KCL: Predicting COVID-19 Severity Based On Early Symptomology

#15,370

Since many COVID-19 cases experience only mild to moderate illness - and can safely be treated at home -  researchers and doctors are working to find ways to predict which patients are the most likely to need more intensive, and preferably early, medical interventions. 

Two primary approaches have been tried with moderate success. 

1. Targeting patients with preexisting conditions (Diabetes, Hypertension, BMI, COPD, etc.), or other risk factors (primarily age) (see CDC Risk factors for Severe COVID-19 Disease).
2. Looking at clinical biomarkers in blood tests (D-Dimer, C-reactive protein, thrombocytopenia, etc.) and/or imaging studies (Chest CT scans) that suggest a more aggressive disease progression (see The role of biomarkers in diagnosis of COVID-19 – A systematic review).

But we've also heard some anecdotal reports suggesting some early symptoms - such as gastrointestinal involvement or CNS symptoms - might hint at a greater risk of severe illness.  

In order to try to identify which early symptoms might presage severe illness, researchers at King's College London (KCL) used data submitted by roughly 1,600 COVID-19 cases who logged their early symptoms into a Smartphone application, and compared that to the eventual severity of their illness. 

The pre-print study, published by MedRxiv, identified 6 general categories of symptoms that they linked to varying levels of disease severity.  First, a link and abstract to the study, followed by some excerpts from a press release by KCL. 

Symptom clusters in Covid19: A potential clinical prediction tool from the COVID Symptom study app

View ORCID ProfileCarole H Sudre, Karla Lee, Mary Ni Lochlainn, Thomas Varsavsky, Benjamin Murray, Mark S. Graham,  View ORCID ProfileCristina Menni, Marc Modat, Ruth C.E. Bowyer, Long H Nguyen, David Alden Drew, Amit D Joshi, Wenjie Ma, Chuan Guo Guo, Chun Han Lo, Sajaysurya Ganesh, Abubakar Buwe, Joan Capdevila Pujol, Julien Lavigne du Cadet, Alessia Visconti,  View ORCID ProfileMaxim Freydin, Julia S. El Sayed Moustafa, Mario Falchi, Richard Davies, Maria F. Gomez, Tove Fall, M. Jorge Cardoso, Jonathan Wolf, Paul W Franks, Andrew T Chan, Timothy D Spector, Claire J Steves, Sebastien Ourselin

doi: https://doi.org/10.1101/2020.06.12.20129056

This article is a preprint and has not been peer-reviewed [what does this mean?]. It reports new medical research that has yet to be evaluated and so should not be used to guide clinical practice.

Download PDF

Abstract

As no one symptom can predict disease severity or the need for dedicated medical support in COVID-19, we asked if documenting symptom time series over the first few days informs outcome. Unsupervised time series clustering over symptom presentation was performed on data collected from a training dataset of completed cases enlisted early from the COVID Symptom Study Smartphone application, yielding six distinct symptom presentations. Clustering was validated on an independent replication dataset between May 1- May 28th, 2020. Using the first 5 days of symptom logging, the ROC-AUC of need for respiratory support was 78.8%, substantially outperforming personal characteristics alone (ROC-AUC 69.5%). Such an approach could be used to monitor at-risk patients and predict medical resource requirements days before they are required.

Competing Interest Statement

Zoe Global Limited co-developed the app pro bono for non-commercial purposes. Investigators received support from the Wellcome Trust, the MRC/BHF, EU, NIHR, CDRF, and the NIHR-funded BioResource, Clinical Research Facility and BRC based at GSTT NHS Foundation Trust in partnership with KCL. RD, JW, JCP, AB, SG and JLduC work for Zoe Global Limited and TDS and PWF are consultants to Zoe Global Limited. LHN, DAD, PWF and ATC previously participated as investigators on a diet study unrelated to this work that was supported by Zoe Global Ltd.

          (Continue . . . . )
 

This research is based on self-reported symptoms from a fairly small study cohort, and should be viewed cautiously.  That said, the full paper is well worth reading, and you'll find a good summation in the press release from King's College London (below). 

17 July 2020

Six distinct 'types' of COVID-19 identified

Analysis of data from the COVID Symptom Study app, led by researchers from King’s College London, reveals that there are six distinct ‘types’ of COVID-19, each distinguished by a particular cluster of symptoms.SARS-CoV-2 (NIAID)

Moreover, the team found that these types differed in the severity of the disease and the need for respiratory support during hospitalisation.

