Derivation and Internal Validation of a Model to Predict the Probability of Severe Acute Respiratory Syndrome Coronavirus-2 Infection in Community People

Abstract

Background: The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) causes COVID-19 disease. There are concerns regarding limited testing capacity and the exclusion of cases from unproven screening criteria. Knowing COVID-19 risks can inform testing. This study derived and assessed a model to predict risk of SARS-CoV-2 in community-based people. Methods: All people presenting to a community-based COVID-19 screening center answered questions regarding symptoms, possible exposure, travel, and occupation. These data were anonymously linked to SARS-CoV-2 testing results. Logistic regression was used to derive a model to predict SARS-CoV-2 infection. Bootstrap sampling evaluated the model. Results: A total of 9172 consecutive people were studied. Overall infection rate was 6.2% but this varied during the study period. SARS-CoV-2 infection likelihood was primarily influenced by contact with a COVID-19 case, fever symptoms, and recent case detection rates. Internal validation found that the SARS-CoV-2 Risk Prediction Score (SCRiPS) performed well with good discrimination (c-statistic 0.736, 95%CI 0.715–0.757) and very good calibration (integrated calibration index 0.0083, 95%CI 0.0048–0.0131). Focusing testing on people whose expected SARS-CoV-2 risk equaled or exceeded the recent case detection rate would increase the number of identified SARS-CoV-2 cases by 63.1% (95%CI 54.5–72.3). Conclusion: The SCRiPS model accurately estimates the risk of SARS-CoV-2 infection in community-based people undergoing testing. Using SCRiPS can importantly increase SARS-CoV-2 infection identification when testing capacity is limited.