Statistical considerations regarding correlated lots in analytical biosimilar equivalence test

Area of Science:

• Biopharmaceutical Sciences
• Statistical Analysis
• Drug Development

Background:

• The Food and Drug Administration (FDA) has recommended an equivalence testing approach for critical quality attributes in biosimilar development since 2014.
• Concerns exist regarding the statistical validity of this approach when reference lot values are potentially correlated, violating independence assumptions.
• Existing methods for analytical similarity evaluation may be compromised by correlated reference lot data.

Purpose of the Study:

• To investigate methods for correcting estimation bias in reference variability when lot values are correlated.
• To evaluate the impact of these correction methods on equivalence margins and confidence intervals.
• To assess the performance of correction methods versus no correction under various correlation structures.

Main Methods:

• Described a method to correct estimation bias of reference variability for correlated lot values.
• Developed modified versions to increase equivalence margins and correct standard errors.
• Compared correction methods with no correction using simulations under known correlation matrices.

Main Results:

• All correction methods significantly increased the type I error rate across simulated scenarios.
• Correction methods offered only slight improvements in statistical power.
• Type I error rates became extremely large when guessed correlation exceeded assumed correlation.

Conclusions:

• Current correction methods for correlated reference lot values in biosimilar analytical similarity testing can dramatically increase type I error rates.
• These methods offer minimal improvement in statistical power.
• Ensuring independent reference lot values, by purchasing lots over a wide time window, is a crucial design remedy for correlated data.