Fig. 2.

Relative feature importance to predict COVID-19 outcome in hospitalized patients. a) boxplot showing the distribution of the top seven estimated feature importance for those features with at least one simulated value higher than zero. b) survival probability distribution for ventilatory support status – ranging from 0 (death) to 1 (recovery) estimated using the XGBoost algorithm. c) relative importance of invasive and non-invasive ventilatory support for predicting COVID-19 mortality. d) relative importance for old (age ≥ 60 years) and young (<60 years) individuals. Dotted line represents the expected identical contribution for those analyzed variables. Values deviating from this line represent a differential contribution.