Bioequivalence and Bioavailability Forum

Dear all,

at the last two conferences in Budapest I discussed the issue of retesting subjects with (former) emplo­yees of the FDA.

• The FDA discourages exclusion of outliers (especially of the “no profile” type) in 2×2 crossover studies.
  
• Performing “mini-studies” (aka retesting) was never an official recommendation. The commonly quoted sizes (at least six subjects or 20% of the main study – which ever is larger) are also an urban myth.
  
• The FDA allowed exclusion of outlying subjects only if
  • retesting was already stated in the protocol as some kind of contingency plan,
    
  • any post hoc procedure was not accepted, and
    
  • not more than 1–2 outliers were excluded.

From time to time people send me data desperately wanting me to “safe” their studies. Sorry folks, there is no free lunch. Observations:

• I never (!) saw a protocol where re-testing was planned.
  
• Some studies were quite large (60–80 subjects). When asked why they didn’t opt for reference-scaling the answer was “FDA’s guidance requires a two-way crossover!” It turned out that high vari­ability is well documented in the literature (well, that’s why the sample size was large). The FDA started updating their product specific guidances (allowing RSABE) in 2010. The one for this product was from 2007. Unlike for the EMA (where a clinical justification is needed), there are no problems expected using RSABE for both AUC and C_max.
  
• Even if no scaling is intended, a fully replicated design could rule a product failure and explore a poten­tial subject-by-formulation interaction. Much easier to deal with “no profiles”. That was already stated in the 2001 biostat’s guidance.
  
• Given the advanced technology in bioanalytics accessible today – requiring much smaller sampling volumes than a decade ago – the main obstacle (total blood loss limited to ~500 mL) in many cases is not issue any more. It is true that more periods increase the chance of drop-outs – but the loss in power is smaller than most people believe. Example: T/R 0.95, CV 35%, target power 90%, ex­pected drop-out rate 5% / washout phase.
    2×2 cross-over: Dose 70, complete 67+, power 89.30%+.
    2×2×4 replicate: Dose 36, comlete 30+, power 86.18%+.
  If the drop-out rate is 10%, power in the replicate design would still be >80%…