Real-world patient pathway of care for diffuse large B-cell lymphoma (DLBCL).

Abstract

272 Background: Use of immunophenotyping and molecular diagnostics is the most important first step in the care process for DLBCL patients according to recommendations by the NCCN and the WHO. Process mining is a novel approach of statistical and graphical analytics and may provide insights on real-world patient pathway of care and guidelines-recommended diagnostic testing for appropriate DLBCL classification in clinical settings. Methods: This retrospective study included patients diagnosed with DLBCL between 01/01/2011 and 12/31/2019, using the Flatiron Health electronic health record-derived de-identified database. The database includes information on the date of diagnostic testing and corresponding results abstracted from pathology reports or clinical visit notes. This study included information on diagnostic testing by immunohistochemistry (IHC) and molecular diagnostics through fluorescence in situ hybridization (FISH) or karyotype for markers with a confirmed known result that can be used to classify cell of origin according to Hans algorithm and to identify double-/triple-hit lymphoma and double expressor lymphoma. We explored the application of process mining analytics to produce patient pathway graphs for visual investigation on the real-world care process for DLBCL patients, from initial diagnosis through diagnostic testing and line of therapy, until death or end of the study follow-up. Results: A total of 5387 patients (female 45%, mean age at diagnosis 66.4±13.6 years) were included. During a median follow-up of 19.2 months (IQR: 6.1-43.8), 4400 (82%) patients had evidence for IHC testing, 3205 (59%) had evidence for molecular testing, and few had information for next-generation sequencing (n = 91). Among those with evidence of diagnostic testing, 3721 (85%) and 1613 (50%) had the first test performed on the day (specimen collection date) of DLBCL diagnosis by IHC and molecular testing, respectively. During the study follow-up, most patients (n = 5005, 93%) started treatment within a median of 24 days (IQR: 14-36) and few participated in clinical trials (n = 131). When we stratified analysis by year of DLBCL diagnosis, patients with evidence of diagnostic testing increased from 67% (249/372) in 2011 to 86% (506/591) in 2019 for IHC and 50% (184/372) in 2011 to 67% (396/591) in 2019 for molecular testing, respectively. Conclusions: In DLBCL, use of recommended diagnostic testing appeared to increase from 2011 to 2019. Clinical insights generated using process mining could be used by institutions for benchmarking and for care pathway optimization.