BACKGROUND: Rivaroxaban PK is highly sensitive to hemodynamic variations due to dual renal and hepatic elimination. While heart failure (HF) alters drug disposition, the impact of quantitative ventricular metrics like left ventricular ejection fraction (LVEF) on PK rivaroxaban remains poorly characterized. This study aims to develop and externally validate a population PK/PD model of rivaroxaban with covariates including cardiac and hepatic markers among atrial fibrillation (AF) patients.
METHODS: A prospective study was conducted among 106 adult rivaroxaban-treated AF patients at the National Heart Institute, Malaysia. Three time-matched blood samples (< 1h, 1-3h, and >3-16h) were collected per patient. Rivaroxaban plasma concentrations were quantified via liquid chromatography-tandem mass spectrometry (LC-MS/MS) assay for PK analysis and pharmacodynamic (PD) response was assessed through prothrombin time (PT). Population PK modelling was performed via NONMEM [v7.5, First-Order Conditional Estimation method with Interaction (FOCE-I)] using a one-compartment model (ADVAN2 TRANS2), evaluating absorption lag-time (ALAG) and mean transit time (MTT) structures, and $PRIOR NWPRI subroutine. Inter-individual variability (IIV) was modelled using an exponential block OMEGA matrix. A sequential two-stage approach linked individual PK parameters to drive PT response, evaluating structural PD models (linear, log-linear, Emax, power) via the Bayesian Information Criterion (BIC). A systematic, stepwise covariate search (forward: ∆OFV > 3.84; backward: ∆OFV > 6.64 (for one degree of freedom)) was executed for modelling. Model evaluation was performed using objective function value (OFV), visual predictive check and diagnostic plots. Parameter precisions were evaluated via sampling importance resampling (SIR). The graphs were prepared and produced using R software.
RESULTS: The cohort was divided into a development dataset (n=53; 150 samples) and an external validation dataset (n=53; 142 samples). Rivaroxaban PK was best characterized by an oral one-compartment structural model. The typical population estimates for apparent clearance (CL/F), apparent volume of distribution (V/F), and the absorption rate constant (Ka) were 4.63 L/h (RSE 4.6%), 43.5 L (RSE 7.1%), and 0.536 h-1 (RSE 10.9%), respectively. IIV was captured via a block OMEGA matrix for CL/F (60.2%), V/F (75.7%), and Ka (67.8%), with a 35.1% proportional residual error. Covariate analysis revealed that LVEF significantly predicted absorption, where every 1% increase above 45% increased Ka by 2.165%. Total bilirubin significantly predicted elimination, with each 1 µmol/L increase above 14 µmol/L resulted in a 0.826% reduction in CL/F. For PD, a linear model best described the concentration-PT relationship from a baseline PT of 12.1 seconds, with the concentration-response slope being significantly intensified by concomitant anti-epileptic drugs (3.77-fold increase) and loop diuretics (1.40-fold increase). The final model successfully passed both internal VPC and external validation.
CONCLUSION: This study establishes that quantitative hemodynamic and metabolic markers provide better predictive value for rivaroxaban PK/PD compared to standard clinical baselines. LVEF determines drug absorption, with higher cardiac output accelerates the Ka, while elevated total bilirubin impairs clearance. Furthermore, the concentration-PT relationship is strongly exacerbated by concomitant loop diuretics and anti-epileptic drugs. Precision anticoagulation in AF patients requires a shift toward quantitative assessments to individualize dosing and mitigate bleeding risks.