Alan V. Boddy, Michael Cole, Peng Bin, Andrew D.J. Pearson and David R. Newell
Cancer Research Unit and Depts. of Medical Statistics and Child Health, University of Newcastle upon Tyne, Medical School, Framlington Place, Newcastle upon Tyne NE2 4HH
Although carboplatin has a lower incidence of non-haematological toxicity than cisplatin, it is still important to ensure that patients receive an optimal dose. The antitumour and toxic effects of platinum drugs have been related to their plasma concentration and this has led to the concept of a target AUC for carboplatin dosing. A formula based on renal function has been successfully applied to carboplatin dosing in adults and modified version has also been proposed for paediatric patients (RF method). We describe a population method based on one or two samples taken during an initial dose of carboplatin with Bayesian estimation for the optimization of dosing based on adaptive control.
The population data were obtained from an initial 24 patients. Carboplatin was administered as a 60 to 90 minute infusion and ultrafiltrable carboplatin was determined by atomic absorption spectrophotometry in u] to 10 plasma samples. A two compartment model was fitted to each data set using the ML estimator of ADAPT and an error model with standard deviation proportional to estimated concentration. These parameter estimates provided the prior means and covariance matrix for the Bayesian estimator of ADAPT, using a lognormal distribution.
Four sampling times were selected using the Sample program within ADAPT, as the basis for the selection of an optimal sampling strategy with one or two data points. The test data sets consisted of ultrafiltrable carboplatin concentrations in 17 patients. One patient (two courses) had impaired renal function. The two compartment model was fitted to data sets containing one or two points using the MAP estimator and an error model with constant coefficient of variation (derived from the population data error model parameters). Performance was judged by bias (mean % deviation of estimate) and precision (mean % absolute deviation) and was compared to the RF method. Overall, the optimal sampling strategy performed slightly better than the RF method which had bias of 10.2% and precision of 20.8%. With one data point at 60 minutes post infusion the bias and precision were 5.0 and 8.3 % respectively. Addition of a second data point at 30 minutes during the infusion improved the estimate slightly (bias = 3.7 and precision = 4.8%). For the patient with renal impairment the optimal sampling times were 480 minutes for one point (92% of true AUC) or end of infusion and 480 minutes for two points (96% of true AUC). The RF method performed better when RF was impaired, but stable (92 and 98% of true AUC). Thus, all methods consistently underestimated AUC for patients with renal impairment, but by less than 10%.
Whether the limited sampling approach is useful in determining carboplatin pharmacokinetics, thereby allowing adaptive dosing, depends on how accurate the target AUC needs to be, the ease of performing the assay and the desirability of achieving an optimal dose of drug on the first administration. It should be remembered that the determination of renal function requires administration of a radiolabelled compound, collection of several blood samples and analysis and interpretation of plasma concentrations. When both techniques are available, initial dosing based on renal function, with confirmation by optimal sampling that the target AUC has been achieved, may provide the maximum of information. However, both approaches are prone to error due to intrasubject variation in pharmacokinetics.
Reference: PAGE 3 () Abstr 873 [www.page-meeting.org/?abstract=873]
Poster: oral presentation