Predictive Performance of a Myelosuppression Model for Dose Individualization; Impact of Inter-Occasion Variability and Level and Type of Information Provided
J.E. Wallin, L. E. Friberg and M. O. Karlsson
Division of Pharmacokinetics and Drug Therapy, BMC, Uppsala University, P.O. Box 591, 751 24 Uppsala, Sweden
Objectives: A previously published myelosuppression model [1] has successfully been applied for several cytotoxic drugs, among them etoposide. Observed plasma drug concentration and/or neutrophil counts from the first cycle and a PKPD model of myelosuppression may be used for dose individualization in order to avoid severe neutropenia or a suboptimal tumor effect. The aim was to evaluate importance of PK and PD information on the predictive performance, and improvement with accumulated information with an increased number of administered courses.
Methods: Multiple course data of etoposide plasma concentrations and neutrophil counts were available for 44 patients from two previously published studies [2,3]. Data was analysed using NONMEM 6. BSV and BOV, including covariances, were estimated. Model performance was evaluated by goodness-of-fit plots and predictive checks.
One thousand patients receiving 5 courses of therapy were simulated from the final model. POSTHOC estimates were obtained providing either only PK data, neutrophil baseline, full course PD or PK+PD profiles from one to four cycles. Predictive performance on the following cycle was evaluated by computing mpe, mae and rmse. POSTHOC estimates were used to find the dose expected to result in a nadir of 1*109 cells/L and what nadir the estimated doses would cause given the true parameters. As a reference, a standard dose reduction of 25% was used when a patient experience severe toxicity.
Results: There was significant BSV in CL (29%), baseline neutrophil count (48%), drug effect slope factor (17%) and mean neutrophil transit time (MTT; 17%). BOV in CL (30%), slope (31%) and MTT (12%) were significant. Predictive performance when based on PD data only was similar to that of full PKPD information, and superior to the use of only PK or baseline data. The difference between PD guided- and standard dosing was less apparent, when comparing mpe and rmse of nadirs. However, patients experiencing severe toxicity was reduced by 8% with PD guided dosing despite that it also allowed for dose escalation (-46 to +23% dose), Adding information from more treatment courses improved the performance.
Conclusions: PK provided little benefit to predictive performance if PD information was available. The ratio of BSV/BOV is of importance in the precision of PD-guided dosing, and the etoposide model contains rather large BOV. Model-based dose individualization was shown to decrease the proportion of patients with severe toxicity despite increasing the overall dose intensity.
References:
[1] Friberg LE et al. J Clin Oncol 20:4713–4721, 2002
[2] Ratain MJ, et al:. Clin Pharmacol Ther 45:226-233, 1989
[3] Ratain MJ, et al: J Clin Oncol 9:1480-1486, 1991
