Control charts for health surveillance based on residuals of negative binomial regression

Abstract

AbstractThis research presents control charts based on one‐sided exponentially weighted moving average (EWMA) control charts of Pearson, studentized Pearson, Deviance, and Studentized Deviance residuals for a negative binomial regression in order to detect increases in count quality variables used in health surveillance. Data of weekly dengue cases for the year 2018 from the coast of Ecuador were used for the design of control charts in phase I because in that year the weather conditions were normal, while dengue cases in 2019 were used for phase II of control. The control charts detected an increase in the weekly cases of dengue during 2019, a year in which the phenomenon of the Niño was present. In this study by simulations, it is shown that the smallest lambda values have a faster speed in detecting changes. If the objective is greater sensitivity, using studentized residuals is better, but if we seek simplicity and interpretation, Pearson residuals can be used.