Rena J. Eudy-Byrne (1), Ahmed Elmokadem (1), Matthew M. Riggs (1), Curtis K. Johnston (1), Jan Marquard (2), Nima Soleymanlou (3), Valerie Nock (2), Karl-Heinz Liesenfeld (2)
(1) Metrum Research Group, Tariffville, CT, USA, (2) Boehringer Ingelheim International GmbH, Ingelheim, Germany (3) Boehringer Ingelheim Canada Ltd./Ltée, Burlington, Canada
Objectives: The objective of this modelling and simulation study was to further characterize the efficacy of empagliflozin (EMPA) 2.5 mg qd dose, independent of data from EASE-3 [1], a phase 3 study which investigated this dose. Specifically, this semi-mechanistic exposure-response modelling study (M-EASE-1) was performed to simulate two scenarios for placebo-corrected HbA1c change from baseline:
1) To assess the effect of insulin dose adjustment on HbA1c
2) To extrapolate the effect on HbA1c lowering in the study population of a 4 week phase 2 trial (EASE-1) [2] by simulating HbA1c lowering to 26 weeks
Methods: M-EASE-1 model development was informed by data from EASE-1 (4 weeks, EMPA 2.5, 10, 25 mg qd) and EASE-2 (52 weeks [1], EMPA 10 and 25 mg qd). Individual predictions of EMPA exposure were taken from a previous population PK analysis. The analysis was conducted in NONMEM Version 7.4 with the FOCEI routine. The exposure-response relationships between longitudinal HbA1c, total daily insulin dose (TDID) and mean daily glucose (MDG) measurements as functions of EMPA exposure at steady state (AUCτ,ss) were parametrically modelled in a step-wise fashion. First, TDID was estimated as a function of EMPA exposure. Next, MDG placebo data and thereafter, the effect of EMPA exposure and TDID on MDG were estimated. In a final step, the time course of HbA1c was estimated using individually derived MDG profiles. For internal and external model evaluation via visual predictive checks, 500 Monte Carlo simulation replicates were generated with parameter uncertainty based on both fixed and random effects. External model qualification, focused on an out of sample prediction using data from EASE-3.
For trial simulations, 500 Monte Carlo simulation trial replicates including inter-individual and residual variability as well as parameter uncertainty were created. The effect of insulin adjustment was based on random sampling from the full data set (EASE-1, -2 and -3 population) with 500 patients per dose group; simulating with and without an EMPA exposure effect on TDID (hypothetical stable insulin). The extrapolated HbA1c time course out to 26 weeks in EASE-1 was based on the study population and the treatment paradigm of this study (19 patients per dose group, 1 week stable insulin, then insulin titration).
Results: TDID was described using a direct response Emax function driven by AUCτ,ss. MDG was affected by three components, 1) EMPA exposure expressed as an Emax function 2) a linear time-dependent placebo effect, and 3) TDID profiles derived from the first model part. Lastly, changes in HbA1c were driven by changes in MDG predicted in the second step. Typical key population parameters (95% CI) were: Baseline HbA1c: 8.15% (8.09%, 8.21%); AUC50 for TDIDEASE-1: 110 (14.3, 836) nmol·h/L; Emax for TDIDEASE-1: 0.186 (0.145, 0.238); AUC50 for MDG: 370 (83.9, 1630) nmol·h/L; and Emax for MDG: 634 (534, 753) mg·day/dL. Inter-individual variability (CV %) for baseline TDID, Emax on TDID, baseline MDG and Emax on MDG were 32.0%, 86.0%, 9.51% and 27.8% respectively. The proportional and additive residual variability estimates (CV% and SD) were 15.6% and 0.0316 for TDID and 16.0% and 0.0316 for MDG, respectively. The simulations performed for external qualification were consistent with EASE-3 results. The simulated median (95% CI) placebo-corrected HbA1c change from baseline at Week 26 for EMPA 2.5 mg qd was -0.29% (-0.40%, -0.10%) and -0.40% (-0.53%, -0.23%) with adjusted and stable insulin therapy, respectively. Simulations of the study population and treatment paradigm of EASE-1 (i.e. 19 patients) showed a median (95% CI) placebo-corrected HbA1c change from baseline at Week 26 of -0.26% (-0.62%, 0.08%) for patients receiving EMPA 2.5 mg qd.
Conclusions: The semi-mechanistic exposure-response model successfully predicted the time-course and dose-related changes of HbA1c for internal (EASE-1 and -2) and external (EASE-3) data.
M-EASE-1 illustrated how pharmacometric analyses can be utilized to simulate untested scenarios (insulin titration, longer treatment duration) to create further evidence of efficacy and substantiate clinical findings.
References:
[1] Rosenstock J, et al. Diabetes Care. 2018;41:2560-2569
[2] Pieber TR, et al. Diabetes Obes Metab. 2015;17:928-35
Reference: PAGE 28 (2019) Abstr 9105 [www.page-meeting.org/?abstract=9105]
Poster: Drug/Disease Modelling - Endocrine