Optimal design and QT-prolongation detection in oncology studies Sylvain Fouliard & Marylore Chenel Department of clinical PK, Institut de Recherches Internationales Servier, Courbevoie, France PODE Meeting – Berlin - 11th June 2010 Clinical PK 1 CONTEXT (1) • QT prolongation, a biomarker of Torsade de Pointes. • QT measured on ECG, then corrected. • Circadian rythm in QT/QTc data • Usually mandatory QT/QTC study performed in healthy volunteers at supratherapeutic dose • Guidelines: mean QTc effect > 5ms Clinical PK 2 CONTEXT (2) • New anti-cancer drug in clinical development - QTc-prolongation = class effect ? • Development of anticancer drugs: patients only • 2 phase I studies: – PK data available population PK model – No QT data available • 2 ongoing phase I/II studies - QTc-prolongation assessment: ECG measurement times already decided without optimization (=empirical design) • Internal QT database in HV (wo drug) population circadian QTc model available Clinical PK 3 CONTEXT (3) EMPIRICAL DESIGN • 2 phase I clinical trials: n = 60 + 40 (=100) patients • Dose regimen : 14 days on / 7 days off, BID administration (4h apart) • 14 ECG measurements per patient • Same measurement times for all patients ECG Dose Inclusion D1 D2 D4 D14 Treatment ECG times : Inclusion 0 Clinical PK D1 D2 D4 0, 1.5, 0, 1.5h 0, 1.5h 4, 5.5, 8 h D22 No Treatment D14 0, 1.5h D22 0, 1.5h 4 OBJECTIVES 1. Evaluate the Empirical design for ECG Times. 2. Calculate the Power of detection of a QTc effect in the on going phase I/II studies. 3. Optimize the ECG Measurement Times for future studies. Clinical PK 5 MATERIALS & METHODS (1) POPULATION QTc MODEL WITHOUT DRUG Assumption: same model to describe the circadian rhythm in QTc in HV and in patients • Model building dataset: 2 thorough QT/QTc studies - 62 + 87 (=149) healthy volunteers - QT data without drug - Fredericia correction: QTc = QT * HR-0.33 t QTLn QTc t QTM 0 1 QTAn cos n 1 24 / 2 n1, 2,3 • Model characteristics - poly-cosine model [1] - IIV on all parameters - Additive error model • Software (estimation method): - NONMEM VI (FOCEI) QTc (ms) … Median 5% - 95 % CI Observations • Criteria : LRT • Evaluation: GOF, RSE, VPC Clinical PK Time (h) [1] Piotrovsky, V. “Pharmacokinetic-pharmacodynamic modeling in the data analysis and interpretation of drug-induced QT/QTc prolongation” (2005) 6 MATERIALS & METHODS (2) POPULATION QTc MODEL WITH DRUG EFFECT Assumptions: • Same model to describe the circadian rhythm in QTc in HV and in patients • Concentration proportional drug effect on Mesor QTM t QTM0 1 Ct t QTLn QTc t QTM (t ) 1 QTAn cos n 1 24 / 2 n1, 2,3 • QTc-prolongation is measured by : QTc t QTM0 C(t ) • Max QTc-prolongation at Cmax (PKPD model) Clinical PK 7 MATERIALS & METHODS (3) POPULATION PK MODEL • Model building dataset: 2 phase 1 studies - 14 patients, IV multiple doses, oral single dose - 35 patients, oral multiple doses F Ka • Model characteristics: - 3-compartments model - First order absorption and elimination Periph. 1 Q2 (V2) - IIV on Ka, F, CL, V1, V2 Central (V1) Q3 Periph. 