SARS-CoV-2 rapid antigen testing in the healthcare sector: A clinical prediction model for identifying false negative results
Johannes Leiner, Vincent Pellissier, Anne Nitsche, Sebastian König, Sven Hohenstein, Irit Nachtigall, Gerhard Hindricks, Christoph Kutschker, Boris Rolinski, Julian Gebauer, Anja Prantz, Jöerg Schubert, Joerg Patzschke, Andreas Bollmann, Martin Wolz
Abstract
OBJECTIVES: SARS-CoV-2 rapid antigen tests (RAT) provide fast identification of infectious patients when RT-PCR results are not immediately available. We aimed to develop a prediction model for identification of false negative (FN) RAT results. METHODS: 2021 were retrospectively analyzed regarding clinical findings. Variables included demographics, laboratory values and specific symptoms. Three different models were evaluated using Bayesian logistic regression. RESULTS: The initial dataset contained 4,076 patients. Overall sensitivity and specificity of RAT was 62.3% and 97.6%. 2,997 cases with negative RAT results (FN: 120; true negative: 2,877; reference: RT-PCR) underwent further evaluation after removal of cases with missing data. The best-performing model for predicting FN RAT results containing 10 variables yielded an area under the curve of 0.971. Sensitivity, specificity, PPV and NPV for 0.09 as cut-off value (probability for FN RAT) were 0.85, 0.99, 0.7 and 0.99. CONCLUSION: FN RAT results can be accurately identified through ten routinely available variables. Implementation of a prediction model in addition to RAT testing in clinical care can provide decision guidance for initiating appropriate hygiene measures and therefore helps avoiding nosocomial infections.