Clinical data bases typically include, for each patient, many heterogeneous features, for example blood exams, the clinical history before the onset of the disease, the evolution of the symptoms, the results of imaging exams, and many others. Using subsets of these features, one can measure the similarity between two patients in several different manners. We here propose to exploit a recently developed statistical approach, the information imbalance, to compare these different similarity measures, and quantify their relative information content. We apply this approach to a data set of ~ 1,300 COVID-19 patients in Udine hospital before October 2021. Using this approach we find (asymmetric) relationships between single features and systematically compare subsets of up to 20 different features as COVID-19 severity predictors. The identified features can be measured at the moment of the admission of the patient and, if used in combination, are maximally informative of the clinical fate and of the severity of the disease. The approach can be used also if the features are available only for a fraction of the patients and, importantly, is able to select automatically features with small inter-feature correlation.