Evento rientra nelle attività organizzate per il 40° Anniversario di Economia

Abstract: Pre-clinical cancer research often involves animal studies to evaluate the safety and efficacy of new treatments before they reach human clinical trials. In the study of cancer growth-curves, in order to preserve animal welfare. animals are euthanized when their pain level is considered too high, as evaluated by a veterinary, or when the tumor size is larger than a set threshold. From a statistical modelling and inferential point of view, this data feature needs to be considered in order to assess, for example, the treatment effect measured by the difference in growth curves between treatment and control groups. In particular, false discovery rates can increase drastically (or test sizes decrease drastically) when the censoring mechanism is not correctly modeled. Therefore, within a general simulation-based framework for inference with complex data settings, we propose a simple statistical procedure to produce correct inferential tools when data are censored, below, above or by intervals, for censoring data mechanisms that can be modeled according to the experimental settings. The proposed approach does not require separate analytical developments for different models and/or censoring data mechanisms, making it easy to implement in a variety of experimental situations. Additionally, the censoring thresholds are also allowed to be random and to depend on the growth size of the curves, which is a more realistic practical situation, as attested by a case study on tumour growth inhibition that is used as a motivating example. The theoretical results are based on Zhang et al. (2023) and Orso et al. (2024)