Venous thromboembolic disease in admitted blunt trauma patients: what matters?

Thrombosis Journal

Table 3 Performance of statistical models in predicting in-hospital VTE events, as measured on a sample testing dataset (20% of patients)

	Logistic	Random forest	GBM	Logistic Synthetic	Random Forest synthetic	GBM synthetic
TPR	0.64	0	0.48	0.64	0.45	0.27
TNR	0.83	1	0.88	0.83	0.93	0.98
FPR	0.95	0	0.94	0.90	0.05	0.88
FNR	0.01	0.02	0.01	0.01	0.08	0.01
Accuracy	0.82	0.98	0.87	0.83	0.92	0.97
Balanced Accuracy	0.80	0.50	0.68	0.74	0.69	0.62
F1 score	0.10	n/a	0.10	0.10	0.16	0.16
AUCROC	0.82	0.78	0.82	0.86	0.84	0.89

Balanced accuracy is defined as (sensitivity + specificity)/2
The F1 score is 2*True Positive/(True Positive + True Positive + False Positive + False Negative)
TPR – true positive rate (i.e. sensitivity), TNR – true negative rate (i.e. specificity), FPR – false positive rate, FNR – false negative rate

ISSN: 1477-9560