Molecular signatures to discriminate patients based on risk of severe disease and mortality from COVID-19 infection are urgently required by the global medical community. Although non-targeted methods are useful for comprehensive ‘omic coverage, targeted MS-based approaches generally provide higher precision, and improved inter-laboratory reproducibility, allowing for more realistic materialization of true biomarkers via validation studies in independent cohorts. We found a relatively small subset of molecular features that can be used to predict the chances of survival of hospitalized COVID-19 patients within the first day of admission, using a robust LC-MRM setup which is already available in many clinical laboratories.