Anosmia, ageusia, and other COVID-19-like symptoms in association with a positive SARS-CoV-2 test, across six national digital surveillance platforms: an observational study

Carole H. Sudre, Ayya Keshet, Mark S. Graham, Amit D. Joshi, Smadar Shilo, Hagai Rossman, Benjamin Murray, Erika Molteni, Kerstin Kläser, Liane S. Canas, Michela Antonelli, Long H. Nguyen, David A. Drew, Marc Modat, Joan Capdevila Pujol, Sajaysurya Ganesh, Jonathan Wolf, Tomer Meir, Andrew T. Chan, Claire J. Steves, Tim D. Spector, John S. Brownstein, Eran Segal, Sébastien Ourselin, Christina M. AstleyThe Lancet Digital Health 2021 · DOI

TL;DR

This study leverages data from six large-scale digital surveillance platforms across three countries to identify the most reliable symptoms for detecting COVID-19. It confirms that anosmia (loss of smell) and ageusia (loss of taste) are exceptionally strong and consistent indicators of a positive SARS-CoV-2 test, outperforming other common symptoms like fever and cough. These findings are crucial for improving real-time public health surveillance, especially in regions with limited testing capacity.

What Problem Does This Paper Address?

The COVID-19 pandemic highlighted the urgent need for rapid, population-level understanding of disease spread. Traditional surveillance methods often lag behind, and the early stages of the pandemic were characterized by limited testing capacity and evolving diagnostic criteria. This created a challenge in accurately identifying infected individuals and tracking the virus's prevalence in near real-time. While various digital platforms emerged to collect symptom data, it was unclear which symptoms were consistently reliable indicators of infection across different platforms, countries, and evolving testing policies. This study aims to address this by systematically comparing the association of various symptoms with positive SARS-CoV-2 test results across multiple diverse digital surveillance systems.

Methods

The researchers conducted an observational study using data from six voluntary digital surveillance platforms across Israel, the UK, and the USA. These platforms included the Carnegie Mellon University and University of Maryland Facebook COVID-19 Symptom Survey, the ZOE COVID Symptom Study app, and the Corona Israel study. The study population comprised adult respondents (aged 18-100) who were not healthcare workers.

The core analytical approach involved logistic regression to model the association between self-reported symptoms and self-reported SARS-CoV-2 test status (positive vs. negative). These models were adjusted for age and sex. To synthesize findings across the different platforms and countries, the researchers performed meta-analyses using a random-effects model. They also employed robust rank aggregation to combine the rankings of symptom-test positivity odds ratios (ORs) from each platform. The study also considered the impact of testing policy changes, COVID-19 incidence, and the duration of symptoms on these associations.

Datasets

The study utilized data from six national digital surveillance platforms:

  • Carnegie Mellon University (CMU) and University of Maryland (UMD) Facebook COVID-19 Symptom Survey: This platform collected data from the USA, UK, and Israel.
  • ZOE COVID Symptom Study app: This platform collected data from the UK and USA.
  • Corona Israel study: This platform collected data from Israel.

In total, the study analyzed data from over 10 million respondents, encompassing 514,459 SARS-CoV-2 tests. The data spanned from April 1 to July 31, 2020.

Key Insights and Ideas

  • Anosmia/Ageusia as the "Super Symptom": The most striking finding is the exceptionally strong and consistent association of anosmia (loss of smell) and ageusia (loss of taste) with a positive COVID-19 test. This symptom consistently ranked highest across all platforms and countries, suggesting its high specificity and utility as an early warning signal.
  • Generality of Key Symptoms: Beyond anosmia/ageusia, fever, shortness of breath, and cough also emerged as consistently strong predictors of a positive test. This reinforces their value as core indicators for syndromic surveillance.
  • Platform and Country Agnosticism: The study demonstrates that these key symptom-test associations hold true across different digital surveillance platforms and national contexts, highlighting the generalizability of these findings.
  • Impact of Testing Policies: The study observed that changes in testing criteria (e.g., broadening the definition of "COVID-19-like illness" to include more symptoms) and periods of higher COVID-19 incidence could influence the strength of symptom associations, though the top-ranked symptoms remained robust.
  • Gastrointestinal Symptoms Less Reliable: Unlike respiratory and sensory symptoms, gastrointestinal symptoms (like nausea and diarrhea) were found to be less consistently predictive of a positive test, particularly in individuals with fewer symptoms or during periods of low incidence.

Key Results

  • Anosmia-ageusia was the strongest and most consistent symptom associated with a positive COVID-19 test, with a meta-analyzed random effects odds ratio (OR) of 16.96 (95% CI 13.13–21.92).
  • Fever ranked second (OR 6.45, 95% CI 4.25–9.81).
  • Shortness of breath ranked third (OR 4.69, 95% CI 3.14–7.01).
  • Cough ranked fourth (OR 4.29, 95% CI 3.13–5.88).
  • The study found that the association of symptoms with test status varied by duration of illness, timing of the test, and broader test criteria, but the relative strength of the top symptoms remained consistent.

Limitations and Future Work

The authors acknowledge several limitations:

  • Selection Bias: Participants in digital surveillance platforms may not be fully representative of the general population, potentially skewing results towards younger, healthier, or more tech-savvy individuals.
  • Measurement Bias: Self-reported symptoms can be subject to misinterpretation, recall bias, or social desirability bias. The study also notes that subjective interpretations of symptoms might differ from clinical assessments.
  • Data Heterogeneity: While efforts were made to harmonize data, differences in survey question wording and data collection methods across platforms (e.g., symptom recall windows) could influence effect estimates.
  • Limited Scope: The study focused on a specific time period (April-July 2020) and did not assess the clinical evaluation of symptoms (e.g., measured fever vs. self-reported). It also did not account for individuals being tested multiple times.

Future work suggested by the authors includes:

  • Discriminating COVID-19 from other respiratory pathogens: Developing methods to differentiate COVID-19 symptoms from those of seasonal illnesses like influenza.
  • Prospective, iterative collaboration: Continuing to use multi-platform data to refine epidemiological understanding and surveillance tools.
  • Investigating dysgeusia: Exploring the role of dysgeusia (distorted taste) as a potential symptom, which might be closely related to anosmia/ageusia but not always captured distinctly.

Summary updated

Abstract

National Institutes of Health, National Institute for Health Research, Alzheimer's Society, Wellcome Trust, and Massachusetts Consortium on Pathogen Readiness.
BibTeX
@inproceedings{sudre2021anosmia,
  author = {Carole H. Sudre and Ayya Keshet and Mark S. Graham and Amit D. Joshi and Smadar Shilo and Hagai Rossman and Benjamin Murray and Erika Molteni and Kerstin Kläser and Liane S. Canas and Michela Antonelli and Long H. Nguyen and David A. Drew and Marc Modat and Joan Capdevila Pujol and Sajaysurya Ganesh and Jonathan Wolf and Tomer Meir and Andrew T. Chan and Claire J. Steves and Tim D. Spector and John S. Brownstein and Eran Segal and Sébastien Ourselin and Christina M. Astley},
  title = {{Anosmia, ageusia, and other COVID-19-like symptoms in association with a positive SARS-CoV-2 test, across six national digital surveillance platforms: an observational study}},
  booktitle = {The Lancet Digital Health},
  year = {2021},
  doi = {10.1016/s2589-7500(21)00115-1},
}
Figure