Exact PK Equivalence for a bridging study Steven Novick, Harry Yang (MedImmune) and Xiang Zhang (NC State) NCB, October 2015 In submission Journal of Biopharmaceutical Statistics • Written with human-subject bridging study in mind. • Easily adapted to animal PK study comparison. Pharmacokinetics: the study of the time course of drug absorption, distribution, metabolism, and excretion. Reference Drug Test Drug Regulatory Requirement: In Vivo Bioequivalence • Change rate and extent of absorption of drug product for which the manufacture has changed. – FDA 1995: Guidance for Industry Immediate Release Solid Oral Dosage Forms, Scale-Up and Postapproval Changes: […] In Vivo Bioequivalence Documentation • A sponsor proposes manufacturing a generic version of an approved off-patent product. – FDA 2006: Guidance for Industry: Bioequivalence Guidance Regulatory Requirement: In Vivo Bioequivalence • […] Demonstrate equivalence […] between the generic medicinal product and a reference medicinal product. – EMEA 2010: Guideline on the Investigation of Bioequivalence Absorption rate=elimination rate Cmax Basic pharmacokinetic (PK) considerations Let the curve be: f( ο±, t ) AUC Tmax T1/2 Bioequivalence definition • EMEA 2010: The plasma drug concentration time curve is […] used to assess bioequivalence between two formulations. • FDA 2006: AUC and CMAX [are considered] as the pivotal parameters for bioequivalence determination. • Area under the curve (AUC): reflects the extent of exposure. • The maximum plasma concentration or peak exposure (CMax) • Statistical evaluations of Tmax and T1/2 are not required. Testing for bioequivalence Show: πππ π΄ππΆ π π /π΄ππΆ π πππ πΆπππ₯ /πΆπππ₯ We add: π π πππ π1/2 /π1/2 π < log 1.25 < log 1.25 < log(1.25) Mice, Rats, and Dogs, oh my • I was often asked to compare PK profiles of mice, rats, and dogs on two different formulations of the same drug compound via AUC, Cmax, Tmax, and T1/2. • Crossover studies • Show statistical equivalence • Sometimes shows statistical superiority Some thoughts • Westlake 1988: Equivalence of the PK metrics does not necessarily imply equivalence in the concentration time profile. • EMEA 2010: The plasma drug concentration time curve is […] used to assess bioequivalence between two formulations. Comparing the mean PK profiles • Chen and Huang (2009) πππ₯ π‘∈(0,π) log( π(π½π , π‘) / π(π½ π , π‘) ) < log(1.25) • We agree! We show: closeness in PK profiles is a more stringent measure than both AUC and CMax. πππ₯ π‘∈(0,π) ==> log( π(π½π , π‘) / π(π½π , π‘) ) < log(πΏ) πππ π΄ππΆ π π /π΄ππΆ π πππ πΆπππ₯ /πΆπππ₯ π < πππ πΏ < πππ πΏ Does not imply anything about Tmax or T1/2. Oral, first-order, single-compartment PK model ππ₯π(−πΎπ π‘ ) − ππ₯π(−πΎπ π‘ ) π(π½, π‘) = πΌπΎπ πΎπ − πΎπ Ka = absorption rate constant Ke = elimination rate constant ο‘ = (Dose)x(Fraction absorbed)/(Volume) A quick counter-example • Reference: πΎπ = 0.331, πΎπ = 0.357, πΌ = 300 • Test: πΎπ = 0.277, πΎπ = 0.301, πΌ = 300 πππ π΄ππΆ π /π΄ππΆ π π πππ πΆπππ₯ / πΆπππ₯ π π πππ π1/2 /π1/2 π = log 1.18 ο = log 1.00 ο = log(1.19) ο Ratio of PK profiles Bioequivalence declared