Uma abordagem bayesiana para avaliação de concordância entre dispositivos de medição de uma variável funcional

Abstract

It is common in the industrial and clinical context to seek more precision, accurate, accessible and cheaper equipments. Therefore, it is important for these fields to have a way to compare the agreement of new devices with one that already exists and that is assumed to be reliable and accurate. On the other hand, it is increasingly common that systems record data over time with a very fine discretization, creating functional data. Considering this scenario, we propose the approach “Predictive Probability of Functional Agreement” (PPFA), a technique for analyzing the concordance between functional data under a Bayesian perspective, which allows determining whether two systems can be used interchangeably (a nomenclature used in studies of comparison of measurement system). The PPFA generates an intuitive measure of agreement (the predictive probability) of the curves from a new subject (from both devices) differ by less than an acceptable amount for the field. An example of equipments that need further analysis on the degree of agreement are the Optotrak and the flexible electrogoniometer. These systems are used in gait studies. The Optotrak is a laboratory device and generates accurate data, while the electrogoniometer is an ambulatory and cheaper equipment than Optotrak. In order to carry out the comparison study between the Optotrak and the electrogoniometer with our Predictive Probability of Functional Agreement tool, we need to assume a model. Therefore, we present the Bayesian Functional Gaussian Process hierarchical model (BFGP). The PPFA is a simple and intuitive approach, while the BFGP is an easily interpreted model for the gait problem.