Identification of serum prognostic biomarkers of severe COVID-19 using a quantitative proteomic approach