Bioequivalence and Bioavailability Forum

❝ If I want to conduct a study with the following criteria:

❝ 1. Conduct dosing on all subjects

Good idea.

❝ 2. Conduct bioanalytical on the first 50% of the study population (i.e, subject 1 – x).

Bad idea.

❝ 3. perform PK and BE assessment, and power:

❝    a. If passes BE and demonstrates acceptable power (>80%), conclude study based on 50% data.

❝    b. If fails BE, conduct bioanalytical on remaining 50% subjects and carry out pk and BE assessment with 100% data.

❝    c. If passes BE but fails power (<80%), conduct bioanalytical on remaining 50% of the subjects and carry out PK and BE with 100% data.

You are basically asking for a Potvin method with the modification that no true dimensioning is done, instead you rather gamble that doubling the number of subjects leads to success.

❝ Questions:

❝

❝ 1. If I end up with option 3b or 3c, do I need to adjust for alpha?

Yes, most likely. You may need to simulate the stuff and prove which alpha would protect against Type I error inflation.

❝ 2. Is this design considered the same as a two-stage design (the tradition one)?

Eye of the beholder.... This is a new design so probably only John the Almighty and a few others have considered it until now. If you wish to call it a two-stage design then it is ok with moi.
Some weirdo published a paper that could have marginal relevance to you. Link here.
One issue: If you only allow for doubling the trial size after stage 1, then the trial will in most practical situations have a very, very low power cf the publication reffed above. So low, that it might very well be a waste of time and money, and likely be unethical.

Hi John,

❝ 1. If I end up with option 3b or 3c, do I need to adjust for alpha?

In 3a as well – if you don’t submit in Canada and/or the US only. 3a is similar to Potvin C, which is taboo in some countries.

❝ 2. Is this design considered the same as a two-stage design (the tradition one)?

It is actually more closely related to Pocock’s group sequential approach. But: Pocock’s method is for superiority testing, parallel groups, known (and equal) variances, normal distributed data, a fixed sample size N and one interim analysis after ½N. For this, the α_adj 0.0294 was developed (Potvin’s 0.0294 is a mere coin­ci­dence). Talking BE, that’s not a different league, but a different sport. Therefore, you would have to find a suit­able α_adj and validate the entire grid of possible combinations (sample sizes vs. CV) in order to show that the patient’s risk is maintained at ≤0.05.
Hint: Modify the code in this thread; function power.2stage.GS() in R-package Power2Stage can exactly do what you need.

Now for the big “But”. It’s a tempting idea to safe money in bioanalytics. I have seen similar attempts in the past. In most cases the protocol was not accepted by the IEC. You’ve put N subjects on risk and – even if you have demonstrated BE in the first half – analyzing all samples would give you a more reli­able answer (GMR closer to the “true” value, narrower CI). Saving money is not a striking argument for the IEC. Even if the protocol gets approved, trouble may come: ~20 years ago one of our sponsors per­formed such a study and the regulatory agency – since the validated storage interval was not ex­ceeded yet – required to get the remaining samples analyzed as well.
Overall, I don’t want to walk that road ever again.