Early detection of COVID-19 in the UK using self-reported symptoms: a large-scale, prospective, epidemiological surveillance study

Abstract

Background Self-reported symptoms during the COVID-19 pandemic have been used to train artificial intelligence models to identify possible infection foci. To date, these models have only considered the culmination or peak of symptoms, which is not suitable for the early detection of infection. We aimed to estimate the probability of an individual being infected with SARS-CoV-2 on the basis of early self-reported symptoms to enable timely self-isolation and urgent testing. MethodsIn this large-scale, prospective, epidemiological surveillance study, we used prospective, observational, longitudinal, self-reported data from participants in the UK on 19 symptoms over 3 days after symptoms onset and COVID-19 PCR test results extracted from the COVID-19 Symptom Study mobile phone app. We divided the study population into a training set (those who reported symptoms between April 29, 2020, and Oct 15, 2020) and a test set (those who reported symptoms between Oct 16, 2020, and Nov 30, 2020), and used three models to analyse the selfreported symptoms: the UK's National Health Service (NHS) algorithm, logistic regression, and the hierarchical Gaussian process model we designed to account for several important variables (eg, specific COVID-19 symptoms, comorbidities, and clinical information). Model performance to predict COVID-19 positivity was compared in terms of sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) in the test set. For the hierarchical Gaussian process model, we also evaluated the relevance of symptoms in the early detection of COVID-19 in population subgroups stratified according to occupation, sex, age, and body-mass index. Findings The training set comprised 182 991 participants and the test set comprised 15 049 participants. When trained on 3 days of self-reported symptoms, the hierarchical Gaussian process model had a higher prediction AUC (0•80 [95% CI 0•80-0•81]) than did the logistic regression model (0•74 [0•74-0•75]) and the NHS algorithm (0•67 [0•67-0•67]). AUCs for all models increased with the number of days of self-reported symptoms, but were still high for the hierarchical Gaussian process model at day 1 (0•73 [95% CI 0•73-0•74]) and day 2 (0•79 [0•78-0•79]). At day 3, the hierarchical Gaussian process model also had a significantly higher sensitivity, but a non-statistically lower specificity, than did the two other models. The hierarchical Gaussian process model also identified different sets of relevant features to detect COVID-19 between younger and older subgroups, and between health-care workers and non-health-care workers. When used during different pandemic periods, the model was robust to changes in populations.Interpretation Early detection of SARS-CoV-2 infection is feasible with our model. Such early detection is crucial to contain the spread of COVID-19 and efficiently allocate medical resources.

Symptoms and SARS-CoV-2 positivity in the general population in the UK

“…REACT also suggested having different symptom sets for adults and children in order to optimise sensitivity, which would require careful public health messaging. ZOE [8] suggested an algorithm also including working in healthcare; whilst this could theoretically be programmed into an online test system, such complexity risks gaming the system if individuals cannot otherwise get a test.…”

“…Most studies to date have restricted to those hospitalised or seeking healthcare, who do not represent most infections [6]. Three recent UK community-based studies suggested that sensitivity could be increased by 10-20% by extending the "classic" symptoms (REACT [7], adding combinations of headache, muscle aches, chills and appetite loss depending on age; ZOE [8], adding different symptoms depending on age, sex, BMI and working in healthcare; VirusWatch [9], adding feeling feverish, headache, muscle aches, loss of appetite or chills) but at a cost of increasing numbers eligible for testing 2-3 fold and tests per positive identified up to 7-fold. However, these studies were mainly before widespread vaccination, and whilst Alpha dominated.…”

Creating symptom-based criteria for diagnostic testing: a case study based on a multivariate analysis of data collected during the first wave of the COVID-19 pandemic in New Zealand

“…To date there have been relatively few other reported studies of COVID-19 symptoms, and even fewer reporting their use to inform public health actions or clinical decision making [ 4 – 7 , 9 , 10 , 29 , 30 ]. A study using data from the UK Covid Symptom Smartphone Application used a similar approach to the New Zealand study, combining symptom data with demographic information to predict COVID-19 PCR test positivity [ 6 ]. Models were trained using data collected at several time points and showed a similar level of prediction of test positivity to the New Zealand random forest model (AUC ~ 0.8).…”

“…Other studies have described the use of self-reported symptoms to develop and train models that can identify infection, with the aim of identifying cases early in the course of infection, enabling prompt self-isolation and testing [ 4 – 6 ]. These include the use of large datasets of self-reported symptoms and PCR test results, collected via mobile phone applications, and an array of modelling methods including logistic regression [ 4 , 5 ] and a hierarchical Gaussian process model [ 6 ] and machine learning models [ 7 , 8 ]. Self-reported symptoms have also been used to identify changes in the symptomatology and disease profiles associated with vaccination [ 9 ] and the introduction of new variants of concern [ 10 ].…”

Deep Learning Based Early Detection Framework for Preliminary Diagnosis of COVID-19 via Onboard Smartphone Sensors

“…Self-reported symptoms during the COVID-19 pandemic have been used to train AI-based models to identify possible COVID-19 infection. Although studies based on early self-reported symptoms, such as [ 23 , 24 ], are widely used for identifying COVID-19 cases, no validated tool exists for surveying COVID-19 infection in the general population. Indeed, COVID-19 data are gathered in many countries to understand the pandemic and be prepared for the future.…”