Aymara Sancho-Araiz (1,2), Belén P. Solans (1,2), VÃctor Mangas-Sanjuán (3,4), Iñaki F Troconiz (1,2).
(1) Pharmacometrics & Systems Pharmacology Group, Department of Pharmaceutical Technology and Chemistry, School of Pharmacy and Nutrition, University of Navarra, Pamplona, Spain. (2) IdiSNA, Navarra Institute for Health Research, Pamplona, Spain. (3) Department of Pharmacy and Pharmaceutical Technology and Parasitology, Faculty of Pharmacy, University of Valencia, Valencia, Spain. (4) Interuniversity Institute of Recognition Research Molecular and Technological Development, Valencia, Spain.
Objectives: Time-to-event (TTE) models allow the characterization of the time-course of a specific response in a population through a hazard function. The time of an event is not observed in all subjects because either the subject is not followed up long enough or drops out from the clinical trial. These two different types of missing data are usually classified as censored information [1]. However, this approach neglects the potential correlation between treatment effect and patients’ drop out. For a TTE model to accurately replicate the outcome of a clinical trial, the correlation of the response with both the event of interest and the drop out might be relevant [2]. The objective of this work is to compare the performance of different TTE model parametrizations on the characterization of death and dropout events with treatment response throughout a clinical trial.
Methods: A total of 140 patients were included in the analysis, of which 25 patients died, 35 patients left the study due to unreported events, and were therefore considered dropouts, and 80 completed the study and were considered survivors. For each type of event, two different parametric TTE analysis, differing on how censored information was handled, were performed. Approach A considers non-censored information individuals that suffered one of the events, thus ignoring the other event, and as censored information patients that survived until the end of the study. The second approach (B) takes into account all the patients. Equally to approach A, censored information includes those individuals that suffered the event of interest. However, the non-censored information includes both patients considered as survivors and those that suffered the other event. Two more parametric TTE analysis were developed where death, dropouts, and survivor information was simultaneously fitted following the two different approaches.
Results: Overall survival (OS) data was best described by a Weibull distribution with an alpha parameter (α) and a base parameter (λ). For model A1, λ was estimated to be 0.00864 and α as 2.32; for model A2, λ and α were 0.0098 and 1.81 respectively, parameter estimates for the combined model A3 were the same than the ones estimated in the separate models. On the other side, model parameter estimates for model B1 were 0.00821 for λ and 2.43 for α, whereas in the case of model B2, λ was 0.00816 and α was 1.74. In the same way than in the A approaches, the estimates of parameters in the combination of models B (B3), were the same as when both events were analysed separately. The two different approximations are not exchangeable. The approximation of model B allows the separation of the influence of death or dropout on the opposite hazard.
Conclusions: This methodologic exercise compares different TTE modelling approaches and their impact on the parameter estimation. This work confirms that dropout individuals contain important information and should be included in the OS analysis. However, there are uncertainties about how to parameterize these events when both hazard functions (drop out and death) are estimated simultaneously. Future analysis are needed to continue to evaluate the impact of the different parametrization in the estimation of the hazard functions for each event.
References:
[1] Schindler E, Amantea MA, Karlsson MO, Friberg LE. A pharmacometric framework for axitinib exposure, efficacy, and safety in metastatic renal cell carcinoma patients. CPT Pharmacometrics Syst Pharmacol. 2017 Jun 1;6(6):373–82.
[2] Zheng Y, Narwal R, Jin CY, Baverel PG, Jin X, Gupta A, et al. Population Modeling of Tumor Kinetics and Overall Survival to Identify Prognostic and Predictive Biomarkers of Efficacy for Durvalumab in Patients With Urothelial Carcinoma. Clin Pharmacol Ther. 2018 Apr 1;103(4):643–52.
Reference: PAGE () Abstr 9512 [www.page-meeting.org/?abstract=9512]
Poster: Methodology - New Modelling Approaches