Predictive model for severe COVID-19 using SARS-CoV-2 whole-genome sequencing and electronic health record data, March 2020-May 2021

Abstract

Objective We used SARS-CoV-2 whole-genome sequencing (WGS) and electronic health record (EHR) data to investigate the associations between viral genomes and clinical characteristics and severe outcomes among hospitalized COVID-19 patients. Methods We conducted a case-control study of severe COVID-19 infection among patients hospitalized at a large academic referral hospital between March 2020 and May 2021. SARS-CoV-2 WGS was performed, and demographic and clinical characteristics were obtained from the EHR. Severe COVID-19 (case patients) was defined as having one or more of the following: requirement for supplemental oxygen, mechanical ventilation, or death during hospital admission. Controls were hospitalized patients diagnosed with COVID-19 who did not meet the criteria for severe infection. We constructed predictive models incorporating clinical and demographic variables as well as WGS data including lineage, clade, and SARS-CoV-2 SNP/GWAS data for severe COVID-19 using multiple logistic regression. Results Of 1,802 hospitalized SARS-CoV-2-positive patients, we performed WGS on samples collected from 590 patients, of whom 396 were case patients and 194 were controls. Age (p = 0.001), BMI (p = 0.032), test positive time period (p = 0.001), Charlson comorbidity index (p = 0.001), history of chronic heart failure (p = 0.003), atrial fibrillation (p = 0.002), or diabetes (p = 0.007) were significantly associated with case-control status. SARS-CoV-2 WGS data did not appreciably change the results of the above risk factor analysis, though infection with clade 20A was associated with a higher risk of severe disease, after adjusting for confounder variables (p = 0.024, OR = 3.25; 95%CI: 1.31–8.06). Conclusions Among people hospitalized with COVID-19, older age, higher BMI, earlier test positive period, history of chronic heart failure, atrial fibrillation, or diabetes, and infection with clade 20A SARS-CoV-2 strains can predict severe COVID-19.

SARS-CoV-2 Genetic Variants and Patient Factors Associated with Hospitalization Risk

“…The GLMM analysis identified several candidates, including one in the Alpha lineage, ORF1b_P1975 (in nsp14 at position 451). The most common genetic mutation for ORF1b_P1975, C19390T, was also found to be weakly associated with severe COVID-19 in a prior GWAS study (14). While we are not aware of papers about this mutation’s functional effects, nsp14 has been found to shut down host protein synthesis and thereby interfere with host innate immune responses (49).…”

“…One possible reason for differences in associations is that our study used a different measure of disease severity and a different study population–hospitalization in persons with co-morbidities. In contrast, many of the prior studies have used symptom information for severity (6, 16), evaluated differences in symptoms among patients that were already hospitalized (13, 14), or included persons without co-morbidities (3, 7).…”

“…Single nucleotide variants are often tightly correlated with each other, and these correlations need to be accounted for to disambiguate associations. Some studies have used methods to analyze one variant at a time, including chi-square or Fisher exact tests that do not account for genetic correlations (2, 3), multiple logistic regression with covariates to capture genetic background like in many human GWAS (4, 14, 15), and phylogeny-based association tests (6). Other studies have used analytical approaches to investigate multiple variant loci at a time, including multiple logistic regression with multiple variants as features (5, 17), neural network approaches (7, 8), random forests (16, 17), and XGBoost (7, 8).…”

The utility of whole-genome sequencing to inform epidemiologic investigations of SARS-CoV-2 clusters in acute-care hospitals

“…Reactive WGS is performed to support epidemiologic cluster investigations at the Microbial Genomics Epidemiology Laboratory (MiGEL) at the University of Pittsburgh. 22 , 23 Reactive WGS was available as a resource for COVID-19 cluster investigations within UPMC facilities through a structured request process (Supplementary Fig. online).…”