Combining Rapid Antigen Testing and Syndromic Surveillance Improves Sensitivity and Specificity of COVID-19 Detection: A Community-Based Prospective Diagnostic Study

Background: Diagnostics for COVID-19 detection are limited in many settings. Syndromic surveillance is often the only means to identify cases, despite lack of specificity and asymptomatic cases. Rapid antigen testing is inexpensive and easy-to-deploy but concerns remain about sensitivity. We examine how combining these approaches can improve surveillance for guiding interventions.Methods: Community-based volunteers were trained to syndromically assess potential COVID-19 cases in low-income communities in Dhaka, Bangladesh. Rapid antigen tests and PCR validation was performed on 1172 syndromically-identified individuals at their households. Statistical models were fit to predict PCR status using rapid-antigen-test results, syndromic data, and their combination. Model predictive and classification performance was examined under contrasting epidemiological scenarios to evaluate their potential for improving diagnoses.Findings: Models combining rapid-antigen-test and syndromic data yielded equal performance to rapid-antigen-test-only models in the “Agnostic” scenario and performed better under scenarios of “Low Incidence” and “Epidemic Growth”. Under “Epidemic Growth”, the combined model’s false negative rate is 26 (IQR:24-29) percentage points lower than the rapid-antigen-test-only model’s, with a false positive rate <20%. Under “Low Incidence” the combined model’s false positive rate is 27 (IQR:23-32) percentage points lower than the rapid-antigen-test-only model’s, with a false negative rate <20%.Interpretation: Drawing on complementary strengths across two rapid diagnostics, we demonstrate improved COVID-19 detection, and reduced false-positive and -negative diagnoses to match local requirements; improvements achievable without additional expense or changes for patients or practitioners. Widespread mobile health applications facilitate these scalable improvements in accessible diagnostics for use in low-income communities.Funding: This work is supported by a grant from the Bill and Melinda Gates Foun dation to FAO (INV-022851). FJC is funded by EPSRC (EP/R513222/1), DJP by the JUNIPER consortium (MR/V038613/1), and KH by Wellcome (207569/Z/17/Z).Declaration of Interest: The authors declare no competing interests.