OpenPMX Software for Nonlinear Mixed-Effect Models in Pharmacometrics: Precision Compared With NONMEM First-Order Conditional Estimation.

Mixed effects models are a backbone of pharmacometrics, and NONMEM software, with the first-order conditional method with interaction having become the de facto industry standard for model estimation. Documentation exists for the general mathematical methodology for estimation, but many technical and implementation details are lacking. OpenPMX aims to enable nonlinear mixed-effects modeling and estimation in a transparent and efficient manner, with open source licensing allowing for broad application and development. Model parameter estimation bias and root mean squared error (RMSE) obtained using OpenPMX were compared to that using NONMEM for five population models and datasets with varying degrees of complexity. For each model and dataset, repeated simulation and estimation were performed, and the per-dataset difference in precision for parameter estimates was calculated for OpenPMX versus NONMEM. We found that the bias and RMSE of OpenPMX are comparable to the industry standard NONMEM, and in some cases slightly better. The project has low complexity, few dependencies, and is open source, with all technical details open for inspection, auditing, and scientific collaboration.