Differential Diagnosis of COVID-19 and Influenza: a Web-Based Predictive Tool Designed with Available Data from China and the U.S. (Preprint)

BACKGROUND Limited guidance is available on how community-based health care providers can diagnose COVID-19 or how employers can monitor employees’ symptoms upon return to work following shelter-in-place or other state-enforced orders. OBJECTIVE This analysis provides a method of assessing risk of COVID-19 in the community as opposed to hospital settings. METHODS Published data shows how asymptomatic COVID-19 patients can be diagnosed based on demographic (age, race, and gender), clinical (i.e. anosmia, blood oxygen levels), and exposure (i.e. close contact in health care setting, use of public transportation, recent travel, social distancing practices) data. For symptomatic patients we contrast symptoms of COVID-19 patients (709 Chinese and 272 United States) patients against 2,885 United States influenza and 884 other influenza-like illnesses patients. Two models were developed for diagnosing COVID-19 in the community-based settings: 1) during flu season and 2) after flu season. Accuracy of the predictions was calculated using the micro average, and the sample weighted micro average, Area under the Receiver Operating Characteristic (AROC) curves. RESULTS Fever and cough were the two most common COVID-19 symptoms. If clinicians use these two symptoms, they would make numerous errors (AROC=0.49). They would correctly diagnose 48% of COVID-19 patients, and correct ruling out of 68% of non-COVID-19 patients. In contrast to the simplified rules, using all patients’ symptoms (not just fever and cough) correctly diagnosed COVID-19 with AROC of 0.79. CONCLUSIONS Community-based health care providers and large employers screening employees returning to work need continuously updated tools that adjust for flu season and the diagnostic value of known signs and symptoms of COVID-19. Simplified rules do not work. Diagnosing COVID-19 in the community settings requires probabilistic inferences that are beyond human capabilities. To assist in the rapid diagnoses, we developed a free web calculator: http://hi.gmu.edu/c19