Prediction of SARS-CoV-2 infection with a Symptoms-Based model to aid public health decision making in Latin America and other low and middle income settings

Abstract

Symptoms-based models for predicting SARS-CoV-2 infection may improve clinical decision-making and be an alternative to resource allocation in under-resourced settings. In this study we aimed to test a model based on symptoms to predict a positive test result for SARS-CoV-2 infection during the COVID-19 pandemic using logistic regression and a machine-learning approach, in Bogotá, Colombia. Participants from the CoVIDA project were included. A logistic regression using the model was chosen based on biological plausibility and the Akaike Information criterion. Also, we performed an analysis using machine learning with random forest, support vector machine, and extreme gradient boosting. The study included 58,577 participants with a positivity rate of 5.7%. The logistic regression showed that anosmia (aOR = 7.76, 95% CI [6.19, 9.73]), fever (aOR = 4.29, 95% CI [3.07, 6.02]), headache (aOR = 3.29, 95% CI [1.78, 6.07]), dry cough (aOR = 2.96, 95% CI [2.44, 3.58]), and fatigue (aOR = 1.93, 95% CI [1.57, 2.93]) were independently associated with SARS-CoV-2 infection. Our final model had an area under the curve of 0.73. The symptoms-based model correctly identified over 85% of participants. This model can be used to prioritize resource allocation related to COVID-19 diagnosis, to decide on early isolation, and contact-tracing strategies in individuals with a high probability of infection before receiving a confirmatory test result. This strategy has public health and clinical decision-making significance in low- and middle-income settings like Latin America.