The findings have major implications for clinical management of COVID-19, and could help doctors predict who is most at risk and likely to need hospital care in a second wave of coronavirus infections.

Although continuous cough, fever and loss of smell (anosmia) are usually highlighted as the three key symptoms of COVID-19, data gathered from app users shows that people can experience a wide range of different symptoms including headaches, muscle pains, fatigue, diarrhea, confusion, loss of appetite, shortness of breath and more. The progression and outcomes also vary significantly between people, ranging from mild flu-like symptoms or a simple rash to severe or fatal disease.

To find out whether particular symptoms tend to appear together and how this related to the progression of the disease, the research team used a machine learning algorithm to analyse data from a subset of around 1,600 users in the UK and US with confirmed COVID-19 who had regularly logged their symptoms using the app in March and April.

The analysis revealed six specific groupings of symptoms emerging at characteristic timepoints in the progression of the illness, representing six distinct ‘types’ of COVID-19. The algorithm was then tested by running it on a second independent dataset of 1,000 users in the UK, US and Sweden, who had logged their symptoms during May.

All people reporting symptoms experienced headache and loss of smell, with varying combinations of additional symptoms at various times. Some of these, such as confusion, abdominal pain and shortness of breath, are not widely known as COVID-19 symptoms, yet are hallmarks of the most severe forms of the disease.

The six clusters are as follows:

•  (‘flu-like’ with no fever): Headache, loss of smell, muscle pains, cough, sore throat, chest pain, no fever.
• (‘flu-like’ with fever): Headache, loss of smell, cough, sore throat, hoarseness, fever, loss of appetite.
• (gastrointestinal): Headache, loss of smell, loss of appetite, diarrhea, sore throat, chest pain, no cough.
• (severe level one, fatigue): Headache, loss of smell, cough, fever, hoarseness, chest pain, fatigue.
• (severe level two, confusion): Headache, loss of smell, loss of appetite, cough, fever, hoarseness, sore throat, chest pain, fatigue, confusion, muscle pain.
• (severe level three, abdominal and respiratory): Headache, loss of smell, loss of appetite, cough, fever, hoarseness, sore throat, chest pain, fatigue, confusion, muscle pain, shortness of breath, diarrhea, abdominal pain.

Next, the team investigated whether people experiencing particular symptom clusters were more likely to require breathing support in the form of ventilation or additional oxygen.

They discovered that only 1.5% of people with cluster 1, 4.4% of people with cluster 2 and 3.3% of people with cluster 3 COVID-19 required breathing support. These figures were 8.6%, 9.9% and 19.8% for clusters 4,5 and 6 respectively. Furthermore, nearly half of the patients in cluster 6 ended up in hospital, compared with just 16% of those in cluster 1.

Broadly, people with cluster 4,5 or 6 COVID-19 symptoms tended to be older and frailer, and were more likely to be overweight and have pre-existing conditions such as diabetes or lung disease than those with type 1,2 or 3.

The researchers then developed a model combining information about age, sex, BMI and pre-existing conditions together with symptoms gathered over just five days from the onset of the illness.

This was able to predict which cluster a patient falls into and their risk of requiring hospitalisation and breathing support with a higher likelihood of being correct than an existing risk model based purely on age, sex, BMI and pre-existing conditions alone.

Given that most people who require breathing support come to hospital around 13 days after their first symptoms, this extra eight days represents a significant ‘early warning’ as to who is most likely to need more intensive care.

(SNIP)

The pre-print, non-peer reviewed paper is available online: Carole H Sudre et al. Symptom clusters in Covid19: A potential clinical prediction tool from the COVID Symptom study app (2020) medRxiv doi.org/10.1101/2020.06.12.20129056

*Researchers have now identified skin rash as a key symptom of COVID-19 in up to one in ten cases. However, it was not recognised as a symptom during the time when the data was gathered for this analysis so it is currently unknown how skin rashes map on to these six clusters.

None of these 6 groupings, by itself, can accurately predict COVID-19 severity.  Even among those who fell into group 6 - the highest risk category - only half required hospitalization. 

But when combined with other patient data (age, gender, BMI, comorbidities, etc.), adding in early symptomology appears to substantially improve the ability of models to predict which patients are most likely to need intensive medical care in the weeks ahead.  

 

While perhaps not a game changer, in this fight against the COVID-19 pandemic, we can use all the advantages we can get.