A model to predict SARS-CoV-2 infection based on the first three-month surveillance data in Brazil.

Fredi A Diaz-Quijano; Jose Mario Nunes da Silva; Fabiana Ganem; Silvano Oliveira; Andrea Liliana Vesga-Varela; Julio Croda

Most of the clinical manifestations included in the model were negatively associated with the SARS-CoV-2 infection. It does not mean that they cannot be presented by patients with COVID-19, but that they are more frequent in other diseases. This finding highlights why the circulation of other infectious agents could be a determinant of the predictors' discriminatory capacity, as has already been suggested for other conditions. 18 Moreover, it is expected that variables determining the notification (e.g., respiratory symptoms and international travel) and, therefore, inclusion in the study, tend to be negatively associated with the outcome due to collider-like phenomena. 19 For this reason, both causal inference interpretation, and extrapolation to the general population of the associations would be biased. Consequently, our model must be considered only for diagnostic prediction in the specific group of reported suspected patients.