Does adding claims to EMR data improve comorbidity burden capture? A Concerto Health AI Definitive Oncology Dataset and Symphony Linkage Study.

Abstract

e18186 Background: Current commercially available oncology electronic medical record (EMR) databases typically have limited data on patient comorbidities which may limit the ability to accurately include/exclude these patients or adjust for these factors in health outcomes research. The objective of this study was to investigate the impact of identifying comorbidities using both EMR and claims data. Methods: The Concerto Health AI Definitive Oncology Dataset is a deeply curated cohort of patients (pts) derived from a wide range of oncology practices throughout the United States. Concerto Health AI has a co-exclusive partnership with ASCO CancerLinQ and collects data from multiple other sources. These data were deterministically linked to US Symphony Health’s (SH) Integrated Dataverse, an anonymized, HIPAA-compliant data set. Advanced (stage IIIb/IV) non-small cell lung cancer (aNSCLC) pts with data curated via manual review by nurse practitioners and diagnosed after 2011 were included. Comorbidities within 1 year prior to advanced diagnosis were categorized using the Charlson Comorbidity Index (CCI). The impact of the claims linkage on inclusion/exclusion (I/E) criteria was assessed by applying criteria from the IMpower150 clinical study and exploring changes to survival estimates, as calculated using Kaplan-Meier method. Results: A total of 11,373 aNSCLC pts were identified, of which 7,887 pts were linked to SH. Using only Concerto, 278 pts (3.5%) had a CCI score ≥1, whereas 3,033 pts (38.4%) had a CCI score ≥1 using linked Concerto and SH. After applying I/E criteria from IMpower150 using available data elements from Concerto only, 88 pts were identified matching the trial criteria, with a median survival of 15.8 mths (95% CI: 14.4, 21.0). When leveraging Concerto and SH, 8 (9.1%) more patients were excluded (n = 80) based on comorbidities (6 pts) and oral medications (2 pts), with a median survival of 16.6 mths (95% CI: 14.6, 23.4). Conclusions: The addition of claims to EMR enhanced the ability to identify patient comorbidities given the longer history and broader physician specialty of claims data. Future real-world studies should explore linkages across datasets to better capture comorbidities for patient selection and adjustment.