Table 2.

Traditional and novel (italicized) data sources that can be used to develop artificial intelligence algorithms to improve patient safety; selected examples.

Patient safety domain	EHR	Claims	Risk scores	Genome sequencing	Sensors	Computer vision	Other
Healthcare-associated infections	X	X		Of pathogens to monitor spread of nosocomial infections	Continuous vitals, Chemical vapor sensors	Of health care setting	Microbiology of random specimens, Smart sinks and dispensers
Adverse drug events	X	X	BADRI model, Trivalle’s risk score	Of patients for Cytochrome P450 polymorphisms	Continuous vitals, Continuous glucose monitoring		Patient report
Venous thromboembolism	X	X	Padua prediction score, Khorana score	Of patients for Cytochrome P450 polymorphisms, Factor V Leiden, Prothrombin 20210 mutations	Activity, Pressure, Location		Novel biochemical analytes
Surgical complications	X	X	Surgical risk scores		Continuous vitals	Of operating room
Pressure ulcers	X	X	Braden score		Activity, Pressure, Location	Of bed	Novel biochemical analytes
Falls	X	X	Hendrich model		Activity, Pressure, Location	Of common spaces
Decompensation	X	X	MEWS, CART score		Continuous vitals, Activity, Continuous glucose monitoring		Novel biochemical analytes
Diagnostic errors	X	X			Chemical vapor sensors		Clinician adjudication, Patient report

Novel data sources italicized.

BADRI Brighton Adverse Drug Reactions Risk, CART Cardiac Arrest Risk Triage, EHR electronic health record, MEW, Modified Early Warning Score.