Table 10.

Metrics for the training and testing (external) cohorts for BERT^a with up-sampling of the minority class.

Training data size	AUCb score	Sensitivity	Specificity	Precision	F1-score
10,000	0.79	0.81	0.58	0.30	0.43
100,000	0.84	0.81	0.68	0.34	0.48
1,000,000	0.85	0.75	0.78	0.41	0.54

^aBERT: Bidirectional Encoder Representations from Transformers.

^bAUC: area under the receiver operating characteristic curve.