(M1530-10-69) A Simulation-Based Decision Guide to Evaluate Missing Data in Bioequivalence Studies

Purpose: In bioequivalence studies with pharmacokinetic endpoints, selecting an adequate and robust sampling time is essential for reliable estimations of PK parameters and accurate BE assessment. Though the derived endpoints are approximations of drug characteristics and data points are inevitably missing in clinical trials, insufficient data undermine the precise determination of PK parameters. Although the impact of missing data on modeling approaches, analysis methods, and how to handle missing data in pharmacokinetic (PK) modeling has been extensively studied, less known about the extent and consequences of incomplete PK data on BE assessment. The most common approach to handle missing data that affects PK parameter estimation in BE studies is to exclude the impacted parameter(s) from the BE assessment. However, some argue that in certain cases, including the affected data might yield advantages for the BE assessment. The criteria used to determine the severity of the impact on PK parameters is also controversial. Therefore, we conducted simulations with various extents of randomly incomplete sampling and investigated their effects on the BE results when impacted parameter(s) were included or excluded. The simulations cover multiple scenarios; however, this poster focuses on the measurement around the maximum concentration [C_max (T_max)] which is a critical parameter exhibiting significant variability in BE assessment.

Methods: In this simulation, we used data from the dasatinib tablet – a highly variability drug with the reported intra-subject variability ranging from 30% up to 70%.¹ The sampling schedule was determined based on regulatory recommendations and dasatinib PK profile¹_, aiming to sufficiently cover its T_max and half-lives, at 0, 0.25, 0.5, 0.75, 1, 1.25, 1.75, 2, 2.5, 3, 4, 6, 8, 12, 16, and 24 hours post-dose. Five in-house studies (N = 280) were selected, which contained replicate dosing of the reference listed product (RLD) under fasting conditions and provided concentration data at the defined sampling schedule mentioned above, to ensure consistent data for comparison. Python programing on Google Colab was utilized to randomly pool data from two replicated-reference periods of 50 subjects, mimicking a 2-period, 2-treatment crossover study (Treatment A and B). Treatment A demonstrated bioequivalence to Treatment B, exhibiting a similar concentration-time profile with an estimated ISCV of up to 36%. Incomplete profiles were simulated using Python programming (NumPy and pandas), generating missing rates of 10% to 60%, and missing time point(s) per subject varying from 1 to 7 within the reported T_max range (0.5 to 3 hours). The non-compartmental analysis to estimate PK parameters and BE assessment was performed using PROC GLM in WinNonlin® Phoenix 8.3. For each simulation, two BE assessments were conducted, one including and one excluding BE dataset with missing data. The acceptance range of 80.00 - 125.00% was applied to the geometric mean ratios and their 90% confidence interval (CI) for AUC_t, AUC_inf, and C_max, which are commonly used to evaluate BE. Simulations failing the criteria were considered to have a critical impact on the BE result. Additionally, mean concentration-time profiles of all simulations were compared to the original dataset to evaluate changes in the shape of the PK profile.

Results: Fourteen simulations and representative scenarios were generated (Figure 1). When subjects with missing data were included in the BE dataset, the outcomes remained unaltered until 40% of subjects had at least 7 missing time points within the same period. Exceeding a missing rate beyond 40% also did not impact the BE results, unless the extent of missing value was substantial ( > 5 time points) which it is not commonly seen in a quality-controlled trials. This suggests that the impact of missing data on the two-sided 90% CI could be mitigated with a robust sampling schedule around T_max. However, incorporating subjects with missing data had minimal effect on the BE conclusion unless the rate and extent of data removal were large, the actual PK parameter values were skewed, and PK profile shape were changed (Figure 3). Conversely, excluding subjects with missing data from the BE dataset did not affect BE conclusion at a 20% missing rate. Nevertheless, excluding subjects with missing data at or above a 40% missing rate led to unmet BE, partly attributed to the low sample size of the excluded population.

Conclusion: While excluding potential unreliable PK parameters from BE analysis may lead to an insufficient samples size for BE assessment, including subjects with missed critical points is unlikely to impact the BE criteria unless the extent of missing data is substantial (40% missing rate with 7 time points around T_max). However, including subjects with missing data introduces bias in the PK parameter estimation and compromises study reliability. Therefore, this simulation supports the approach to exclude the impacted PK parameter(s) from BE assessment and emphasizes the significance of incorporating a sufficient buffer when estimating the sample size of a study. Please look forward to our future publication for more discussions and simulations of other scenarios.

References: 1. FDA Clinical Pharmacology and Biopharmaceutics Review(s). Sprycel® film coated tablets. Published online 2005: 1-132.
2. Bristol Myers Squibb Pharmaceuticals Limited. Summary of Product Characteristics – Sprycel® 140 mg film coated tablets. Published online Oct 19, 2022.

Figure 1: Illustrated flowchart of the simulation procedure and representative scenarios of missing data.

Figure 2: Forest Plot - Comparative BE results (Cmax and AUCt) when including and excluding the impacted subjects with missing data.

Figure 3: Visualized plasma concentration-time profile and descriptive statistics of PK parameters in complete and incomplete data with 40% missing rate and varying degrees of missing time point(s).