to the study, calculation would have showed 300 patients were needed, but the study proceeded with 250 and stat significance was still shown.
Thank you!
1. If you want sufficient power (say 80%) to detect effects at a given alpha level of significance (say 5%) for all outcome parameters then calculate sample size seperately for each outcome parameter and take the largest. However, if you have budget or other logistical constaints , select sample size to have sufficient power for the most clinically important outcome parameters.
2.no.
However, if the results were borderline, or for some reason tighter confidence limits were required, or a negative result was important, the study may need to be repeated with a larger sample size.
I agree almost completely with James. However, if you plan to test more than one outcome parameter, you have to face the awkward difficulties of multiple testing and cumulation of alpha error. Therefore most clinical trials are planned with one primary outcome parameter. The other, less important outcome parameters are classified as "secondary" and significance tests on them are often not adjusted for multiple testing. Normally this classic approach spares patients and a significant result for the primary outcome is more trustworthy.
Thank you!
Follow up question: does the sample size calculation assumes a normal distribution of a given parameter?
Keep in mind that you may need different sample sizes depending on your goals. If you are aiming for statistical significance, you may choose one sample size. If you are aiming for robust enough results to, say, impact clinical practice, you may need a much larger sample size. So start from knowledge of what it is you seek to do.
Bear in mind, also, that 80% power means guaranteed 20% chance of failing to find a clinically significant effect when such an effect exists. Ethics committees would much rather 90%, and it's good to run 95% power calculations to see the minimum difference you get.
80% power is too great a risk of wasting research resources.
Hello, Anna!
Sorry if I am wrong, but I think this website might be useful to you:
http://www.sample-size.net/
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size