COVID Symptoms Analysis

COVID symptoms analysis is part of the work developed in collaboration between King's College London, Harvard Medical School and ZOE Limited to understand the COVID symptoms and their impact on the population and healthcare. This is a population-based study for disease monitoring using data collected via a mobile app:

The COVID-19 Symptom Study app has been developed by the health science company ZOE aiming at the collection of positive COVID-19 patients’ symptoms and, among other research questions, the development of diagnostic and monitorisation tools based on self-reported symptoms (Cristina Menni, 2020).

Main Projects

Early detection of COVID-19 in the UK using self-reported symptoms

Develop a Bayesian framework to predict positive Covid-19 diagnosis, based on the biomarkers input in the Covid-19 Symptoms App. This approach models the correlation between the several markers provided by the app, including both pre-conditions and symptoms, while providing a probabilistic label of the status of the subject.

Study published in Lancet Digital Health: Early detection of COVID-19 in the UK using self-reported symptoms: a large-scale, prospective, epidemiological surveillance study - The Lancet Digital Health

Features Relevance

Sensitivity analysis by age:

Younger individuals report mostly loss of smell (neurological) and respiratory symptoms;
Older individuals show a higher incidence of gastrointestinal symptoms.

Sensitivity analysis by gender:

Male individuals report a wider range of symptoms, with a high incidence of cardiac/respiratory and flu-like symptoms.

Sensitivity analysis by occupation:

The healthcare workers show a wider range of symptoms, selected by the model as relevant.

Predictions uncertainty

Given the Bayesian nature of the model, it is possible to retrieve the likelihood of the predictions and use it as a surrogate measure of certainty of the model:

Younger to middle-aged patients:

Reduced differences in model performance from 2 to 3 days;

Older patients:

The model is more certain, but highly benefits from 3 days of symptoms.

Profiling post-COVID syndrome across different variants of SARS-CoV-2

This project aims:

Assess differences in symptom profiles and duration for Post COVID-19 syndrome (PCS);
Assess if PCS symptom profiles vary with SARS-CoV-2 variants and an individual’s vaccination status at the time of infection.

Study available: Profiling post-COVID syndrome across different variants of SARS-CoV-2 | medRxiv

Main symptoms clusters per variant

Wild-type:

Four symptom clusters for subjects infected by the wild-type variant:

The largest cluster, cluster wild-A (n=138, 51%) was characterised by upper respiratory and central neurological symptoms.
Cluster wild-D (n=37, 14%) exhibited abdominal symptoms as well as central neurological (headache and anosmia), and upper respiratory (sore throat) symptoms also common to other clusters.

Unvaccinated Alpha-variant:

Seven symptom clusters for subjects infected by the alpha variant:

The largest cluster, cluster alpha-A (n=47, 30%), was dominated by anosmia, whereas the smallest, cluster alpha-G (n=13, 8%), was highly heterogeneous and polysymptomatic.
Difficult to optimize due to the smaller sample size.

Vaccinated Delta-variant:

Five symptom clusters for subjects infected by the delta variant:

The largest cluster, cluster delta-A (n=431, 49%) included mainly central neurological symptoms, similar to alpha (Figure 8, Supplementary Figure 9);
The smallest cluster, cluster delta-E (80 subjects, 9%), contained predominantly abdominal symptoms.