Enoxaparin is a low molecular weight heparin used in the treatment of thrombosis. The current approved treatment dose of enoxaparin is based on total body weight and its dosing frequency is based dichotomously on creatinine clearance. Recent evidence has shown these dosing strategies to be suboptimal and Bayesian dose-individualisation has been proposed as a safer and a more effective alternative. A population pharmacokinetic model for enoxaparin has been developed (Green, 2003; Green, 2005) and coded into a Bayesian dose-individualisation software (TCIWorks). The predictive performance of this dosing method has not yet been evaluated.
To evaluate a computerised Bayesian dose-individualisation method for enoxaparin.
Demographic data, dosing history, and anti-Xa measurements (a surrogate for enoxaparin plasma concentration) of 109 patients who received enoxaparin treatment (Barras, 2008) were entered into TCIWorks. The mean error (ME) and root of mean squared error (RMSE) for the prior predictions (calculated from patient covariates) and posterior predictions (estimated from the posterior parameter estimates) to the future observed anti-Xa observations were calculated to determine the bias and precision of model predictions.
There were a total of 238 anti-Xa measurements in the dataset: 109 first observations (mean = 4.1 mg/L), 98 second observations (mean = 8.6 mg/L), 26 third observations (mean = 6.9 mg/L), and 5 fourth observations (mean = 8 mg/L). The RMSEs of the posterior predictions were 3.3, 1.7, and 1.8 mg/L for the second, third, and fourth observations, respectively. The RMSEs for the prior predictions were 2.5, 4.2, 2.8, and 2.8 mg/L for the first, second, third, and fourth observations, respectively.
Both prior and posterior predictions were negatively biased but the bias of the posterior predictions decreased with more observations and became non-significant after the third observation (95% CI -2.3 to 1.1). The MEs of the posterior predictions were -2.2, -0.6, and -0.6 mg/L for the second, third, and fourth observations, respectively. The MEs for the prior predictions were -1.5, -2.9, -1.6, and -2.3 for the first, second, third, and fourth observations, respectively.
The Bayesian estimation performed relatively well given the low bias especially after the second and third observations. Imprecision was also reduced after the second observation. The prior model suffered from significant bias and imprecision.
TCIWorks provided acceptably accurate predictions of anti-Xa activity. There appears to be limited benefit in obtaining more than two observations during dose-individualisation.
Green & Duffull. Br J Clin Pharmacol 2003 Jul;56 (1): 96-103.
Green et al. Br J Clin Pharmacol 2005 Mar;59 (3): 281-90.
Barras et al. CPT 2008 ; 83: 882-8