Towards a COVID-19 symptom triad: The importance of symptom constellations in the SARS-CoV-2 pandemic

Abstract

Pandemic scenarios like SARS-Cov-2 require rapid information aggregation. In the age of eHealth and data-driven medicine, publicly available symptom tracking tools offer efficient and scalable means of collecting and analyzing large amounts of data. As a result, information gains can be communicated to front-line providers. We have developed such an application in less than a month and reached more than 500 thousand users within 48 hours. The dataset contains information on basic epidemiological parameters, symptoms, risk factors and details on previous exposure to a COVID-19 patient. Exploratory Data Analysis revealed different symptoms reported by users with confirmed contacts vs. no confirmed contacts. The symptom combination of anosmia, cough and fatigue was the most important feature to differentiate the groups, while single symptoms such as anosmia, cough or fatigue alone were not sufficient. A linear regression model from the literature using the same symptom combination as features was applied on all data. Predictions matched the regional distribution of confirmed cases closely across Germany, while also indicating that the number of cases in northern federal states might be higher than officially reported. In conclusion, we report that symptom combinations anosmia, fatigue and cough are most likely to indicate an acute SARS-CoV-2 infection.

Fig 1. Comparison of symptom distribution between patients with and without confirmed contact.

A) Single symptoms: Color coded Cramer’s V correlation of symptoms with the confirmed contact variable (dark red tones) & symptom frequency count in percent of positive statements broken down by groups with and without confirmed contact. Anosmia seems to be the strongest predictor followed by fever and dyspnea at rest. On the contrary, the least single important symptoms are sniff, fatigue and cough by itself. B) Complex symptoms: Symptom frequency count (total) with combinations. The symptom combination fatigue, anosmia and cough has been highlighted in red to illustrate the shift of importance between the two groups. The age distribution of the two groups is depicted in each case above graph B. A random sample of 19128 was taken from the population without confirmed contact for comparison. The percentage of positive cases in the total number of participants without contact was 6·24% whereas the percentage of positive cases in the group with confirmed contact was 23·21%.

Fig 2

Maps illustrating the total number of users of COVID-Online (A), the number of predicted infections (B), the number of confirmed infections (C), the frequency of fatigue (D), the frequency of anosmia (E) and the frequency of fever (F). Data of maps A–F is based on the time period 03.04–10.04.2020. A, B, D, E, F: The district “Marburg-Biedenkopf” has been excluded from these charts as it contained too many records from internal tests carried out by associated personnel of COVID-Online and was also influenced by regional media reports. Unfortunately, due to the great time pressure in times of crisis, no test or staging instance could be installed for such purposes.