Table 1.

Study variables and EHR data source.

EHR Source Information Type	Variables Needed for Analysis	Curation Approaches
Structured data (e.g., date of birth)	Diagnoses (i.e., ICD codes) Gender Birth year Race Ethnicity Practice type ECOG performance status Medication order date Medication administration date Visit date Mortality date a	Transformation, harmonization, and deduplication
Unstructured data (e.g., clinic notes, PDF lab reports, radiology images, etc.)	NSCLC diagnosis NSCLC diagnosis date Advanced NSCLC diagnosis Advanced NSCLC diagnosis date ROS1 test result ROS1 test date ALK test result BRAF test result EGFR test result ALK test date BRAF test date EGFR test date PD-L1 percent staining PD-L1 test result date Group stage Histology Line of therapy b Line of therapy start date b	Expert abstraction OR ML-extraction

Abbreviations: ALK: anaplastic lymphoma kinase; EGFR: epidermal growth factor receptor; ECOG: Eastern Cooperative Oncology Group; ML: machine learning; NSCLC: non-small cell lung cancer; PD-L1: programmed death-ligand 1; ^a mortality date is a composite variable based on multiple data sources (structured and unstructured EHR data, commercial sources, and Social Security Death Index) [16]. ML extraction was not used to define this variable. ^b Line of therapy and line of therapy date are a derived variable based on both structured and unstructured data inputs.