Bioequivalence and Bioavailability Forum

Sample size, ratio, CV ↔ power [Power / Sample Size]

posted by Helmut  – Vienna, Austria, 2012-11-22 14:35 (4496 d 16:54 ago) – Posting: # 9571
Views: 10,565

Hi Torn!

❝ Just as a follow-up, would it be fair to twist a bit what you said and express it this way: "...that if one increases the sample size sooner or later any study will show a significant treatment difference if the ratio is not exactly one." Or is that too much wishful thinking?

For a suitable wording ask ElMaestro.
I once had a study with the minimum sample size of 12 and very low variability (~5%). As expected the study passed in full glory, but with a significant treatment effect. Deficiency letter. Reply similar to the one above. I also said that the result was expected because the sample size for even 90% power would be just four (4!)… The assessor wanted to see bootstrapped studies with a sample size of six as a sensitivity analysis. Guess the outcome.

❝ The reason I'm asking is that the significant treatment effect is what seems to bother the assessor and not the fact that the ratio doesn't include unity (maybe both are linked statistically, but that's beyond my understanding).

Let’s look at my example again:



A significant treatment effect will show up in all studies where the confidence interval does not include unity. Keeping the CV and ratio constant all studies with sample sizes <48 will show no significant treatment effect (CI includes 100%). Any sample size ≥48 will show a significant treatment effect. Different ratios will shift the curves. If the ratio was lower than the 95% in the example, a significant effect will show up earlier (the upper CI intersects 100% at a smaller sample size).

There are two possibilities for a significant treatment effect to show up:

• A very low CV (≤ ~10%). For e.g., an expected ratio of 90% and CV 9% the sample size for ≥80% power is 10. If we perform the study in the regulatory minimum of 12 subjects power will be 91%. Chances are high to see a significant treatment effect – the study is overpowered.
  
• We planed the study for a given ratio and CV. Both are assumptions. If the ratio turns out to be closer to unity and/or the CV is lower than expected, chances are high to see a significant treatment effect – the study is overpowered again.

—
Dif-tor heh smusma 🖖🏼 Довге життя Україна! 
Helmut Schütz


The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes

Complete thread:

• Significance The Outlaw Torn 2012-11-20 07:56 [Power / Sample Size] 
  • Significance ElMaestro 2012-11-20 08:17
    • Significance The Outlaw Torn 2012-11-20 08:55
      • Quality vs. Clinical ElMaestro 2012-11-20 09:04
  • Sample size? Helmut 2012-11-20 20:10
    • Sample size? The Outlaw Torn 2012-11-21 07:52
    • Sample size? The Outlaw Torn 2012-11-22 10:55
      • Sample size, ratio, CV ↔ powerHelmut 2012-11-22 13:35
        • Sample size, ratio, CV ↔ power The Outlaw Torn 2012-11-23 14:47
        • Sample size, ratio, CV ↔ power BEQool 2025-01-31 13:12
          • TOST != treatment effect mittyri 2025-02-20 13:13
          • Sample size, ratio, CV ↔ power zizou 2025-02-21 13:40
            • Not vice versa mittyri 2025-02-22 21:08
              • Not vice versa? zizou 2025-02-24 11:02
                • Not vice versa? mittyri 2025-02-25 06:12
  • Sample size calculation using Marzo and Balant method kumarnaidu 2013-01-31 07:48
    • Marzo / Balant formula d_labes 2013-01-31 10:05
      • Marzo / Balant formula kumarnaidu 2013-01-31 10:54
        • Marzo / Balant formula kumarnaidu 2013-02-04 05:48
          • Marzo / Balant formula for parallel groups d_labes 2013-02-04 08:44