Risk tools for predicting long-term sequelae based on symptom profiles after known and undetected SARS-CoV-2 infections in the population.

Abstract

The aim was to determine the profile of long-term symptoms after known and undetected SARS-CoV-2 infections and to generate tools for risk and diagnostic assessment of Post-COVID syndrome (PCS). In the population-based Gutenberg COVID-19 Study (N = 10,250), sequential, systematic screening for SARS-CoV-2 was performed in 2020/2021. Individuals received a standardized interview on newly occurred or worsened symptoms since the infection or the pandemic. Robust Poisson regression models were fit to compare the frequency of symptoms between groups. Two scores were developed using machine learning techniques and prospectively validated in an independent cohort. Among n = 942 individuals, prevalence of long-term symptoms was 36.4% among individuals with known SARS-CoV-2 infection, 25.0% in those unknowingly infected, and 28.1% among the controls. Individuals with known infection more often reported smell (Prevalence ratio [PR] = 13.66 [95% confidence interval 4.99;37.41]) and taste disturbances (PR = 5.57 [2.62;11.81]), forgetfulness (PR = 2.88 [1.55;5.35]), concentration difficulties (PR = 2.83 [1.55;5.16], trouble with balance (PR = 2.74 [1.18;6.35]), and dyspnea (PR = 2.22 [1.18;4.19]) than controls. The risk score for predicting long-term sequelae based on symptoms during the acute infection had a cross-validated AUC of 0.74 and 0.72 when applied in an independent cohort (N = 6,570). The diagnostic score providing a probability of the presence of PCS had a cross-validated AUC of 0.66 and of 0.64 in the validation cohort (N = 3,176). Individuals with and without SARS-COV-2 infection reported persistent symptoms, but symptoms attributable to PCS were identified. The data-driven scores may help guide further diagnostic decisions in the initial management of PCS.