Table 1:

Purpose of NLP/ML applications for analyzing PRO data among the 79 studies included in the systematic review^*

Classifications	Specific roles of NLP/ML tasks	Task description	N	%
Information extraction/text identification			74	93.7
	PRO content detection, identification, extraction	Detect or identify PRO keywords or terminologies from free text	47	59.5
	PRO annotation	Perform semi-automated or manual annotation for PROs in free text	37	46.8
	PRO affirmation/ negation	Declare whether symptoms or symptom-related outcomes exist or equivalent expression or negative statement for having symptoms	38	48.1
	Vocabulary mapping	Map or assign PROs or PRO-related vocabulary words to appropriate indexes or labels	13	16.5
Classification/phenotyping/clustering			26	32.9
	PRO classification	Assign or classify extracted PROs into specific categories	16	20.3
	PRO phenotyping	Indicate specific characteristics of single or multiple PROs features	9	11.4
	PRO clustering	Identify two or more PROs that are related to each other or co-occur	5	6.3
Develop or validate NLP/ML pipelines			19	24.1
	Development of NLP/ML pipelines	Develop new NLP/ML pipelines or build NLP software	10	12.7
	Evaluation/validation	Evaluate and validate the performances of NLP system/pipeline	12	15.2
Risk prediction or stratification for clinical outcomes			25	31.6
	Risk prediction	Predict the risk of outcomes using extracted PROs based on unstructured narratives	22	27.8
	Risk stratification	Identify the right level of care and services for distinctive subgroups of patients.	3	3.8
Investigate associations between PROs and clinical outcomes			5	6.3
	Relationship detection	Detect semantic associations or relationships between unstructured PROs	5	6.3

^*Each study may include multiple study purposes and NLP/ML tasks.