Table 2.

Guidelines for prespecification and reporting of data creation.

		Item number	Recommendation
Data harmonization
	Target RCTa study design	1	Prespecification: Indicate the source of the study design document (protocol, reporting paper, or others). Reporting: Describe the sections and tables from which the relevant variables are recognized.
	RCT variable list	2	Prespecification: Specify the method for recognizing variables from the RCT document and matching to relevant EHRb features. Reporting: Define all end points, interventions, eligibility criteria, and other baseline characteristics recognized from the RCT document along with the matched EHR features.
Cohort construction
	Data mart	3	Prespecification: Specify the method for compiling the broad list of EHR features indicating the condition or disease of interest. Reporting: List the EHR features, and state the algorithm to define inclusion in the data mart. Reporting: Report the size of the data mart.
	Disease cohort	4	Prespecification: Specify the method for ascertaining the phenotype of the condition or disease of interest. Reporting: Describe the input EHR features of the phenotyping algorithm. Reporting: State the phenotyping algorithm with chosen parameters. Reporting: Report the AUCc of prediction and the accuracy of the disease cohort.
	Treatment arms	5	Reporting: Explain how treatment initiation time is determined. Reporting: Explain how treatment arms are defined with the list of involved EHR features, time windows, and the algorithm.
Variable curation
	End points	6	Prespecification: Specify the method for ascertaining the end points. Reporting: State the end point algorithm with chosen parameters. Reporting: Explain how the end point is defined.
	Baseline characteristics (eligibility criteria and confounders)	7	Prespecification: Specify the variable curation plans for each class of baseline characteristics. Reporting: List the baseline characteristics considered in the RWEd and define how they are created with input EHR features, time windows, groupings, and other transformation. Reporting: Explain how eligibility criteria will be matched according to the curated baseline characteristics. Reporting: Present the summary statistics of the baseline characteristics in treatment arms filtered by eligibility criteria.
	Additional confounders	8	Prespecification: List the other confounders considered in the RWE and define how they are created with input EHR features, time windows, groupings, and other transformation.
	Missing data	9	Reporting: Describe how missing information on variables are handled.
Validation
	Sampling strategy	10	Prespecification: Specify the sampling strategy for the validation set. Reporting: Report the sizes and list of variables (in data mart, disease cohort, and arms) of the validation.
	Data accuracy	11	Reporting: Report the agreement between gold-standard data from validation and curated data. Reporting: Explain how inaccurate data are dealt with.
Publication		12	Reporting: Export the final curation models for the condition or disease of interest, end points, and other variables curated through machine learning methods.

^aRCT: randomized controlled trial.

^bEHR: electronic health record.

^cAUC: area under the curve.

^dRWE: real-world evidence.