Perspectives on Non-Inferiority Clinical Trials – based on draft FDA guidance doc DSBS 20 May 2010 H. Lundbeck A/S 7-Apr-15 1 Key contents in FDA draft guidance • Margins – M1 = effect of active control – M2 = ”clinical” margin (fraction of M1) • Analysis methods – Fixed Margin method – Synthesis method H. Lundbeck A/S 7-Apr-15 2 Notation • δCP = effect of active control vs placebo • δTC = effect of test drug vs active control • δTP = effect of test drug vs placebo = δTC + δCP NB: A positive figure indicates ”better” H. Lundbeck A/S 7-Apr-15 3 Base logic of NI test • If δCP = M1, then showing that δTC > -M1 amounts to showing that δTP > 0, since δTP = δTC + δCP • Showing that δTC is greater than some fraction f of M1, δTC > -fM1 (=M2), amounts to showing that δTC + fδCP > 0 or δTC + δCP = δTP > (1-f)δCP , i.e. that a fraction (1-f) of the effect of C has been preserved H. Lundbeck A/S 7-Apr-15 4 Inferences based on 95% CI for δTC T superior to P demonstrated M1(δTP=0) Preserved:p=0% H. Lundbeck A/S Preservation of p % benefit demonstrated M2 0%<p<100% 7-Apr-15 δTC=0 p=100% 5 T superior to C demonstrated Analysis methods • Fixed Margin – Use historical estimate of δCP and its SE to derive a fixed margin for the test (typically lower limit of 2sided 95% CI) • Synthesis – Treat historical estimate of δCP as a random variable – Assume independence between historical data and NI trial and use as analysis basis • δTP = δTC + δCP • V(δTP) = V(δTC)+ V(δCP) H. Lundbeck A/S 7-Apr-15 6 Fixed Margin versus Synthesis criteria Fixed margin method: δˆTC f δˆCP 1.96 SETC f SECP Synthesismethod: δˆ f δˆ CP TC SE f SE 2 TC H. Lundbeck A/S 7-Apr-15 2 2 CP 1.96 7 Fixed Margin versus Synthesis criteria • Fixed margin method inefficient and overly conservative (Rothmann et al., Stats in Med 22: 239-264, 2003) • Synthesis method more efficient and takes the variability of both historical data and NI trial data into account in a natural way H. Lundbeck A/S 7-Apr-15 8 More conservative method FDA position on margins and methods Fixed Margin M1 Fixed Margin M2 Synthesis M1 Synthesis M2 Preferred option H. Lundbeck A/S 7-Apr-15 More conservative margin 9 The case for ”One Standard of Evidence” (PhRMA PISC Expert Team White Paper, BASS XV, 2008) • The traditional standard of evidence for efficacy of a new treatment T is statistically significant evidence that δTP > 0 • Why should an arbitrarily higher standard of evidence (δTP > δ > 0) be used when an active-controlled (AC) trial has been used? • Preservation margin is arbitrary – Preserving less than p% does not imply ineffectiveness of T – In contrast, δTP = 0 has a definite objective clinical meaning • Requiring a higher standard of evidence for AC trials institutes a regulatory bias in favor of the first drug to be approved (requiring preservation of p% may lead to rejection of T even thought T is better than C) H. Lundbeck A/S 7-Apr-15 10 Hypothetical example: 95% CIs relative to P T and C are both superior to P and data suggests that T might be better than C but because C was approved first and T does not meet the p% margin T can’t be approved C vs P T vs P δTP=0 H. Lundbeck A/S p % margin 7-Apr-15 11 Example: Metastatic Bladder Cancer • Randomized trial* of Gemzar + cisplatin compared to MVA + cisplatin • An earlier randomized trial** showed superioty of MVA + cisplatin to cisplatin • In our notation T=Gemzar, C=MVA, P=cisplatin – δTC = log hazard ratio (MVA over Gemzar) – δCP = log hazard ratio (”no treatment” over MVA) – δTP = log hazard ratio (”no treatment” over Gemzar) * Von der Masse et al.: JCO, 18:3068-3077 ** Loehrer et al.: JCO, 10:1066-1073 H. Lundbeck A/S 7-Apr-15 12 Metastatic Bladder Cancer cont’d • Point estimates ˆCP 0.421(variance0.0181) ˆTC 0.039(variance0.0147) ˆTP 0.382(variance0.0328) fromsynthesisanalysis • Synthesis method estimated 90.7% preservation of benefit but lower 95% bound was only 11.7% • So ”preservation of 50% benefit” criterion was not met but Gemzar+cisplatin was statistically superior to cisplatin alone (2-sided pvalue=0.035) H. Lundbeck A/S 7-Apr-15 13 Metastatic Bladder Cancer cont’d • Assuming constancy of δCP across trials – Gemzar improves survival when added to cisplatin (p=0.035) – Gemzar+cisplatin was estimated to have similar efficacy as MVA+cisplatin (estimated HR=0.96) • Why do a test for preservation of effect? H. Lundbeck A/S 7-Apr-15 14 Conclusions • One standard of evidence for efficacy should be used – superiority to placebo – regardless of the design used (placebo- or activecontrolled) • The synthesis method should be used rather than the fixed margin method for better efficiency H. Lundbeck A/S 7-Apr-15 15