Table 2.

List of 15 studies included in the review and their characteristics.

Author (year)	Study setting (country)	Dataset sizea	Labels included	Aim	Triage phase
Anthony et al (2021) [26]	South Africa	93	3 binary	Classify critical conditions in emergency calls	Dispatch
Ceklic et al (2022) [27]	Australia	11,971	1 binary	Classify severity of traffic crash incidents in dispatch calls	Dispatch
Chin et al (2022) [28]	Taiwan	114	1 binary	Classify severity of traffic crash incidents in dispatch calls	Dispatch
Cotte et al (2022) [29]	Germany	385	1 multiclass	Classify triage decisions using a symptom assessment app	Chatbot
Ferri et al (2021) [30]	Spain	1,244,624	3 binary and multiclass	Classify emergency incidents in dispatch calls	Dispatch
Gatto et al (2022) [31]	United States	574	1 binary	Classify severity in patient’s text-based inquiries	Chatbot
Inokuchi et al (2022) [32]	Japan	15,442	1 binary	Identify undertriage in prehospital telephone triage	Nurse-led phone line
Lai et al (2020) [33]	United States	—b	—	Classify triage for prehospital COVID-19 cases	Chatbot
Marchiori et al (2021) [25]	Switzerland	>900,000c	1 multiclass	Evaluate AId-powered chatbot for symptom-checker triage	Chatbot
Morse et al (2020) [34]	United States	26,646	—	Evaluate AI-powered chatbot for symptom-checker triage	Chatbot
Pacula et al (2014) [35]	United States	427	2 multiclass	Classify triage and distress indicators in crisis hotline chats	Crisis hotline
Spangler et al (2019) [3]	Sweden	68,668	1 continuous	Validate MLe-generated risk scores for prehospital care	Dispatch (operated by nurses)
Tollinton et al (2020) [5]	United Kingdom	1,188,509	1 binary	Classify triage of unconscious patients in dispatch calls	Dispatch
Veladas et al (2021) [36]	Portugal	269,669	1 multiclass	Classify clinical pathways from text data	Nurse-led phone line
Yunoki et al (2014) [4]	Japan	61,927	1 multiclass	Classify triage categories from phone call data	Dispatch

^aNumber of patient records included.

^bDataset size used for model development was not stated for this study or information on labels was not included.

^cThe study stated that “more than 900,000 case records” were included.

^dAI: artificial intelligence.

^eML: machine learning.