Anticipating the curve: can online symptom-based data reflect COVID-19 case activity in Ontario, Canada?

Abstract

BackgroundLimitations in laboratory diagnostic capacity and reporting delays have hampered efforts to mitigate and control the ongoing COVID-19 pandemic globally. Syndromic surveillance of COVID-19 is an important public health tool that can help detect outbreaks, mobilize a rapid response, and thereby reduce morbidity and mortality. The primary objective of this study was to determine whether syndromic surveillance through self-reported COVID-19 symptoms could be a timely proxy for laboratory-confirmed case trends in the Canadian province of Ontario.MethodsWe retrospectively analyzed self-reported symptoms data collected using an online tool – Outbreaks Near Me (ONM) – from April 20th to Oct 11th, 2020 in Ontario, Canada. We estimated the correlation coefficient between the weekly proportion of respondents reporting a COVID-like illness (CLI) to both the weekly number of PCR-confirmed COVID-19 cases and the percent positivity in the same period for the same week and with a one-week lag.ResultsThere were 314,686 responses from 188,783 unique respondents to the ONM symptom survey. Respondents were more likely to be female and be in the 40-59 age demographic compared to the Ontario general population. There was a strong positive correlation between the weekly number of reported cases in Ontario and the percent of respondents reporting CLI each week (r = 0.89, p <0.01) and with a one-week lag (r = 0.89, p <0.01).InterpretationWe demonstrate a strong positive and significant correlation (r = 0.89, p <0.01) between percent of self-reported COVID-like illness and the subsequent week’s COVID-19 cases reported, highlighting that a rise in CLI may precede official statistics by at least 1 week. This demonstrates the utility of syndromic surveillance in predicting near-future disease activity. Digital surveillance systems are low-cost tools that may help measure the burden of COVID-19 in a community if there is under-detection of cases through conventional laboratory diagnostic testing. This additional information can be used to guide a healthcare response and policy decisions.

Consistent comparison of symptom-based methods for COVID-19 infection detection

“…In this sense, various predicting rules have been proposed that combine multiple individual features (age, gender, symptoms, demographic data, etc. ), to characterize COVID-19 infected people [6,7,8,9,10,11,12,13,14,15,16,17,18,19]. These detection methods typically rely on the presence of the most predictive combination of symptoms in order to identify a COVID-19 active case.…”

Predicting COVID-19 case status from self-reported symptoms and behaviors using data from a massive online survey

“…Gomathi et al (2020) used similar features, along with details regarding international travel to predict COVID-19 in India. Maharaj et al (2021) (in Canada) and Menni et al (2020) (in the UK) also used symptoms to predict COVID-19. One limitation of these studies is the use of small datasets (100-100,000 samples).…”

Consistent Comparison of Symptom-based Methods for COVID-19 Infection Detection

“…In this sense, various predicting rules have been proposed that combine multiple individual features (age, gender, symptoms, demographic data, etc. ), to characterize COVID-19 infected people [6,7,8,9,10,11,12,13,14,15,16,17,18,19]. These detection methods typically rely on the presence of the most predictive combination of symptoms in order to identify a COVID-19 active case.…”