False-negative RT-PCR for COVID-19 and a diagnostic risk score: a retrospective cohort study among patients admitted to hospital

ObjectiveTo describe the characteristics and outcomes of patients with a clinical diagnosis of COVID-19 and false-negative SARS-CoV-2 reverse transcription-PCR (RT-PCR), and develop and internally validate a diagnostic risk score to predict risk of COVID-19 (including RT-PCR-negative COVID-19) among medical admissions.DesignRetrospective cohort study.SettingTwo hospitals within an acute NHS Trust in London, UK.ParticipantsAll patients admitted to medical wards between 2 March and 3 May 2020.OutcomesMain outcomes were diagnosis of COVID-19, SARS-CoV-2 RT-PCR results, sensitivity of SARS-CoV-2 RT-PCR and mortality during hospital admission. For the diagnostic risk score, we report discrimination, calibration and diagnostic accuracy of the model and simplified risk score and internal validation.Results4008 patients were admitted between 2 March and 3 May 2020. 1792 patients (44.8%) were diagnosed with COVID-19, of whom 1391 were SARS-CoV-2 RT-PCR positive and 283 had only negative RT-PCRs. Compared with a clinical reference standard, sensitivity of RT-PCR in hospital patients was 83.1% (95% CI 81.2%–84.8%). Broadly, patients with false-negative RT-PCR COVID-19 and those confirmed by positive PCR had similar demographic and clinical characteristics but lower risk of intensive care unit admission and lower in-hospital mortality (adjusted OR 0.41, 95% CI 0.27–0.61). A simple diagnostic risk score comprising of age, sex, ethnicity, cough, fever or shortness of breath, National Early Warning Score 2, C reactive protein and chest radiograph appearance had moderate discrimination (area under the receiver–operator curve 0.83, 95% CI 0.82 to 0.85), good calibration and was internally validated.ConclusionRT-PCR-negative COVID-19 is common and is associated with lower mortality despite similar presentation. Diagnostic risk scores could potentially help triage patients requiring admission but need external validation.