Table 5.

Model performance with augmented training data.

Model	Data split strategy	Accuracy (%)
		Entity replacement	Back translation	Entity replacement+back translation
BERTa	1	79.33	77.33	79.33
BERT	2	79.33	85.33	77.33
Clinical BERT	1	88.66	86.66	82.66
Clinical BERT	2	84	89.33	88

^aBERT: Bidirectional Encoder Representations from Transformers.