Virtual twin approach using physiologically based pharmacokinetic modelling in hospitalized patients treated with apixaban or rivaroxaban.

Aims: In a large cohort of hospitalized patients, previously validated physiologically based pharmacokinetic (PBPK)-based models for apixaban and rivaroxaban are being assessed for their performance in predicting individual pharmacokinetics, aiming to identify patients at high risk of under- or overdosing based on demographic, physiological and CYP-related phenotypic characteristics.

Methods: Clinical data were collected from hospitalized patients treated with apixaban (n = 100) or rivaroxaban (n = 100) at the Geneva University Hospitals (HUG). These patients were recruited in the OptimAT trial (NCT03477331). PBPK modelling created virtual twins for each patient, integrating demographic, kidney function, P-glycoprotein (Pgp) and cytochrome P450 (CYP450) 3A phenotyping. Individual PK profiles were simulated for every patient and compared to actual drug exposure, as assessed with LC/MS-MS.

Results: Mean fold error (MFE) (95% CI) for the apixaban and rivaroxaban models integrating demographic and kidney function was within the pre-required bioequivalency criteria with 1.10 (1.04-1.16) and 0.97 (0.93-1.02), respectively. Adding individual Pgp and CYP3A phenotypes led to a slight overprediction 1.25 (1.17-1.33) and 1.30 (1.21-1.39), but patients at risk for bleeding were correctly predicted with MFEs of 0.90 (0.76-1.04) and 1.15 (1.11-1.20).

Conclusions: In a large cohort of hospitalized patients, a PBPK model incorporating demographic characteristics and kidney function can accurately predict, within bioequivalency criteria, an individual's apixaban and rivaroxaban plasma exposure. The added value of individual Pgp and 3A phenotypes on the predictive performance need to be further explored, although patients at higher risk for bleeding may benefit. This innovative approach represents an important step towards the application of PBPK at bedside.