Background: National cause-specific death rates are based on cause of death as reported on death certificates.
Objectives: We sought to determine whether cause-specific death rates can feasibly be developed from claims near the date of death.
Methods: This work utilized the Optum Research Database (ORD), which contains pharmacy and medical claims from a large commercial insurer. Month and year of death in the ORD during 2020 were identified based on data from the Social Security Death Master File, supplemented by publicly available obituary data and International Statistical Classification of Diseases, 10th Revision (ICD-10) codes R99 (Ill-Defined/Unknown Cause of Mortality) and G93.82 (Brain Death). The top 10 causes of death nationally for 2020 were identified based on ICD-10 codes reported by the Centers for Disease Control and Prevention. For each cause of death, national death rates for individuals >1 years of age were age-standardized to the 2020 estimated US Census population. Corresponding ORD disease-specific death rates were estimated from a simple algorithm based on an ICD-10 code for the disease in the month of death. Person-time at risk was estimated using a mid-year count of patients enrolled in the ORD.
Results: For patients >1 years old, age-standardized death rates for the top 10 causes of death (per 100,000 US standard population) were as follows: diseases of the heart, 204.3 nationally, 367.4 ORD; malignant neoplasms, 176.6 nationally, 112.8 ORD; COVID-19, 103.2 nationally, 98.6 ORD; accidents, 58.5 nationally, 34.1 ORD; cerebrovascular diseases, 46.6 nationally, 93.2 ORD; chronic lower respiratory diseases, 44.6 nationally, 154.4 ORD; Alzheimer disease, 39.7 nationally, 40.0 ORD; diabetes mellitus, 29.8 nationally, 176.9 ORD; influenza/pneumonia, 15.9 nationally, 172.3 ORD; nephritis/nephrotic syndrome/nephrosis, 15.9 nationally, 242.6 ORD.
Conclusions: These results suggest that a claims-based cause of death assessment leads to cause-specific death rates that are similar to those based on death certificate data. Whether they are similar enough depends on the research question and specifics of study design. It may be feasible to accurately estimate disease-specific death rates in claims; however, more complex claims-based algorithms could be used to improve accuracy, particularly for deaths due to chronic manageable disease.
