Eduard Schmulenson (1), Linda Krolop (1), Sven Simons (1), Susanne Ringsdorf (1), Yon-Dschun Ko (2), Ulrich Jaehde (1)
(1) Institute of Pharmacy, Clinical Pharmacy, University of Bonn, Bonn, Germany, (2) Department of Internal Medicine, Evangelische Kliniken Bonn gGmbH, Johanniter Hospital, Bonn, Germany
Objectives: The inclusion of the patient’s perspective has become increasingly important when reporting adverse events and may assist in management of toxicity during anticancer therapy. Treatment with capecitabine is especially associated with the incidence of the dose- and treatment-limiting hand-foot syndrome (HFS). The relationship between drug exposure and toxicity can be quantified by combining Markov elements with pharmacometric models. The aim of this project was to develop a modeling and simulation framework in order to describe and predict patient-reported HFS severity in patients treated with capecitabine. Based on this example, the suitability of Markov models to simulate the time course of patient-reported toxic symptoms should be assessed.
Methods: Patient-reported HFS grades over time of 150 capecitabine-naïve patients from two open, prospective multi-centered observational studies [1, 2] were analyzed using a minimal continuous-time Markov model (mCTMM) [3]. Grading of HFS severity was based on the Common Terminology Criteria for Adverse Events developed by the National Cancer Institute. Patients were observed up to six cycles and they were asked to grade HFS severity after each conducted cycle. Different covariates were investigated as predictors of variability. To evaluate the model fit, visual predictive checks (VPC) were generated. Model performance was analyzed by calculation of a positive predictive value which indicated the ability of predicting clinically relevant toxicity (≥ grade 2). A negative predictive value was calculated to assess prediction of the absence of relevant toxicity (grade ≤ 1).
Results: The parameters of the developed mCTMM (three intercept parameters describing the cumulative probabilities of suffering from an individual adverse event grade and the mean equilibration time between two adjacent grades) were precisely estimated. The VPC of this dose-toxicity model indicated an accurate description of HFS severity over time. Individual absolute daily dose was found to be a predictor for HFS. The positive predictive value was 36.7% whereas the negative predictive value was 74.5%, indicating the ability of the model to predict the absence of toxicity ≥ grade 2.
Conclusions: This is the first study which evaluated patient-reported adverse event severity in clinical routine with a Markov modeling approach. In general, our study shows that this Markov modeling approach is suitable for predicting adverse event severity during drug therapy. However, it is limited by the adverse event rate and the observation frequency. Since observations of toxicity ≥ grade 2 only comprised of about 22% of the total observations, further toxicity data may be needed to improve the prediction of clinically relevant HFS. In conclusion, the developed modeling and simulation framework may assist in prediction of the time course of patient-reported HFS severity grades. It can potentially serve as a basis for optimizing anticancer and supportive therapy during but also before starting therapy.
References:
[1] Krolop L, et al. BMJ Open 2013; 3: e003139.
[2] Simons S, et al. Support Care Cancer 2011; 19: 1009–1018.
[3] Schindler E, et al. AAPS J 2017; 19: 1424–1435.
Reference: PAGE () Abstr 9247 [www.page-meeting.org/?abstract=9247]
Poster: Drug/Disease Modelling - Oncology