Risk factors for adverse outcomes during mechanical ventilation of 1152 COVID-19 patients: a multicenter machine learning study with highly granular data from the Dutch Data Warehouse

Intensive Care Medicine Experimental

Table 2 Model performance for the different outcomes and day of IMV

	Overall	Day 1	Day 7	Day 14
ICU mortality (AUROC ± 95% confidence interval)
Decision tree	0.695 ± 0.027	0.668 ± 0.042	0.718 ± 0.013	0.739 ± 0.051
Logistic regression	0.744 ± 0.023	0.710 ± 0.035	0.766 ± 0.024	0.782 ± 0.028
XGBoost	0.774 ± 0.023	0.732 ± 0.04	0.806 ± 0.025	0.817 ± 0.013
ICU-free days (R2 ± 95% confidence interval)
Lasso	0.118 ± 0.009	0.086 ± 0.024	0.147 ± 0.016	0.067 ± 0.100
Ridge	0.179 ± 0.050	0.140 ± 0.065	0.196 ± 0.071	0.229 ± 0.081
XGBoost	0.212 ± 0.028	0.148 ± 0.029	0.267 ± 0.090	0.263 ± 0.077
Ventilator-free days (R2 ± 95% confidence interval)
Lasso	0.169 ± 0.015	0.112 ± 0.012	0.209 ± 0.050	0.231 ± 0.024
Ridge	0.217 ± 0.038	0.147 ± 0.018	0.263 ± 0.108	0.303 ± 0.039
XGBoost	0.250 ± 0.033	0.160 ± 0.019	0.319 ± 0.080	0.352 ± 0.038

Model performance is shown for ICU mortality, ventilator-free days at day 30, and ICU-free days at day 30 across the days of IMV
AUROC area under the receiver operating characteristic