Building a tobacco user registry by extracting multiple smoking behaviors from clinical notes

Table 2 Summary Statistics from Smoking Status Algorithm Testing

a)	i2b2 trained and tested
	Micro F1: 0.90
	Precision	Recall	F1-Score	N notes	Sensitivity	Specificity
Never	0.94	0.94	0.94	16	94%	94%
Ever	–	–	–	25	94%	99%
Former	0.73	0.73	0.73	11	73%	97%
Current	0.62	0.73	0.67	11	73%	95%
Smoker	0.0	0.0	0.0	3	0%	99%
Unknown	1.00	1.00	1.00	63	100%	100%
b)	Local record trained and tested
	Micro F1: 0.90
	Precision	Recall	F1-Score	N notes	Sensitivity	Specificity
Never	0.83	0.98	0.90	51	98%	94%
Ever	–	–	–	64	90%	96%
Former	0.93	0.83	0.88	30	83%	99%
Current	0.79	0.84	0.81	31	84%	96%
Smoker	0.33	0.33	0.33	3	33%	99%
Unknown	0.99	0.92	0.95	108	92%	99%
c)	Local record trained, i2b2 record tested
	Micro F1: 0.88
	Precision	Recall	F1-Score	N notes	Sensitivity	Specificity
Never	0.75	0.94	0.83	16	94%	94%
Ever	–	–	–	25	80%	94%
Former	0.88	0.64	0.74	11	64%	99%
Current	0.86	0.55	0.67	11	55%	99%
Smoker	0.00	0.00	0.00	3	0%	94%
Unknown	1.00	1.00	1.00	63	100%	100%

Overall F1-score, and by smoking status precision, recall, F1-score, sensitivity, and specificity for a) i2b2 note trained and tested b) local note trained and tested and c) local note trained and i2b2 note tested data sets

ISSN: 1472-6947