Abstract
INTRODUCTION: Methods for assessing change in tumor burden differ between clinical trial and routine clinical care settings, presenting a unique opportunity to design novel methods to capture clinical outcomes from electronic health record (EHR) data. We adapted a previously established approach for solid tumors and modified it to capture real-world progression (rwP) and real-world insufficient response (rwIR) events in non-Hodgkin lymphoma (NHL). METHODS: This study used a nationwide EHR-derived deidentified database. The first phase assessed the rwP/rwIR approach in patients with follicular lymphoma (FL) and diffuse large B cell lymphoma (DLBCL). The second phase assessed the approach in a larger cohort of patients with mantle cell lymphoma (MCL). Performance was assessed through inter-abstractor agreement on event occurrence, date, and type, clinician assessment and source data completeness, downstream events after an rwP/rwIR event, and time-to-event analyses (real-world progression-free [rwPFS] and event-free [rwEFS] survival) and their correlation with real-world overall survival (rwOS). RESULTS: A total of 6162 patients with NHL were included in the study, comprising 672 patients with FL (median age, 64 years; female, 49%; male, 51%), 405 patients with DLBCL (median age, 70 years; female, 42%; male, 58%), and 5085 patients with MCL (median age, 69 years; female, 27%; male, 73%). Inter-abstractor agreement among all cohorts was 96-97% for event occurrence and 85-89% for event date within 30 days. The proportion of patients with an rwP/rwIR event was 26% in the FL cohort, 27% in the DLBCL cohort, and 42% in the MCL cohort. Clinically relevant downstream events were observed in 58% of the FL cohort, 63% of the DLBCL cohort, and 73% of the MCL cohort. In the MCL cohort, median rwPFS was 34.5 months, and median rwEFS was 32.2 months. Real-world OS correlated more strongly with rwPFS (85%) than with rwEFS (80%). CONCLUSIONS: The NHL-specific rwP/rwIR approach is feasible, reliable, and scalable. Observed inter-abstractor agreement and rwP/rwIR event frequency show applicability across NHL cohorts. Endpoint analyses and correlations with rwOS in a real-world population demonstrate the clinical relevance of this approach.