Machine Learning for COVID-19 Patient Management: Predictive Analytics and Decision Support

BackgroundThe global impact of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has profoundly affected economies and healthcare systems around the world, including Lebanon. While numerous meta-analyses have explored the systemic manifestations of COVID-19, few have linked them to patient history. Our study aims to fill this gap by using cluster analysis to identify distinct clinical patterns among patients, which could aid prognosis and guide tailored treatments. MethodsWe conducted a retrospective cohort study at Beiruts largest teaching hospital on 556 patients with SARS-CoV-2. We performed cluster analyses using K-prototypes, KAMILA and LCM algorithms based on 26 variables, including laboratory results, demographics and imaging findings. Silhouette scores, concordance index and signature variables helped determine the optimal number of clusters. Subsequent comparisons and regression analyses assessed survival rates and treatment efficacy according to clusters. ResultsOur analysis revealed three distinct clusters: "resilient recoverees" with varying disease severity and low mortality rates, "vulnerable veterans" with severe to critical disease and high mortality rates, and "paradoxical patients" with a late presentation but eventual recovery. ConclusionsThese clusters offer insights for prognosis and treatment selection. Future studies should include vaccination data and various COVID-19 strains for a comprehensive understanding of the diseases dynamics.