Self-reported symptoms and their relation to COVID-19 infection and its severity: A Swedish pilot study

Abstract

On March 11th, 2020, the World Health Organization officially recognized COVID-19 as a global pandemic. Health systems worldwide were overwhelmed, in part due to the variety of new variants that began to emerge, presenting diverse symptom profiles and levels of infectiousness. This study aimed to develop a predictive scale based on Rasch analysis, using symptoms reported by individuals with positive PCR tests. Additionally, the project sought to evaluate the scale's effectiveness in screening for positive PCR tests among those tested and in predicting hospitalization among those who tested positive. Data were obtained from the COVID Symptom Study smartphone application in the United Kingdom, United States of America, and Sweden, with a focus on Swedish data from the COVID Symptom Study Sweden. Early symptoms, within the first 5 days, were of particular interest for early prediction of COVID-19 infection severity. A Rasch analysis was used to investigate whether the symptoms could form a measurement scale related to COVID-19 infection. This could indicate the importance of the symptoms in regard to severity of the infection and regarding predictability. A low location of the symptoms on the the scale indicates that they are common and could indicate low severity. A high scale location would then indicate more rare symptoms that could be connected to more severe infection. Logistic regression was used to predict positive PCR tests among individuals who underwent PCR testing, as well as hospitalization among those with positive PCR tests. The scale's fit to the Rasch model was moderate, its predictive ability for hospitalization among individuals with positive PCR tests was acceptable, as indicated by an area under the curve (AUC) of 0.7 (Mandrekar, 2010), but maybe not clinically useful (Fan et al., 2006). However, the representation of COVID-19 cases in the Swedish data during the first half of 2020 was limited to more severe cases, primarily reflecting individuals with severe symptoms. 5 (38) In conclusion, symptom clustering holds promise in understanding patterns in COVID-19 symptoms and could serve as a valuable screening tool for identifying severe cases. Further research, particularly focusing on predictive models and comparative analyses, is necessary to fully understand these symptom patterns and their practical applications. Our findings indicate that predicting COVID-19 severity is feasible, making continued research in this field imperative.