Time to event models of survival in cancer of pancreas : confirmation of explanatory variables pre-selected by bootstrap analysis.
C.Pobel(1), M.Guery(1), G.Herbin(1), E.Kiep(1), C.Donamaria(2)
(1) Clinical pharmacology, CH Saintonge, Saintes, France; (2) Clinical pharmacology, Institut Bergonié, Bordeaux, France
Objectives: Treatment of pancreas cancer often involves protocols out of references. These are argued by phase 2 studies, small cohorts studies or case reports. Important goal is to rank these protocols according to efficacy and safety.This ranking and associated statistical models are of potential interest for patient, help to proritize phase 3 studies to undertake, help to design phase 3 studies. Multivariate statistical analysis lead to select appropriate explanatory variables, with events of interest like progression free survival, tumor size kinetics, score of toxicity, and for this study time to death. Here modelisation involve logistic regression, Weibull model, in order to confirm explanatory variables pre-selected by 2-stage bootstrap analysis.
Methods: The population multivariate analysis was performed using NONMEM based on datas from a cohort of 42 unselected patients. Logistic regression and Weibull models were implemented in NONMEM with prediction of probability of death (at 12 months)(1)(2) , and probability of non observed event, respectively. Confirmation of explanatory variables pre-selected was test on individuals predictions.
Results: 42 patients were analysed, with combinations of 12 different protocols of chemotherapy. 16 pre-selected variables were : age ( more or less 65 ), stage at diagnosis ( local or metastatic ) , number of treatment lines ( more/less 2), first protocol schedule, prior surgery, dose reduction, platinium salt introduction, gemcitabin-oxaliplatin protocol exposure, erlotinib exposure. Check of individual probabilities predictions against the variables showed agreements with 2-stage bootstrap selection.
Conclusions: Such models as part of more global analysis strategy can take into account separation of sources of variability ( by use of mixed effects models), and interactions between explanatory variables. This techniques are efficient for sparse and heterogenous datas like outliers of references in cancer treatment. This methodology will improve the determination of the prognostic indicators. Such approach could also be part of methods be used for clinical trial simulation, a technique allowing to avoid clinical studies having a high risk of failure, and to properly design future clinical trials.
References:
[1] A pharmacodynamic Markov mixed-effects model for the effect of temazepam on sleep Mats O. Karlsson and all. Clin Pharm Ther 2000/08 175-188
[2] Pharmacokinetic/Pharmacodynamic and Time-to-event models of ribaviri-induced anaemia in chronic hepatitis C Michel Tod and all. Clin Pharmacokinet; 44 (4): 417-428
