Comparison of two Bayesian approaches to dose-individualization for once-daily aminoglycoside regimens
Article first published online: 2 OCT 2003
British Journal of Clinical Pharmacology
Volume 43, Issue 2, pages 125–135, February 1997
How to Cite
Duffull, S. B., Kirkpatrick, C. M. J. and Begg, E. J. (1997), Comparison of two Bayesian approaches to dose-individualization for once-daily aminoglycoside regimens. British Journal of Clinical Pharmacology, 43: 125–135. doi: 10.1046/j.1365-2125.1997.05341.x
- Issue published online: 2 OCT 2003
- Article first published online: 2 OCT 2003
- Cited By
- Markov Chain;
- computerized dose prediction;
Aims Bayesian dose-individualization methods have been shown to have good predictive performance using minimal data points, and are now used widely in clinical practice. This study was designed to compare two computerised Bayesian dose-individualization methods, ABBOTTBASE and SeBA-GEN, in once-daily dosing of aminoglycosides.
Methods ABBOTTBASE uses the maximum a posteriori estimator (MAP) algorithm which analyses all available serum drug concentration data for individual patients simultaneously, while the prior model remains unchanged. SeBA-GEN analyses each data set sequentially while continually modifying the individual patient's prior model, allowing within-patient variability to be modelled. One hundred consecutive patients who received once-daily dosing of aminoglycosides were prospectively dose-individualized using either of these methods. Retrospectively the alternative dosing method was used to provide comparative data. The ability of the methods to predict subsequent serum aminoglycoside concentration data was assessed in terms of their predictive performance, bias and precision.
Results From the 100 patients, 277 serum aminoglycoside concentrations were available. Ninety-eight patients had serum concentrations available from the first dose and 55 from the second dose. Gentamicin was used in 96 patients. There was no significant bias when predicting peak concentrations from the prior model using either SeBA-GEN or ABBOTTBASE. The prior model used by ABBOTTBASE did, however, significantly underpredict the mid-dose concentration (mean bias=−0.79 mg l−1, 95% Confidence Interval [CI]: −1.3 to −0.3). When using the Bayesian algorithms for prediction of the second set of concentrations neither method was biased when predicting the peak concentration. ABBOTTBASE significantly overpredicted the mid-dose concentration (mean bias=0.38 mg l−1, 95% CI: 0.03 to 0.74). The prior model used by SeBA-GEN was more precise at predicting both peak and mid-dose concentrations (P<0.01), indicating better use of covariates. There was no difference between the methods in terms of estimation of the value of volume of distribution, but they differed significantly in the estimation of clearance (mean difference=0.24 l h−1 for SeBA-GEN-ABBOTTBASE, 95% CI: 0.05 to 0.43).
Conclusions Bayesian techniques appear to work well with once-daily aminoglycoside dosing. The method of incorporation of individual patient information into the prior model appears to be important in the optimum choice of the first dose. SeBA-GEN has an advantage in this and in the lack of bias related to predicting low concentrations compared with ABBOTTBASE.