Table 1.
Variations across data sources and implementation of end‐point definition
| Data set | Description of data source | Population | Derivation of date of death | Censor date |
|---|---|---|---|---|
| A | Structured and unstructured EHR data and commercial obituary data. | Academic and community practice patients in the United States; outpatient; initiated frontline therapy between 2015 and 2018. | An algorithm is used. If dates agree across the three data sources, the date is selected. If discrepancy <7 days exists, EHR data is preferentially captured. If discrepancy >7 days exists, EHR data with accompanying source documentation (e.g., death certificate) is prioritized, otherwise commercial obituary data is captured. | Date of last clinical activity prior to data cutoff, defined as in‐person visit event with healthcare provider such as treatment administration or collected test. |
| B | Structured and unstructured EHR data, commercial obituary data, and Social Security Death Index (SSDI) data. | Academic and community practice patients in the United States; outpatient; initiated frontline therapy between 2015 and 2018. |
An algorithm is used. If all dates agree across the three data sources, the date is selected. If any two dates agree, that date is selected. If all three dates disagree, the following hierarchy is applied: SSDI, obituary, EHR data. If a day level DoD is available in abstracted EHR data, that date is selected over the consensus structured date. Exact date was used, where available. If only month‐level date was available, it was generalized to the end of month. If only year‐level date was available, if was generalized to the end of year. |
Date of last structured activity, defined as the most recent visit prior to data cutoff. |
| C | Structured and unstructured EHR data; hospital‐based, enterprise‐wide, and national cancer registries; commercial obituary data; and digitized obituaries. | Community practice patients in the United States; inpatient and outpatient; initiated frontline therapy between 2015 and 2018. | An algorithm is used. Tumor registry (hospital, enterprise‐wide, national) dates were preferentially selected, followed by structured EHR, followed by commercial obituary data. | Date of last contact (physical encounter, medication order, or medication administration, from structured or unstructured EHR data) prior to the data cutoff. |
| D | Structured EHR data. | Community practice patients in the United States; outpatient; initiated frontline therapy between 2015 and 2018. | Actual date of death documented from EHR or DMF. | Date of last structured clinical activity prior to data cutoff. |
| E | Structured EHR data, structured claims data. | Academic and community practice patients in the United States; primarily outpatient; initiated frontline therapy between 2017 and 2018. | Mortality algorithm incorporating EHR and Claims. | Date of last clinical encounter with healthcare provider prior to data cutoff. |
DMF, death master file; DoD, date of death; EHR, electronic health record.