Bioequivalence and Bioavailability Forum

Hi Naghma,

❝ Thanks for a comprehensive reply,

You are welcome.

❝ We are focusing WHO regulations …

Did you bother reading my entire post? There are nothing like WHO “regulations”. It might well be that you pick something out of the WHO’s GLs and at the end of the day the agency of your country will not accept it.

❝ … and as per WHO there is another approach for highly variable drug (HVD) i.e. sequential design approach

❝ Some regulatory agencies (Canada, Japan, and the World Health Organization (WHO)) permit ‘add-on’ designs.

Not precisely. Let’s see what’s written in your reference (of 2010) about add-on designs:

If the bioequivalence study was performed with the appropriate number of subjects but bioequivalence cannot be demonstrated because of a larger than expected random variation or a relative difference, an add-on subject study can be performed using not less than half the number of subjects in the initial study, provided this eventuality was anticipated and provided for in the study protocol. […] Add-on designs must be carried out strictly according to the study protocol and SOPs, and must be given appropriate statistical treatment.

Which translates into:

• The CV_wR of C_max given in the SOF-GL is based on numerous studies and hence, can be considered reliable. An “appropriate number of subjects” means the sample size which as expected will result in sufficient power (i.e., ≥80%). As stated in my previous post a 2×2 crossover (ABE) would require 230 subjects.
  
• “Appropriate statistical treatment” is the crucial point. You must not (‼) evaluate the study with a nominal α of 0.05 (i.e., the conventional 90% CI) and afterwards add more subjects and pool the data. The consumer’s risk will not be preserved (= inflation of the Type I Error of more than 50%). This was shown for the regulations of Japan^1,2 and for the ones of Argentina, Korea, and Mexico.² Canada² had a similar approach but it is not acceptable any more. In these add-on designs (second group ½, ¾, or equal to the first group) Bonferroni’s 0.025 (95% CI) would preserve the consumer risk. Note that the sizes of the first and second groups are fixed and have to be given already in the protocol. Since the evaluation will be done based on 95% CIs it would increase the sample size of the first group (n₁) to 292 subjects. Say you stated that the second group will be ½n₁. Even if you miss BE by a very small margin, you would have to perform the second part full throttle in yet another 146.
  That’s something nobody likes. Better approaches are a “classical” group sequential design (fixed total sample size and one interim analysis) or adaptive two-stage sequential designs (fixed stage 1 group size with sample size re-estimation in the interim).

Now for the bad news: All this stuff so far has only be explored for 2×2 crossovers and parallel designs for conventional ABE (no replicate designs for RSABE or ABEL; see this thread). In your case the sample sizes would be a nightmare.
Note that add-on designs were dropped in the WHO’s working draft (2014) and two-stage designs suggested instead. Similar in the current (2015) version.* Although a paper was published³ for an expected T/R-ratio of 0.90 and high variability, it does not cover such extreme sample sizes. If one wants to walk that stony path, simulations have to be performed to demonstrate that the chosen adjusted α preserves the patient’s risk.

❝ my query: Which one will be the most appropriate approach for BE of Sofosbuvir?

❝

❝ 1. Sequential Design Approach (initially conducted on 24 subjects and if bioequivalence cannot be demonstrated than an add-on subject will be done as per guideline)

Forget it. See above. BTW, 24 subjects would not be “appropriate” anyway. n₁ 292 in the obsolete add-on (α 0.025) and ~120 in the two stage design (α 0.028). If you start a TSD in a too small sample size, you will be penalized (high expected average total sample size E[N]). Likely you will have to proceed to the second stage since the power to show BE already in the first stage is low:

 n1  E[N]  in stage 2  power1
 24  282     99.6%      0.4%
 48  262     89.2%     10.8%
 72  228     72.8%     27.2%
 96  215     62.7%     37.3%
120  211     54.8%     45.2%
144  212     47.9%     52.1%
168  217     41.6%     58.4%
192  226     35.9%     64.1%
216  238     30.6%     69.3%
240  252     25.3%     73.7%

❝ OR

❝

❝ 2. Reference Scaled Average Bioequivalence (RSABE), a full replicate 2 × 4 cross-over design Which can be conducted on minimum of 22 subjects?

How shall I know? Although I’m a gifted tassologist and fortune teller, my crystal ball is in the laundry. Otherwise it easily could whisper in my ear what your regulators will accept.
You should ask them, not me.

Scientifically only RSABE (n = 22) or ABEL (n = 28) makes sense.

---

References:

1. Wonnemann M, Frömke C, Koch A. Inflation of the Type I Error: Investigations on Regulatory Recommendations for Bioequivalence of Highly Variable Drugs. Pharm Res. 2015;32(1):135–43. doi:10.1007/s11095-014-1450-z
   
2. Schütz H. Two-stage designs in bioequivalence trials. Eur J Clin Pharmacol. 2015;71(3):271-81. doi:10.1007/s00228-015-1806-2
   
3. Montague TH, Potvin D, DiLiberti CE, Hauck WW, Parr AF, Schuirmann DJ. Additional results for ‘Sequential design approaches for bioequivalence studies with crossover designs’. Pharm Stat. 2012;11(1):8–13. doi:10.1002/pst.483

---

• Don’t estimate the sample size of the second stage based on the point estimate of the first stage as stated in the GL. That’s nonsense. Use only the CV. Why? See Ref. 2.