2 (V3) CL - Combined error model • Software (estimation method): NONMEM VI (FOCEI) • Criteria : LRT • Evaluation: GOF, RSE, VPC more Clinical PK 8 MATERIALS & METHODS (5) CALCULATION OF FISHER INFORMATION MATRIX Sequential pop PKPD modelling PK model PK parameters QTc model without treatment (not estimated) Mesor, 3 Cosine amplitude terms Drug effect 3 Cosine Lagtime γ (estimated) (estimated) QTc model under treatment 8 parameters + Additive error QTM0, QTA1, QTA2, QTA3, QTL1, QTL2, QTL3, γ Clinical PK 9 MATERIALS & METHODS (6) EVALUATION OF THE EMPIRICAL DESIGN To find the range of relevant γ values corresponding to a range of relevant QTc prolongations QTc t QTM0 C(t ) Range of relevant QTc-prolongation values Range of relevant γ values [1 ms, 100ms] [0.01, 1] • Calculation of the population Fisher Information Matrix – – – Parameters of QTc model without drug γ = {0. 01, 0.02, 0.05, 0.1, 0.15, 0.2, 0.3, 0.4, 0.5, 0.8, 1} IIV on γ = 30 % • Output results: – Clinical PK SE, RSE, DET (determinant of the population FIM) 10 MATERIALS & METHODS (7) POWER DETECTION OF DRUG EFFECT For each value of γ, SE(γ) is computed from FIM Wald test is performed, with a 5 % type I error. SE( ) - Null hypothesis H0 : ~ N ( 0 ,1) no QTc effect of the drug, 0 = 0 - Alternative hypothesis H1: QTc effect of the drug, 0 > 0 Then power is computed from the type II error β. Power = 1- β. 1 95% Clinical PK 0 0 1 1 11 MATERIALS & METHODS (8) ECG DESIGN OPTIMIZATION • Design characteristics : - 1 group of patients - = 0.05, 30 % IIV - Same days* & number of measurement per day* as the empirical design, design domain = [0-10h] for D1 = [0-8h] for each other ECG measurement day 5 ECG on D1, 2 ECG on D2, 2 ECG on D4, 2 ECG on D14, 2 ECG on D22 * • Output results : – Optimal ECG times – SE, RSE, DET (determinant of the population FIM) Clinical PK 12 MATERIALS & METHODS (9) • Software: - PopDes [2], version 3.0 under MATLAB • Design options: -Local, Population, Univariate (design variable = ECG measurement time only, i.e. PK fixed) • Optimisation method: Fedorov Exchange • Criteria : D-Optimality [2] Gueorguieva, K. Ogungbenro, G. Graham, S. Glatt, and L. Aarons. A program for individual and population optimal design for univariate and multivariate response pharmacokinetic and pharmacodynamic models. Comput. Methods Programs Biomed. 86(1): 51-61 (2007) Clinical PK 13 RESULTS (1) EMPIRICAL DESIGN EVALUATION (1) 1 QTM0 QTA1 QTL1 10 QTA2 QTL2 QTA3 1 QTL3 SE (gamma) RSE (%) of fixed effects param. 100 0.1 0.01 0.001 0.1 0.01 0.1 Gamma value 1 0.01 0.1 1 Gamma value Whatever the values (i.e. drug effect), there is low impact on the RSEs of baseline QTc model parameters. SE() increases with ; RSE is below 20 % for > 0.05 (QTc-prolongation of 5 ms). Clinical PK 14 RESULTS (2) EMPIRICAL DESIGN EVALUATION (2) RSE of QTc model parameters for a drug effect () of 0.05 (corresponding to a QTc prolongation of about 5 ms). RSE (%) QTM0 (ms) QTA1 QTA2 QTA3 QTL1 (hr) QTL2 (hr) QTL3 (hr) Add_Err (ms) 0.37 27.1 8.4 10.8 5.9 13.6 2.9 4.85 11.7 The RSEs of QTc model parameters are always lower than 20% for fixed effects, except for QTA1, for which there are around 25%. Clinical PK 15 RESULTS (3) POWER DETECTION OF DRUG EFFECT Power of drug effect detection versus value (drug effect size) Power 1 Série1 0 0.01 0.1 1 10 Gamma Power > 90 % for > 0.02, corresponding to a 2 ms average QTcprolongation. Clinical PK 16 RESULTS (4) ECG TIME OPTIMIZATION (1) RSE comparison for each parameter of the empirical and the optimal designs Empirical design (Det = 2.37 x 1040) QTA3 QTL1 (hr) QTL2 (hr) QTL3 (hr) Add_Err (ms) 0.37 27.1 8.4 10.8 Optimal design (Det = 2.22 x 1064) 5.9 13.6 2.9 4.85 11.7 QTM0 (ms) QTA1 QTA2 RSE (%) RSE (%) QTM0 QTA1 QTA2 QTA3 QTL1 QTL2 QTL3 Add_Err 0.29 8.99 3.35 3.49 1.64 0.37 0.66 5.25 0.26 Sampling times : D1 