Table 3.

The task category list by grouping relevant customization tasks

Cat. ID	Customization Task Category	Customization Task Example	Estimated time consumption (median)	Score	Count
Knowledge aspect
C1	Map source vocabulary to target vocabulary (among different terminologies or same terminology different versions)	Vocabulary mapping to get non-ingredient RxCUI from ingredient RxCUI since medication may also be coded using brand drug codes Vocabulary mapping to get ICD10 procedure codes from provided ICD9 procedure codes considering the data contains ICD10 coded data	<1day	K 0	185
C2	Map free text to target vocabulary	Vocabulary mapping to get RxCUI from provided medication names	<1week	K 1	88
C3	Define an operational definition of a specific EHR data element	Find “provider specialty” that links to procedure to check eye exam from ophthalmology department	<1week	K 1	138
C4	Define an operational definition of a non-EHR data element event	Define “Continuous enrollment/contact” for implementation	<1day - <1week	K 2	20
C5	Locate the data source for a group of data	Identify where to find “carotid imaging study”	<1day	K 1	89
C6	Retrieve data attribute representation and contextual knowledge through exploring structured data	Find lab unit, categorical range for urine protein tests	<1hour	K 1	61
C7	Retrieve data attribute representation and contextual knowledge through exploring unstructured data	Explore radiology reports to validate the local use of the “intravenous contrast” keywords provided and their occurrence prevalence	<1week	K 2	18
C8	Acquire knowledge of unstructured clinical data from domain expert and through programming	Find relevant “note types” and “service groups” which clinical notes may contain PAD information	<1week and <1month	K 2	14
Interpretation aspect
C9	Understanding phenotype algorithm pseudocode clause	Understand if “ever” from “Taking ARBs (angiotensin receptor blockers) ever” means both structured and unstructured medication lists should be used	<1day	I 0–2	46
Programming aspect
C10	Compile machine readable input file	Compile ICD codes or code groups provided in a pdf pseudocode appendix to a machine-readable file	<1day	P 0	258
C11	Pre-processing data by simple programming	Find if relevant pathology reports exist from clinical data warehouse programmatically	<1week	P 1	11
C12	Search keywords from unstructured data	Find at least 2 unique DSM-IV social interaction terms from notes	<1week	P 1	56
C13	Search keywords with modifier from unstructured data	Identify non-negated diverticulosis terms (e.g., diverticulitis, diverticula) from relevant radiology reports	<1week	P 2	16
C14	Extract information from unstructured data using advanced NLP implementation	Extract heart rate from ECG report	<1month	P 2	16
C15	Extract information from NLP tool processed documents	Search heart disease concepts (UMLS CUIs) from MedLEE parsed ECG report	<1day	P 1	11
C16	Configure, install and execute NLP tools	Install cTAKES for parsing clinical notes	<1week	P 2	23
C17	Populate NLP search results for SQL query	To exclude patient with cancer using ICD-9 and keywords, keywords search results from unstructured data need to be imported to the database	<1week	P 1	37
C18	Implement complex SQL query	Implement extrapolating height at serum creatinine measurement time from its pre- and post- height measurement based on a formula	<1day and <1week	P 1	41
Other (not specific to the phenotype)
C19	Check the availability and completeness of the needed data element	Potentially unavailable information of medication administration route, which is required for glaucoma phenotyping	<1week		22
C20	Implement another existing phenotype	Use existing eMERGE T2DM algorithm to check if a patient has type 2 diabetes	<1week		3
Total (1153)