Purpose: The Phase 3 Mylight study was designed to confirm clinical equivalence of proposed biosimilar aflibercept (SOK583A1; Sandoz [proposed biosimilar aflibercept, SDZ-AFL]) to its reference biologic (Eylea; Regeneron Pharmaceuticals, Inc; Bayer AG [reference aflibercept, Ref-AFL]). Method: Mylight was a prospective, double-masked, 2-arm, parallel Phase 3 study. Participants with neovascular age-related macular degeneration were randomized 1:1 to receive eight injections of SDZ-AFL (n = 244) or Ref-AFL (n = 240) over 48 weeks. The primary endpoint was mean change in best-corrected visual acuity score from baseline to Week 8. Secondary endpoints included anatomical outcomes, best-corrected visual acuity at Weeks 24 and 52, safety, and pharmacokinetics. Results: Similarity in mean change in best-corrected visual acuity score was established between SDZ-AFL (n = 235) and Ref-AFL (n = 226) at Week 8 (difference: −0.3 [90% CI, −1.5 to 1.0]) and Week 52. No clinically meaningful differences occurred between groups in anatomical outcomes. Safety profiles were similar, with comparable incidences of treatment-related adverse events (SDZ-AFL: 2.5%; Ref-AFL: 2.9%). The incidence of anti-drug antibodies was similar between groups. Systemic free aflibercept concentrations 24 hours postdose were low and comparable between SDZ-AFL and Ref-AFL. Conclusion: Proposed biosimilar aflibercept matched reference aflibercept in efficacy, safety, and pharmacokinetics in participants with neovascular age-related macular degeneration. Therefore, this Phase 3 study confirmed biosimilarity of SDZ-AFL to Ref-AFL.
Background/Objectives: Biosimilar studies use overall response rate to assess clinical similarity. Sample size and power depend on the equivalence margin, defined in either risk difference or risk ratio scale. This manuscript investigates how different evaluation metrics and varying response rates affect study power. Methods: Two numerical simulations are conducted. The first is designed to test in the risk difference scale, while the second tests in the risk ratio scale. Both simulations consider no difference between the biosimilar and reference product. Response rates vary from 0.1 to 0.9, and all scenarios are repeated 10,000 times. Results: The study shows inconsistent results in testing the equivalence of overall response rate across the risk difference and risk ratio scales, even when the hypotheses are mathematically equivalent. Consequently, the study is often under powered for testing in both scales. Additionally, study power is sensitive to outcome response rate deviation, with different directions of change in the two different evaluation metrics. Conclusions: Biosimilar study design should avoid the concept of converting equivalence margins between risk difference and risk ratio scales, assuming no change in study power. Careful strategies should be planned for estimating overall response rates for sample size assessments.