Table 4.

AUC of signal detection performance for FAERS, healthcare data, and combined systems

ADR	Experiment 1. Combining FAERS and NYP/CUMC EHR
	FAERS	EHR	Combined
Acute renal failure	0.89	0.61	0.89
Acute liver injury	0.70	0.45	0.68
Acute myocardial infarction	0.65	0.53	0.70
Upper GI bleeding	0.83	0.54	0.83
Total	0.75	0.51	0.74

ADR	Experiment 2. Combining FAERS and GE EHR
	FAERS	GE	Combined
Acute renal failure	0.91	0.68	0.92
Acute liver injury	0.71	0.63	0.76*
Acute myocardial infarction	0.72	0.80	0.82
Upper GI bleeding	0.80	0.77	0.87*
Total	0.76	0.76	0.82*

ADR	Experiment 3. Combining FAERS and the claims data
	FAERS	Claims	Combined
Acute renal failure	0.91	0.83	0.93
Acute liver injury	0.72	0.69	0.79*
Acute myocardial infarction	0.71	0.77	0.82*
Upper GI bleeding	0.81	0.83	0.86
Total	0.76	0.78	0.82*

The bold values indicate the highest AUC performance but not necessarily significantly higher than comparators except for those also marked with * where the performance difference is statistically significant

ADR adverse drug reaction, AUC area under the receiver operator characteristics curve, EHR Electronic health record, FAERS FDA Adverse Event Reporting System, GE EHR GE Healthcare MQIC (Medical Quality Improvement Consortium) database, NYP/CUMC New York Presbyterian Hospital at Columbia University Medical Center