Bioequivalence and Bioavailability Forum • Bad idea

Bad idea [Two-Stage / GS Designs]

posted by Helmut – Vienna, Austria, 2013-10-19 17:04 (4619 d 11:12 ago) – Posting: # 11710
Views: 29,822

Hi Kumar,

❝ We have already included Potvin's method C in our statistical analysis plan in the protocol. According to our client we should change our design to normal 2x2 crossover design.

The study’s power will drop – in some cases substantially – which may render the study unethical. I hope the study was planned in such a way, that the sample size in stage 1 was large enough to show BE if the assumption about the CV turns out to be correct (that’s the idea behind Method C). If the sponsor played Bingo (“We guess that the CV will be 20%, but maybe we will pass with only 16 – or even 12 – subjects. Let’s try.”) you will be in trouble if you switch to a conventional 2×2 design.
Example (T/R 95%, CV 20%, target power 80%): The sample size in a conventional 2×2 cross-over will be 20 (power 83.5%). If you perform the study in 16 subjects, power will be 73.5% (unethical). With only 12 power will be 56.6% (crazy). Compare this to Method C (10⁶ simulations, exact* power estimation):

Stage 1 sample size 20:
Overall power 86.3%, power in stage 1 75.1%.
Average total sample size 22 (n_5% 20, median 20, n_95% 32, n_max 64).
18.7% of studies proceed to the second stage.
Stage 1 sample size 16:
Overall power 85.2%, power in stage 1 61.9%.
Average total sample size 20 (n_5% 16, median 16, n_95% 36, n_max 80).
34.1% of studies proceed to the second stage.
Stage 1 sample size 12:
Overall power 84.0%, power in stage 1 41.2%.
Average total sample size 21 (n_5% 12, median 18, n_95% 40, n_max 98).
56.5% of studies proceed to the second stage.

In scenario I (stage 1 sample size = fixed design sample size) you have a fairly high chance (75.1%) to show BE already in the first stage; only 18.7% chance to proceed to the second stage. Note that power in the first stage (75.1%) is lower than in a fixed sample design (83.5%) because in some studies you have to use the wider 94.12% CI ⇒ less likely to pass than with the conventional 90% CI.
In scenario II (= gambling) the chance to show BE in the first stage is only 61.9%; chance to proceed to the second stage is 34.1%. Note the distribution of total sample sizes – a consequence of the penalty.
Let’s be silent about scenario III.

❝ Now the problem is we have dosed volunteers and I think CRO has started bioanalyis part (but not statistical analysis). In this situation can we ammend the protocol or we need to repeat the study with ammended protocol.

I would do neither; as shown above power may drop – ethical problems. If the sponsor is wary about EMA’s acceptance of Method C, consider switching to Method B instead.
Can you post your assumed CV and stage 1 sample size?

Slightly different values for n₁ 12 compared to results reported by Potvin et al. since they used in their simulations for speed reasons an approximate power estimation (based on the shifted central t-distribution).

—
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:

Acceptance by FDA/CDER Mahesh 2013-08-06 06:12
- Two stage BE design ElMaestro 2013-08-06 10:05
- Two stage BE design Helmut 2013-08-06 15:31
  - Two stage BE design Mahesh 2013-08-07 06:51
    - Two stage BE design kumarnaidu 2013-10-17 15:08
      - Two stage BE design ElMaestro 2013-10-17 17:28
        
        Two stage BE design kumarnaidu 2013-10-18 05:59
        
        EMA vs. rest of the world Helmut 2013-10-18 14:42
        
        Two stage BE design — cold water thrown on our head Shuanghe 2013-10-18 10:05
        
        Open your mouth and forcefully spit the water out Helmut 2013-10-18 13:29
        
        Change in statistical design kumarnaidu 2013-10-19 07:52
        
        Bad ideaHelmut 2013-10-19 15:04
        
        Bad idea kumarnaidu 2013-10-21 06:01
        
        Change to Method B? Helmut 2013-10-21 14:51
        
        Change to Method B? kumarnaidu 2013-10-22 06:15
        
        Simulations Helmut 2013-10-22 13:20
        
        Open your mouth and forcefully spit the water out Shuanghe 2013-10-22 09:03
        
        Reach for the stars Helmut 2013-10-22 13:05
        
        Alpha inflation in two stage design kumarnaidu 2013-12-04 13:43
        
        Deficiency letters Helmut 2013-12-04 14:50
        
        Deficiency letters kumarnaidu 2013-12-05 06:18
        
        Simulations Helmut 2013-12-05 15:51
        
        Simulations kumarnaidu 2013-12-05 16:07
        
        Simulations Helmut 2013-12-05 16:55
        
        Simulations kumarnaidu 2013-12-06 06:07
        
        Ejaculatio praecox Helmut 2013-12-06 14:30
        
        Ejaculatio praecox kumarnaidu 2013-12-06 15:13
        
        All effects fixed Helmut 2013-12-06 15:31
        
        All effects fixed kumarnaidu 2013-12-09 07:33
        
        EMA… Helmut 2013-12-09 12:45