Bioequivalence and Bioavailability Forum

Hi Laura,

❝ We are planning a XO BE study to support a manufacturing change, and have only data from a single administration (from which we extracted %CV).

That’s the total CV. It is pooled from the between- and within-subject variability. The fact that you administered the product only once, does not mean that the within-subject variability (or inter-occasion variability) does not exist.
Only from a Xover you could estimate (besides CV_total) both CV_intra and CV_inter (see this presentation, slides 12–13).

❝ In order to calculate sample size, do I need information regarding expected correlation between PK parameters in period 1 (test) vs period 2 (ref)?

In a Xover you have two sequences (RT|TR). Hence, T in ½N in period 1 and ½N in period 2. Same for R: ½N in period 1 and ½N in period 2. Administering T and R without randomization (T in period 1 and R in period 2) would be a paired design which is not acceptable in BE (requires lacking period effects).

❝ What would be a reasonable assumption when there is no prior data?

None. You need the within-subject CV which you cannot obtain from the data you have. In most (!) cases CV_intra < CV_inter but the actual relationship is unknown till you have data from a Xover. I know, that’s a vicious circle.
Don’t be tempted to design the Xover based on your CV_total (e.g., your company is “wealthy” enough and the guy in the Armani-suit tells you “We have the budget. Go ahead!”).

• Bad statistical practice and against most guidances.
  
• Unethical. Example polymorphism: CV_total will be extreme though CV_intra might be pretty low. Say CV_total 60% and CV_intra 12%. To obtain 90% power for an expected GMR 0.90 you would falsely design the study with 182 (!) subjects instead of the required 12…
  Hopefully the IEC will have enough statistical expertise to reject the protocol.

Sorry, there are no shortcuts. Either perform a pilot study or – if you don’t have to deal with a HVD(P) – a two-stage design.