through AUC, CMax, and T1/2 Bioequivalence not declared through PK profile comparison Hypothesis testing • Chen and Huang (2009): parametric bootstrap (probably ok) • We propose Bayesian posterior probability πππ₯ π πΏ, π = ππ( π‘∈(0,π) log( π(π½π , π‘) / π(π½π , π‘) ) < log(πΏ) | πππ‘π ) • If π πΏ, π > 0.95 (say), the PK profiles are declared bioequivalent • JAGS (Plummer, 2003), OpenBUGS (Lunn, Thomas, Best, and Spiegelhalter, 2000) or STAN (Stan, 2014) A small simulation • Examine characteristics of p(ο€, ο΄) vs. traditional bioequivalence testing. • Two arms: Reference (R) and Test (T) – 20 subjects per arm – 11 time points: 0.5 – 16 hours • Hierarchical model Data Generation for Simulation log10 ( πΆπππ‘ ) = log10 {π(π½ππ , π‘)} + ππππ‘ – Between-subject: log e (π½ππ )~π log π π½π , π½ – Within-subject: ο₯ijt ~ N(0, ο³2) 0.01 π½= 0 0 0 0 0.01 0 0 0.01 and ο³=0.022 Priors • Centered at their true values π½π ~ normal with SD = 0.5 – In practice, π½π may be known with greater precision. V-1 ~ Wishart with 20 degrees of freedom – We assumed knowledge of V. ο³ ~ U( 0, 1 ) Mean Reference PK curve Showing time points Example of 20 subjects on Reference drug Testing Traditional Test: πππ π΄ππΆ π π /π΄ππΆ < log 1.25 π < log 1.25 π < log(1.25) πππ πΆπππ₯ /πΆπππ₯ π π πππ π1/2 /π1/2 Proposed Test: πππ₯ π‘∈(0,π) log( π(π½π , π‘) / π(π½π , π‘) ) < log(1.35) Six Scenarios (ο‘=300 for all cases) Ka Ke Max Ratio Reference 0.257 0.294 Test 1 0.257 0.294 log10(1) = 0 Equal Test 2 0.248 0.294 log10(1.1) Equiv Test 3 0.257 0.316 log10(1.1) Equiv Test 4 0.228 0.294 log10(1.35) Borderline Test 5 0.257 0.370 log10(1.35) Borderline Test 6 0.198 0.294 log10(1.88) Not Equiv Bayesian model fitting • 4 independent MCMC chains via JAGS • 200K posterior samples (total) – Burn in = 10,000 – Thinning = 200 • Effective sample sizes > 10,000 • Simulation ran for four weeks! Simulation Results Alternative #1 Close…most of the time (q=70%) 100q% Curves are close together for 100q% time Let π π‘ = log( π(π½π , π‘) / π(π½ π , π‘) ) τ πΌ{ 0 π π‘ < log πΏ } ππ‘ = proportion of t ο (0, ο΄) such that π π‘ < log πΏ Metric: τ π πΏ, π, π = ππ( πΌ{ π π‘ < log πΏ } ππ‘ ≥ π | πππ‘π ) 0 q = 80%, 90%, 95%, 100% Simulation Results Alternative #2 Subset of time points 60% of points < 1 π πΆπππ₯ 10 1 10 π πΆπππ₯ are not equivalent Subset of time points • Let Cο΄ be a subset of time points ο (0, ο΄). • For example Cο΄ = set of time points such that π ππ , π‘ ≥ 1 10 π πΆπππ₯ • A claim of |log( f(ο±R, t) / f(ο±T, t) )| < ο€ for all t ο Cο΄ π π implies | log(πΆπππ₯ /πΆπππ₯ ) | < log(ο€) and implies closeness in the AUC results on the set Cο΄. Alternative #3: Differences • The difference of the two curves are close together; i.e., not on the log scale. πππ₯ π‘∈(0,π) 0.2 π(π½πΉ , π‘ ) − π(π½ π , π‘ ) < π΄ππΆ π π ==> πππ π΄ππΆ π π /π΄ππΆ π πππ πΆπππ₯ /πΆπππ₯ π < log 1.25 < log 1.25 • Posterior probability that log-ratio of two PK curves are close together for all time points in a time range. • Method provides better control over consumer’s risk from a compliance point of view. • Discussed three related, alternative metrics. Thank you! Questions?