Identification of critical SARS-CoV-2 amino acids associated with COVID-19 hospitalization rate using machine learning and statistical modeling: An observational study in the United States

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

“…A few of these have been associated with disease severity in prior studies. Sokhansanj et al identified mutations at S_G142 as contributing to prediction of disease severity in a deep learning model using S protein features (8), and Liang et al identified both S_T95 and S_L212 as associated with COVID-19 disease severity in a study of associations between spike amino acid mutation prevalence in GISAID data and concurrent hospitalization rates in CDC data (7). Obermeyer et al also found that the amino acid substitution S T95I was associated with increased SARS-CoV-2 transmission (22), and other studies point to associations of S_T95 with increased transmission (47).…”

“…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).…”

“…in a study of associations between spike amino acid mutation prevalence in GISAID data and concurrent hospitalization rates in CDC data (7). Obermeyer et al also found that the amino acid substitution S T95I was associated with increased SARS-CoV-2 transmission (22), and other studies point to associations of S_T95 with increased transmission (47).…”

“…Multiple studies have investigated associations between SARS-CoV-2 genomic variation and COVID severity in different times and places during the pandemic and collectively have identified a diverse set of candidate genetic variants (2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)). These studies have either used data from GISAID (2, 4-6, 9, 11, 16-18) or from individual hospitals (3,10,(12)(13)(14)(15)).…”

“…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).…”