Pharmacokinetic Scientist BioPharma Services Inc. Mississauga, Ontario, Canada
Background: Clinical studies are, albeit well-controlled, not immune to missing data, and it is crucial to retain statistical power as a few missing values can disrupt a study. Currently, there is no universal best approach to handle these problematic data in the BA/BE realm.1,2 To study the impact of included or excluded incomplete data on BA/BE results, we simulated different scenarios of missing data. Missing TP(s) around peak time was previously assessed, this poster will focus on missing individual Cmax values. Methods: A master set of 280 subjects with intact data was pooled from the datasets of 5 in-house dasatinib studies. Python was utilized to generate 3 “control groups” mimicking 2-way crossover studies of 50 subjects with the same 16-TP sampling schedule. All control studies showed BE with ISCVs of up to 37%. Simulations with low to high rates of subjects missing Cmax were also created. NCA and ANOVA were performed using WinNonlin® Phoenix 8.3. Scenarios failing FDA-defined BE criteria were considered to critically impact BE results. Concentration-time curves were reviewed to address changes in PK profiles. Results: 33 simulations were generated and presented. When subjects with missing Cmax values were included, the second-highest concentrations were treated as the new Cmax for BE analysis. BE was achieved even at 70% missing rates. When excluding impacted subjects, similar remarks were observed up to 30% missing rate. Beyond this point, BE was not met due to the significantly reduced sample size. (Fig. a) This finding echoes our prior works on missing TPs within the Tmax range, emphasizing that the inequivalence threshold is much lower when excluding impacted subjects than when keeping them.3 However, significant changes in PK profiles in the latter approach can forfeit the study’s conclusion. (Fig. b) Conclusion: Many believe data loss around the absorption peak to be sensitive to study outcome, but our simulations showed that BE remains unaffected unless a substantial amount of data is missing. Although including subjects allows greater tolerance to missing data, it is crucial to strike a fine balance between preserving study power and minimizing impact on PK profiles. Statistical compliance must also involve addressing the significance of any notable change in PK profiles, considering their impact on study’s reliability and practical implications.
