Enhancing cause-of-death attribution through real-world data integration: A framework to improve cancer mortality identification.

Abstract

e23418 Background: Precise cause of death (COD) is essential for pharmaceutical and healthcare research in evaluating treatment effectiveness, product safety, outcomes, and health-economic analyses. This is amplified in oncological diseases where death, comorbidities, and time from treatment are critical variables for research. However, numerous studies have demonstrated substantial inaccuracies in death certificates, including misclassification of underlying causes, incomplete documentation, and major errors occurring in approximately 10-40% of U.S. certificates driven by limited clinical context and difficulty classifying multimorbid patients - which further undermine COD accuracy. Autopsy-based studies show discordance between recorded and actual causes of death, showing the need for more precision in identify cancer mortality. However, death certificates remain indispensable for mortality surveillance and public health decision-making, necessitating improved approaches in augmenting COD data. These issues are especially impactful in oncology, where misclassification of cancer-related deaths can distort mortality patterns, limit evaluation of treatment effectiveness, and obscure disparities. Methods: The Veritas COD solution was developed utilizing a real-world-data (RWD) model incorporating up-to three years of longitudinal claims, diagnoses, procedures, and Unified Medical Language System (UMLS) concept mapping. This framework creates standardized, contextualized COD profiles that bridge aggregate reporting (e.g., CDC) and ad-hoc data sources. By using UMLS concept mapping as a bridging tool, the solution can draw together multiple disparate sources and calculate likely cause of death and comorbidity information. To evaluate its impact, privacy-preserving record linkage (PPRL) was applied to combine Connecticut death records with Veritas calculated COD values. Cancer-related mortality values sourced from: (1) state-reported COD alone versus (2) enhancement of state-reported data by Veritas COD. Results: Integration of Veritas COD profiles with state mortality data produced a 33% increase in the number of decedents showing cancer as an underlying COD compared with state reporting alone. Newly attributed cases often involved individuals with multimorbidity or ambiguous certificate entries - categories known to exhibit high misclassification rates in research. This demonstrates the model's ability to recover cancer-related deaths overlooked in traditional reporting systems. Conclusions: Using an RWD-anchored COD solution addresses key weaknesses of death certificate-only systems and substantially improves cancer COD ascertainment. This framework supports more accurate epidemiologic insights, strengthens oncology outcomes research, and offers a scalable foundation for enhanced mortality surveillance.