Current Trends in COVID-19 Severity: A Machine Learning Approach Based on RecentVariants, Genomic Mutations and Demographic Information

Abstract

Abstract With the recent global surge in COVID-19 cases, understanding the current drivers of severe COVID-19 is critical as novel variants continue to emerge. This study aimed to investigate the impact of current SARS-CoV-2 variants, genomic mutations and patient demography on COVID-19 severity using robust prediction models. The training and testing dataset (n=11,181 and 2796 each) consisting of SARS-CoV-2 genomic information with patient metadata and severity status were obtained from Global Initiative on Sharing all Influenza Data (GISAID) spanning samples deposited from 1 January 2024 to 15 May 2025. Six machine learning algorithms were employed for model construction. SHapeley Additive exPlanation (SHAP) was used for model interpretability and identifying severity related features. GPBoostmodel (including country and region metadata) performed the bestin terms of classification performance metrics and clinical utility (AUROC=0.895, Brier score=0.095) closely followed by GPBoost (mixed effects with country and region as random effect groups) (AUROC=0.892, Brier score=0.092) as evaluated by both statistical and decision curve analysis. In addition to patient age, gender and variant, more than twenty mutations in the 3’ untranslated region, S gene, ORF1ab and M gene were observed to have a significant impact on severity. We audited data quality and representativeness, evaluated domain generalization (leave-one-country-out) and temporal robustness (rolling-origin-splits)and outlined a simple updating policy to remain calibrated under variant turnover. Our model is positioned as a mutation aware screening tool and provides a foundation for prospective clinical integration.