Fuzzy Model to Support Covid-19 Epidemiological Surveillance in Hospitalized Patients with Severe Acute Respiratory Syndrome

Background One of the limitations of the molecular reference test for the diagnosis of COVID-19 is its sensitivity of around 70% and the possibility of false-negative results due to the pre-analytical conditions of the test. In an epidemiological scenario of high prevalence of COVID-19, the isolated interpretation of negative results in the hospitalized population may underestimate the number of cases of the disease. Objective Present a fuzzy model developed, in support of epidemiological surveillance, to assess the COVID-19 possibility in hospitalized patients with a negative molecular test in a university hospital in Rio de Janeiro, Brazil. Methods A Mamdani-type fuzzy inference model was developed based on the knowledge of three infectiologists in 2020 and implemented in a computational environment using MATLAB® R2019a. Results Seven variables with linguistic terms represented by fuzzy sets were included for model input: oxygen saturation, period of test performance, period of clinical worsening, history of exposure, dyspnea, thromboembolic events and chest tomography. The model output was the COVID-19 possibility represented by four fuzzy sets: very likely, likely, unlikely, and discarded. The model was tested from simulation with five hypothetical cases, and cutoff points related to the degrees of membership to the output sets were defined to support decision-making to classify the case as confirmed or discarded. Conclusion The application of fuzzy logic proved possible and valuable to improve the criteria for confirming COVID-19 based on the indication by specialists of relevant variables for the diagnosis. The model brings possible conditions for the conclusion of cases without robust defining criteria for association with the confirmed and discarded categories: patients with a negative molecular test. The model also qualifies the information that subsidizes implementing public policies to control the disease and has the possibility of practical application through a computational interface.