A comparison of pharmacometric software programs for atezolizumab population pharmacokinetic simulation

Abstract

Purpose We compare our experience with three pharmacometric modeling workflows for simulating alternative dosing regimens of atezolizumab: (1) the gold-standard, NONMEM software used in combination with R, (2) the R-based package RxODE, and (3) the recently developed Julia-based software Pumas, discussing the advantages and limitations of each. Methods Our prior work demonstrated that an extended-interval dosing regimen (840 mg q6w) following two standard loading doses maintained efficacy while having a nonsignificant exposure-response relationship with adverse events. In the original analysis, the virtual population was generated in R, simulations performed using NONMEM, and data analysis and visualization then conducted in R. In the present study, we perform the full workflow within R using RxODE for simulation and also recreate this workflow using Pumas in Julia. Pharmacokinetic parameters and graphical output, as well as the processing speed for each method were compared. Results All three approaches generated comparable virtual populations, key exposure metrics of CMAX, CMIN, and Weekly AUC, and data visualizations of the simulated serum concentrations. However, there were differences in how quickly each software simulated the entire seven cycle dataset, with Pumas simulating 33,273 obs/second, NONMEM 4,782 obs/sec, and RxODE 251 obs/sec. Due to this large difference, the dataset was broken into individual cycles, where NONMEM and RxODE performed comparably at 2041-3337 obs/sec, while Pumas simulated 48,122-69,168 obs/sec. Conclusion All three software produced comparable results. Ultimately, the choice should be based on the modeler’s specific needs and limitations.

Bibliographic Information