Técnicas estatísticas para avaliação de novos marcadores de risco: aplicações envolvendo o Modelo de Cox

Abstract

New risk prediction markers have been proposed in the literature. Thus, new formulations of risk are investigated with the intention of incorporating these markers. It is important to evaluate the usefulness of a new marker for risk prediction. The ROC curve has shown limitations in the evaluation of such models. We present in this dissertation the following recent statistical techniques for markers evaluation and models of risk prediction: the risk stratication table, the predictiveness curve, the net reclassication improvement (NRI) and the integrated discrimination improvement (IDI). Monte Carlo simulations were used in order to evaluate the performance of tests related to NRI and IDI. Under the null hypothesis, the NRI test showed rejection rates closer to the nominal level than the IDI one. Under the alternative hypothesis, the IDI test had more power in relation to the NRI test and is less sensitive to the sample size variation. An important aim of this work was to assess the importance of the E/E ratio in the performance of the model to predict risk of early death in patients with cardiac Chagas disease. All techniques, except for the IDI, pointed out that the E/E ratio does not add predictive value to the model