Development and Evaluation of a Simulation Platform for Malaria Volunteer Infection Study (VIS) Designs
Background: Previous work1 demonstrated viability of adaptive reduced trial designs for VIS in malaria, reducing traditional three single-dose-level, 8 subject cohorts (n = 24) to a three-dose-level single-cohort (n=8) design. The prior proof of concept work used a single cohort, single subset (n=8), favoring the optimal dose (2-2-4). The goal of this work is to select cohort data stochastically to assess performance under uncertainty as to the optimal dose. A simulation-analysis platform was developed for rapid iterative model performance assessment and refinement in support of malaria research.
Methods: An R based platform was developed to rapidly simulate, estimate, and summarize VIS study results for alternate designs. The platform generates analysis datasets and NONMEM control files for estimation with NONMEM 7.3 using KIWI 3. The platform was tested with a literature model for quinidine (3). Ten traditional design studies (8 subjects at each of 3 single dose-level cohorts (n=24)) were simulated from which 100, 2-2-4 (low-middle-high dose) and 4-4-8 single-cohort design trials were selected and evaluated for bias and imprecision of parameter estimates.
Results: Central tendency of PD parameter estimates was comparable between full and reduced study designs. Parameter bias was small and comparable across designs [e.g., Gordi Model, mean transit time (MTT) of parasite life-cycle: -3.0 & -3.3% for reduced vs full-design, respectively]. Precision was rank order for 2-2-4 and 4-4-8 versus traditional 8/8/8 designs (MTT: 2.35, 0.791, & 0.511 %RSD, respectively)
Conclusions: Simulations demonstrated similar accuracy, but reduced precision of PK/PD model parameters at a much reduced trial cost and duration. Further clinical assessment of single cohort design is warranted.
By James Clary, Andrew Castleman, Thaddeus Grasela, Joel Owen