Addition of Open Administrative Claims Significantly Improves Capture of Mortality in Electronic Health Record (EHR) – Focused Real-World Data (RWD)

OBJECTIVES: Overall survival (OS) is the gold-standard endpoint in observational studies utilizing real-world data (RWD). However, OS relies on complete and accurate capture of mortality information for the population in question. Previous work has shown that supplementing EHR-focused RWD with publicly and commercially available mortality data improves completeness of the composite mortality endpoint, but no studies have evaluated open administrative claims to increase the capture of mortality.

METHODS: We generated an All-Source Composite Mortality Endpoint (ASCME) drawing on EHR structured data, unstructured EHR documents, Social Security Administration Limited-Access Death Master File, commercially available obituary and internment data, and open claims. We used National Death Index (NDI) data to validate ASCME in data sourced for ConcertAI Patient360^TM, a large, representative de-identified oncology database of human-curated comprehensive RWD, sourced from academic and community EHRs and linked to open claims. Patients were selected randomly from early, advanced and metastatic disease groups in Patient360^TM, with 33,000 patients sampled from 10 solid cancers (bladder, breast, gastroesophageal, hepatocellular carcinoma, melanoma, NSCLC, pancreatic, prostate, renal cell carcinoma, SCLC).

RESULTS: Without open claims, overall sensitivity and specificity were 89.8% and 96.7%, respectively, which aligns with previously published composite mortality endpoint results. The inclusion of open claims improved mortality capture by 5%, with ASCME sensitivity and specificity increasing to 94.9% (95%CI: 94.4-95.1) and 96.3% (95%CI: 96.0-96.6), respectively. Notably, increases in sensitivity were observed in the early disease setting, where RWD death capture has been most challenging, with overall early-stage sensitivity of 93% (range across cancers: 89.9%-94.3%). The inclusion of claims did not alter the date agreement between Patient360^TM and NDI data, with match of 96.4%, 97.8%, 98.8% and 99.2% for 5-day, 15-day, 30-day, and 90-day agreement, respectively.

CONCLUSIONS: The optimal capture of mortality requires open claims, and ASCME represents the best-in-class mortality endpoint available in EHR-focused RWD.