OBJECTIVES: In oncology populations, treatment and line of therapy information is crucial in understanding the patient journey and is necessary to understand exposure and subsequent outcomes. To better understand potential gaps in EHR data, this study utilized an administrative health claims linked EHR dataset in an aNSCLC population to identify antineoplastic exposure in claims prior to the first recorded EHR documented exposure.

METHODS: Patients diagnosed with aNSCLC between 2013 and 2022 who were treated with a common rule-based National Comprehensive Cancer Network designated first line of therapy (1L) were selected from the linkage of the nationwide de-identified EHR-derived Flatiron Health Research Database and Komodo Health's Healthcare Map of de-identified patient-level claims data. Antineoplastics were identified in claims between 14 days prior to advanced diagnosis through the day before EHR-based 1L start to identify potential missingness in EHR-derived 1L treatment data.

RESULTS: A total of 5,863 patients were included, of whom 876 (14.9%) were treated with 1L oral tyrosine kinase inhibitors, and 4,987 (85.1%) were treated with 1L intravenous (IV) regimens consisting of platinum-based chemotherapy, immunotherapy and/or anti-VEGF agents. Overall, 221 patients (3.8%) had evidence of claims-derived antineoplastic therapy that was not observed in EHR-derived 1L during the window of interest. Of those 221 patients, the majority of the observed antineoplastic components did not have a labeled indication for aNSCLC; common agents observed included imatinib (n=56), hormonal therapies (n=33), and methotrexate (n=27).

CONCLUSIONS: A low proportion of claims-derived antineoplastic drug exposures were observed prior to the EHR-based 1L start date that were not present in the EHR. Of those observed many were antineoplastics typically used for the treatment of other malignancies, indicating a low likelihood of missingness for the selected 1L regimens investigated in this dataset. These findings can inform exposure misclassification quantitative bias analysis.