A Sociedade Portuguesa de Estatística (SPE) xunto coa SGAPEIO organizan a seguinte sesión paralela especial de Biometría o venres 24 de outubro, 12:30 - 13:30.

• Inferring Infectious Disease Risk in Real Time – Methodology and Public Health Implications

• Ricardo Águas, Nuffield Department of Medicine, University of Oxford, United Kingdom

RESUMO

Accounting for heterogeneity has been shown to be critical in models of infectious disease transmission. Traditional compartmental models, often defaulted to in outbreak scenarios, oversimplify disease dynamics by assuming homogeneous susceptibility within, and homogeneous mixing between, modelled populations. We consider an extension of the traditional susceptible-infected-recovered model that allows for heterogeneity in susceptibility and test our ability to statistically infer its parameters in real-time using simulated data with state-of-the-art software. A simple retroactive validation and real-time limit test across a set of reasonable parameter values demonstrate our ability to infer these parameters with a high degree of certainty. We quantify the value of detecting this heterogeneity in real-time and discuss the implications on the timing and scope of public health mitigation policies. An instantaneous measure of population-level heterogeneity in the context of a rapidly evolving infectious disease landscape with quantifiable uncertainty adds value to health policy decision-making. Milder interventions can be most effective for relatively high levels of heterogeneity due to the infection-by-selection phenomenon, which disproportionately decreases the mean level of susceptibility in the population as the disease dynamics evolve.

• Nonparametric Model Check for Cure Rate Quantile Regression

• Mercedes Conde-Amboage^1,2, Wenceslao González-Manteiga^1,2, César A. Sánchez-Sellero ^1,2

• ¹  Department of Statistics, Mathematical Analysis and Optimization. Faculty of Mathematics. Universidade de Santiago de Compostela (USC), Santiago de Compostela, Spain.

• ² Galician Center for Mathematical Research and Technology (CITMAga), Santiago de Compostela, Spain.

RESUMO

In classical survival analysis, a fundamental assumption is that all individuals will eventually experience the event of interest. However, it often occurs that a subset of subjects will never experience the event. These individuals are typically considered to have infinite survival times and are classified as “cured”. To deal with this phenomenon, classical survival models have been extended to what is commonly referred to as cure models. A thorough review of this kind of models from the standpoint of classical mean regression can be found in Amico and Van Keilegom (2018). On the other hand, in the context of quantile regression (which aim is to provide a more detailed description of the conditional distribution of the response variable) the problem of estimating a cure rate model has been scarcely addressed in the literature. Specifically, only the work of Wu and Yin (2013), and more recently that of Narissety and Koenker (2022), can be highlighted. Throughout this talk, a new lack-of-fit test for cure models in the context of quantile regression is presented. This new proposal represents the first contribution in the literature to test the effect of a group of covariates on a survival time using empirical processes marked by residuals. The asymptotic behaviour of the test statistics will be derived. In addition, an extensive simulation study and a real data application will be presented to show the performance of the new proposal in practice.