A comparative study of pre-trained language models for named entity recognition in clinical trial eligibility criteria from multiple corpora

Table 5 The strict performance of the PubMedBERT model for each main entity across the three corpora

Main entities	Covance			Main entities	EliIE			Main entities	Chia
Main entities	P	R	F1	Main entities	P	R	F1	Main entities	P	R	F1
Condition	0.783	0.806	0.795	Condition	0.871	0.892	0.881	Condition	0.742	0.773	0.757
Drug	0.734	0.762	0.748	Drug	0.850	0.881	0.865	Drug	0.747	0.798	0.771
Qualifier_Modifier	0.597	0.599	0.598	Qualifier	0.780	0.814	0.796	Qualifier	0.444	0.486	0.462
Measurement	0.786	0.818	0.801	Measurement	0.863	0.871	0.866	Measurement	0.669	0.689	0.678
Procedure	0.651	0.674	0.662	Procedure_device	0.725	0.765	0.742	Procedure	0.574	0.630	0.600
Observation	0.651	0.679	0.664	Observation	0.754	0.792	0.771	Observation	0.278	0.260	0.267
Temporal_constraint	0.717	0.751	0.733	Temporal_measurement	0.807	0.829	0.815	Temporal	0.552	0.638	0.592
Anatomic_location	0.458	0.407	0.429	Anatomic_location	0.519	0.499	0.507	Negation	0.569	0.626	0.595
Negation_Cue	0.500	0.502	0.501					Device	0.528	0.515	0.520
Event	0.814	0.848	0.830					Multiplier	0.374	0.406	0.388
Permission_Cue	0.578	0.635	0.604					Person	0.795	0.824	0.808
Demographics	0.714	0.743	0.727					Value	0.727	0.745	0.735
Device	0.565	0.567	0.559					Visit	0.504	0.579	0.530
Refractory_condition	0.519	0.586	0.547					Mood	0.302	0.360	0.325
Investigational_product	0.657	0.630	0.641					Reference_point	0.398	0.524	0.453

ISSN: 1472-6947