Publication: Journal of Pharmacokinetics and Pharmacodynamics
Abstract
Population pharmacokinetic modeling relies on computationally intensive parameter optimization algorithms, and faster methods for computing derivatives could meaningfully improve modeling efficiency. This methods paper describes the first implementation of automatic differentiation, termed ‘automatic-differentiation-assisted parametric optimization’ (ADPO), within Phoenix NLME 8.6, applied to several widely used estimation algorithms including FOCE ELS. Benchmarking across four PK/PD models showed ADPO maintains the accuracy and robustness of traditional finite-difference-based methods while reducing total computation time by roughly 20-50%, with a 95% runtime reduction observed in one realistic voriconazole model. This advance offers substantial efficiency gains for population pharmacokinetic modeling workflows without sacrificing accuracy.
Author(s): Chen R, Sale M, Mazur A, Tomashevskiy M, Hu S, Craig J, Dunlavey M, Leary R, Nieforth K.
Published: September 22, 2025
Phoenix NLME: Faster Estimation, Same Rigor
This methods paper benchmarks a new automatic-differentiation approach built into Phoenix NLME, cutting population PK computation time by up to 95% without sacrificing accuracy. As part of a Phoenix platform cited in more than 33,000 peer-reviewed articles, Phoenix NLME remains the engine behind population PK/PD modeling, combining the flexibility of code with the ease of a guided interface.
