(142) Assessing the potential value of commercial mortality data combined with publicly available death data for pharmacoepidemiology research

Background: Amalgamating death information from oncology electronic health record (EHR), commercial, and US Social Security Death Index (SSDI) data sources has been shown to result in a highly reliable mortality variable for oncology populations, but requires access to EHR data. The National Death Index (NDI) is considered a gold standard for mortality data, but its utility in research is limited due to its two-year time lag.

Objectives: In this study, we examined the reliability of a combined commercial mortality database, drawn from public obituary data (OD) and SSDI alone, as a viable option for research in which EHR data is unavailable for potential applications in non-oncology research.

Methods: This study used the nationwide Flatiron Health EHR-derived de-identified database, which contains OD collected since 2010. This OD, along with the Social Security Death Index (SSDI) and electronic health record (EHR) data (structured and unstructured), are curated to derive a real-world composite mortality variable (CMV). While the latest CMV has been validated against the gold standard of the National Death Index (NDI) (91%+ sensitivity and 98% specificity) in oncology populations, it hasn’t been evaluated for non-oncology populations.

Results: In an analysis of 160,436 patients across 18 cancer types, we found that the combination of OD and SSDI (without EHR data) had a 74.5% sensitivity and 98.3% specificity when compared to the NDI as a gold standard. This was more complete than the OD alone (66.3% sensitivity) or the SSDI alone (26.6% sensitivity) when compared to the NDI.

Conclusions: The combination of the OD and SSDI identified vital status with high accuracy. Given the flexibility of this data to be used in direct linkage or via tokenization, it is likely that when combined with any other dataset (e.g., medical claims, trials, etc.), it would serve as a viable measure of mortality for pharmacoepidemiology real-world research.