Comparison and selection of dosing regimens by discretization, optimizing efficacy and evaluating exposure demonstrated for KAE609 (Cipargamin)
Thomas Bouillon (1), P. Timothy Pollak (2)
(1) Academic Affiliation: Department of Anesthesia, University of Berne, CH, (2) Department of Medicine and Cardiac Sciences, University of Calgary, CA
Objectives:
Dosing regimen finding by discretization was formally proposed by D’Argenio in 1994 [1]. Here, we demonstrate an approach to evaluate different dosing regimens by optimizing for target attainment in virtual individuals while comparing the respective cumulative dose for target attainment (CDTA). The practicable dosing regimen with the lowest exposure that meets the efficacy target “wins”. Our test case is KAE609/cipargamin, an antimalarial with high potency (kill rate) and intermediate half-life, for which population PKPD models in healthy volunteers [2] and patients [3] have been published. The CDTA for a single dose regimen is compared to that of several multiple-dose regimens.
Methods:
The proposed approach consists of 8 sequential tasks. 1: Based on the published PKPD models of KAE609 ([2,3], supplements essential), 1000 virtual patients each were drawn from the parameter distributions (no truncation). 2: The efficacy criterion to be met was selected as maximal parasite reduction ratio (PRRmax) of 5*10^11), assumed to be curative in severe adult malaria (10^5 parasites/µL blood, blood volume 5 L). 3: Dataset for estimation of individual doses: each virtual patient received a unit dose of 1 at t=0. The efficacy criterion was set as dependent variable at t=28d (1 datum per patient, 11.7 in log10 units). The individual PKPD parameters except F (bioavailability) were treated as regressors. Different dosing scenarios were introduced via II and ADDL, i.e. each virtual individual had one dosing and one observation record. 4: The typical value and between subject variability of F were estimated with the residual error fixed to 0.001 and individual estimates of F (= individual dose, for which PRRmax equals 11.7 in log10 units) obtained. 5: Target achievement was checked (DV vs. IPRED). 6. Individual CDTAs were calculated (F*(ADDL+1)). 7. The distribution of CDTA ratios were compared between regimens. 8. The results were summarized in language accessible to non-modelers. All operations were done in R using the public version of IQdesktop [4] with Monolix [5] as “estimation engine”.
Results:
Investigated regimens and adjustments: For each parameter set [2,3], single dose (SD), q12h for 3d, q24h for 3d and q24h for 5d regimens were investigated. The EC50 of [3] was adjusted to yield plausible results (most likely unit error, 0.354 mcg/L→0.354mg/L).
Median CDTA [2]: 227mg(SD); 84mg(12h for 3d); 100 mg (24h for 3d); 46 mg (24h for 5d). Median CDTA [3]: 153mg(SD); 80mg(12h for 3d); 85 mg (24h for3d); 60 mg (24h for 5d). Median(IQR) CDTA ratio [2]: 0.39(0.33-0.46) 12h for 3d/SD; 0.45(0.40-0.52) 24h for 3d/SD; 0.22(0.15-0.33) 24h for 5d/SD. Median(IQR) CDTA ratio [3]: 0.53(0.35-0.70) 12h for 3d/SD; 0.56(0.40-0.72) 24h for 3d/SD; 0.46(0.24-0.65) 24h for 5d/SD.
Conclusions:
The approach demonstrates the “exposure cost” of an effective regimen based on within virtual subject comparison followed by aggregation. The median effective doses can be used to assess plausibility of results, which is difficult when comparing ratios only. The distribution of ratios offers high granularity information. In simple terms: Regardless of scenario [2 vs. 3], an effective single dose regimen requires at least double the cumulative dose in approx. 50% of the virtual patients compared to an effective q24h for 3d regimen. Further fractionation (q12h for 3d) does not yield relevant benefits wrt. cumulative dose. q24h for 5d yields diverging results for [2] and [3]. The quality of predictions and resp. recommendations wrt. dosing regimens is, as always, limited by the quality of the available models.
References: [1] D’Argenio DZ, Rodman JH. Controlling the systemic exposure of anticancer drugs: The Dose Regimen Design Problem (p 363-378). In: Pharmacodynamics and Drug Development: Perspectives in Clinical Pharmacology. 1994. ISBN 0-471-95052-1 [2] McCarthy JS et al., AAC 65(2021): p 1-11 (+ suppl.) [3] Hien TT et al., AAC 61(2017): p 1-10 (+ suppl.) [4] IQdesktop. A Qualified Virtual Modeling & Simulation Environment Supporting Efficient Model Informed Drug Development. iqdesktop.intiquan.com. [5] Lixoft/SimulationsPlus
