Statistical Tools for Biosimilarity Assessment: A Perspective Based on WHO-recommended Approaches.

Abstract

Biosimilars expand access to biologic therapies by offering high quality, cost-effective alternatives to originator products. Analytical similarity assessment is central to biosimilar development, yet biosimilarity outcomes are strongly influenced by reference product (RP) lot numbers, assay variability and RP variability, and statistical methodology. The World Health Organization (WHO) recommends statistical tools for analytical biosimilarity evaluation, including X-sigma, min-max, and tolerance interval test, however their sensitivity to variability has not been systematically examined. In this perspective, we examine how reference lot selection, assay replication, and RP heterogeneity affect biosimilarity conclusions when applying WHO-recommended methods. Using analytical data across glycosylation, size heterogeneity, and charge variant, four biosimilars were compared with an innovator RP under different variability scenarios. Similarity outcomes were quantified using a normalized scoring framework. Increasing RP batches and incorporating broader intrinsic RP variability increased apparent biosimilarity by yielding acceptance limits that better reflected reference dataset dispersion, whereas limited sampling produced unstable variance estimates and restrictive criteria. Increased assay replication provided modest, attribute-dependent changes in biosimilarity outcomes. Across scenarios, X-sigma approaches were more robust to limited sampling and changes in variance estimates, whereas min-max and tolerance interval methods were more sensitive to dataset expansion and distributional extremes. These findings support data-driven justification of reference characterization and statistical strategy in biosimilarity assessment.