Between-Batch Pharmacokinetic Variability Inflates Type I Error Rate in Conventional Bioequivalence Trials: A Randomized Advair Diskus Clinical Trial

Abstract

We previously demonstrated pharmacokinetic differences among manufacturing batches of a US Food and Drug Administration (FDA)-approved dry powder inhalation product (Advair Diskus 100/50) large enough to establish between-batch bio-inequivalence. Here, we provide independent confirmation of pharmacokinetic bio-inequivalence among Advair Diskus 100/50 batches, and quantify residual and between-batch variance component magnitudes. These variance estimates are used to consider the type I error rate of the FDA's current two-way crossover design recommendation. When between-batch pharmacokinetic variability is substantial, the conventional two-way crossover design cannot accomplish the objectives of FDA's statistical bioequivalence test (i.e., cannot accurately estimate the test/reference ratio and associated confidence interval). The two-way crossover, which ignores between-batch pharmacokinetic variability, yields an artificially narrow confidence interval on the product comparison. The unavoidable consequence is type I error rate inflation, to ∼25%, when between-batch pharmacokinetic variability is nonzero. This risk of a false bioequivalence conclusion is substantially higher than asserted by regulators as acceptable consumer risk (5%).

Figure 1

Plasma concentration‐vs.‐time profiles for fluticasone propionate (FP; 100 μg) and salmeterol (50 μg) following single‐dose dry powder oral inhalation to healthy adult subjects as Advair Diskus 100/50 (gray scale) or the test product (red).

Figure 2

Pharmacokinetic comparisons between individual Advair Diskus 100/50 (reference) batches are shown with geometric mean ratios (GMRs) and 90% confidence intervals (CIs) for fluticasone propionate (FP) and salmeterol maximum observed plasma concentration (C_max) and area under the plasma concentration‐vs.‐time curve (AUC). Individual reference batches are indicated as “R1,” “R2,” or “R3.” A ratio value of 1.00 is shown via a horizontal red line. The 0.80–1.25 bioequivalence region is crosshatched. The ratio (y) axis is plotted on a log scale.

Figure 3

Distributions of the test/reference ratio estimate from a two‐way crossover bioequivalence study design in which a single randomly selected test batch is compared to a single randomly selected reference batch in 26 subjects. On the logarithmic scale, the within‐subject residual error variance is assumed to be 0.04. On the natural scale, the true test/reference ratio is assumed to be 1.05. Specific distributions are shown for between‐batch variance values $({\sigma }_b^2$) on the log‐scale of zero (blue), 0.01 (red), 0.03 (green), and 0.06 (gray). The expected range of the 90% confidence interval of the test/reference ratio assuming ${\sigma }_b^2=0$ is shown as a shaded area to illustrate the coverage of a 90% confidence interval derived from a two‐way crossover design. For non‐zero ${\sigma }_b^2$, the two‐way crossover design 90% confidence interval provides only a fraction of the coverage provided for ${\sigma }_b^2=0$.

Figure 4

The type I error rate from a two‐way crossover bioequivalence study design in which a single randomly selected test batch is compared to a single randomly selected reference batch in 26 subjects. On the logarithmic scale, the within‐subject residual error variance ( ${\sigma }_e^2$) is assumed to be 0.04. Log‐scale between‐batch variance ( ${\sigma }_b^2$) is assumed to vary from zero to 0.10 (corresponding to ${\sigma }_b^2/{\sigma }_e^2$ variance ratios ranging from zero to 2.5), with equal between‐batch variance on test and reference products. To assess the type I error rate, the true test/reference ratio is assumed to be 1.25 on the natural scale. Simulation results (green circles) are compared to the approximate analytical solution (blue line).