D2 D4 D14 D22 Phase I/II design 0, 1.5, 4, 5.5, 8h 0, 1.5h 0, 1.5h 0, 1.5h 0, 1.5h Optimized design 4, 8, 8.2, 8.8, 9.6h 1.5, 5.6h 3.8, 5.2h 0, 0.6h 1, 1.5h The optimal design is better than the empirical one, especially for QTA1. Clinical PK 17 CONCLUSIONS INTERESTS & LIMITS This work reassured us on the capability of the empirical design to detect any potential drug effect. The empirical design should allow an accurate estimation of the parameters of the QTc model under treatment. Clinical PK Several assumptions have been made clinicians not ready yet to have an adaptive design within a study. 18 CONCLUSIONS (2) NEXT STEPS • Assumptions made will be challenged with first clinical data coming. – PK model – QTc baseline model parameter values – Linear drug effect • Optimization of the ECG measurement times with different clinical constraints (days, times, number of group, doses, number of measurements) for further studies. • Interest in having an integrated tool for estimation and optimization. Clinical PK 19 ACKNOWLEDGMENT Sylvain Fouliard pharmacometrician at Servier France Mentré Clinical PK 20 BACK-UP Clinical PK 21 RESULTS MODEL BUILDING Population PK model Parameter estimates and RSE of the population PK model Parameter CL KA F V1 V2 V3 Q2 Q3 ErrA ErrP Estimates (RSE %) 54 (10.1) 0.74 (12) 0.30 (10.3) 45 (14.6) 630 (11.7) 61 (11.7) 12 (12.8) 35 (12.8) 0.0092 (32.2) 0.31 (6.36) IIV (RSE %) 0.114 (38.8) 0.342 (32.2) 0.277 (28) 0.202 (35.5) . 0.143 (46.9) . . Back Clinical PK 22 RESULTS MODEL EVALUATION Population PK model Normalized dose Visual predictive checks Normal scale Log scale Time (h) … Clinical PK Median 5% - 95 % CI Back Observations 23 RESULTS MODEL EVALUATION Population PK model Numerical predictive checks Observed Values compared to Simulated Confidence Interval CI Obs below CI (%) Obs in CI (%) Obs above CI (%) MEDIAN 61.1 . 38.8 [P1-P99] 1.7 97.6 0 [P5-P95] 6.1 91.7 2.2 [P10-P90] 10.4 85.3 4.3 [P25-P75] 26.4 60.2 13.4 Back Clinical PK 24 RESULTS MODEL BUILDING Baseline poly-cosine QTc model Parameter estimates and RSE of the baseline poly-cosine QTc model Parameter QTM0 (ms) QTA1 QTL1 (hr) QTA2 QTL2 (hr) QTA3 QTL3 (hr) ErrA (ms) Estimates (RSE %) 400 (0.214) 0.011 (12) 12 (1.98) 0.0103 (7.75) 7.66 (1.04) 0.0073 (3.8) 5.73 (0.61) 5.35 (2.5) IIV (RSE %) 0.0008 (10.7) 0.084 (32.3) 2.4 (24.3) 0.029 (43.2) 0.488 (26.2) 0.047 (40.9) 0.091 (23.7) . Back Clinical PK 25 RESULTS MODEL EVALUATION Baseline poly-cosine QTc model Visual predictive checks QTc (ms) Median … 5% - 95 % CI Observations Time (h) Back Clinical PK 26 RESULTS MODEL EVALUATION Baseline poly-cosine QTc model Numerical predictive checks Observed Values compared to Simulated Confidence Interval CI Obs below CI (%) Obs in CI (%) Obs above CI (%) MEDIAN 49.2 . 50.8 [P1-P99] 0.6 97.6 1.8 [P5-P95] 3.8 90.6 5.6 [P10-P90] 9.4 80.6 10 [P25-P75] 23.3 51.9 24.8 Back Clinical PK 27 CONTEXT (1’) • P wave: auricular depolarisation • QRS complex: ventricular depolarisation • T wave: auricular repolarisation Clinical PK 28 CONTEXT (1’’) • Relationship between QT and RR (=60/HR1000) QT versus RR AT BASELINE ALL DATA QT CORRECTION AT BASELINE ALL DATA QTc vs. RR 475 450 450 425 QTc_BASELINE (msec) QT_BASELINE (msec) QT vs. RR 425 400 375 350 400 375 350 325 325 300 300 500 750 1000 1250 RR (msec) 1500 1750 500 750 1000 1250 RR (msec) 1500 1750 • Compare QT before and after treatment, once QT is corrected for HR (QTc) Clinical PK 29