Comparison Between using Continuous and Categorical Toxicity Data for Estimation with a Model for Continuous Data
Angelica L. Quartino (1), Lena E. Friberg (1), Sharon D. Baker (2) and Mats O. Karlsson (1)
(1) Division of Pharmacokinetics and Drug Therapy, Uppsala University, Uppsala, Sweden; (2) The Sidney Kimmel Comprehensive Cancer Center, Johns Hopkins University, Baltimore, Maryland, USA
Objectives: Access to toxicity data is sometimes limited to the categorized severity of the side-effect. The aim of this project is to explore the possibilities to use categorized data to estimate parameters of a model developed for continuous data.
Methods: A semi-mechanistic model of myelosuppression [1] and a dataset not used in the development of the model [2] were used. The dataset comprised of 77 cancer patients treated with docetaxel (75, 50 or 40 mg/m2 according to liver function). Absolute neutrophil count (continuous) and grade 0 to 4 neutropenia (categorical) was recorded on day 0, 7, 14 and 21 during one cycle of treatment. Individual concentration-time profiles were generated using the published population pharmacokinetic model for unbound docetaxel. The drug-related effect in the myelosuppression model was described using a sigmoid Emax model and an additive residual error on Box-Cox scale was applied.
The data was analyzed in four ways; 1) as continuous data, 2) as categorical data where all parameters were estimated, 3) as categorical data using prior information on the population parameters for baseline neutrophil count and residual error and 4) as categorical data using individual baseline values [3] and prior information on the residual error. The model parameters were estimated with NONMEM VI and estimation method Laplace (continuous data) or Laplace LIKE (categorical data). The categorical data was predicted as interval observations using the method for integration (M3) [4] with modification. The models were compared with respect to population parameter estimation and population and individual predictions.Results: The estimated population parameters were similar with the four different ways of analysing the data. The individual predictions of the neutrophil time profiles obtained with continuous data (model 1) and the categorical data (model 4) corresponded very well. Model 2 and 3 were not able to characterize the individual baseline values, thus observations of grad 0 neutropenia (mainly day 0 and 21) were predicted to be around the population mean. However, the models were able to capture the neutrophil profile around nadir accurately.
Conclusions: Categorical data may be used for estimation of population parameters of a model for continuous data with reasonable accuracy. The approach allows individual prediction of the neutrophil time-course even though only categorical data was used in the analysis. It also allows a combined analysis of continuous and categorical data.
References:
[1] Friberg, LE et al. Model of chemotherapy-induced myelosuppression with parameter consistency across drugs. J. Clin. Oncol. 2002, 20:4713-4721
[2] Hooker, A et al. Population Pharmacokinetic Model for Docetaxel in Patients with Varying Degrees of Liver Function: Incorporating Cytochrome P450 3A Activity Measurements. Clin Pharmacol Ther. 2008, Jan 8
[3] Dansirikul, C et al. Comparison of models for baseline. PAGE 15 (2006) Abstr 968 [www.page-meeting.org/?abstract=968]
[4] Beal, S.L. Ways to fit a PK model with some data below the quantification limit. J Pharmacokinet Pharmacodyn, 2001. 28(5): p. 481-504.