Differences in Physiologically Based Pharmacokinetic Predictions between Software Platforms Can Be Clinically...

Physiologically based pharmacokinetic models are increasingly applied in drug development and regulatory submissions. Differences in predictions between software platforms may challenge reproducibility and interpretation of results. We compared two widely used platforms, Simcyp and PK-Sim, using levonorgestrel and ethinylestradiol as model compounds, with parameters sourced from the literature, and implemented the simulations without data fitting.

Systematic reconstruction of drug, system, and virtual population models revealed structural and functional differences, including the number of compartments (12 in Simcyp versus 19 in PK-Sim), absorption model options, partition coefficient methods, and enzyme abundances. The clinical relevance of differences was also demonstrated in case of drug-drug interaction (DDI) assessment. Pharmacokinetic (PK) profiles were simulated and area under the curve (AUC) and peak plasma concentration (C_max) ratios computed for scenarios where ethinylestradiol and levonorgestrel were co-administrated with itraconazole and carbamazepine, the well-established inhibitor and inductor of cytochrome P450 (CYP)-mediated metabolism, respectively.

Despite harmonized and consistent inputs, predicted pharmacokinetic metrics diverged and were clinically relevant. For levonorgestrel, Simcyp yielded higher C_max (1.23 versus 0.59 ng/mL, C_max ratio: 2.084) and AUC (10.79 versus 6.75 ng/mL/h, AUC ratio: 1.59), while ethinylestradiol results were more consistent (C_max 0.17 versus 0.13 ng/mL, C_max ratio: 1.30; AUC 1.04 versus 1.15 ng/mL/h, AUC ratio: 0.90). The most substantial differences were obtained with carbamazepine: The C_max ratio was 0.78 with Simcyp and 0.61 with PK-Sim, and the AUC ratio was 0.61 with Simcyp and 0.85 with PK-Sim.

These findings show that reproducing physiologically based pharmacokinetic (PBPK) models across platforms requires more than inputting identical/consistent parameters: Platform-specific defaults and algorithms substantially influence outcomes, in particular in case no parameter is optimized with observed data. Beside the key role of the PBPK expert in the adequate use of the respective platforms, our results highlight the importance of observed data used for parameter adjustment, when needed, and the key role of ensuring model fitting performances on well qualified data. From a regulatory perspective, extrapolating model qualification between platforms should be approached cautiously. Transparent reporting of assumptions, platform-specific sensitivity analyses, and enhanced collaboration between developers, users, and regulators are essential to ensure reproducibility and credibility of PBPK applications in high-impact contexts such as drug-drug interaction assessment.