Abstract P199: Development of an Algorithm for Identifying Fatal Cardiovascular Disease in Medicare Claims
Introduction: Administrative claims data including Medicare are often used to study cardiovascular disease (CVD). Approximately one-third of CVD events are fatal, but claims data lack information on cause of death and claims-based analyses only include non-fatal outcomes.
Objective: To develop claims-based algorithms to discriminate fatal stroke, fatal coronary heart disease (CHD), and fatal CVD (fatal CHD or stroke) from deaths due to other causes.
Methods: We analyzed data for REasons for Geographic and Racial Differences in Stroke (REGARDS) study participants who died between January, 1, 2003 and December, 31, 2013, and were age ≥65.5 years and had at least 182 consecutive days of fee-for service Medicare coverage prior to their death. Deaths within 28 days following an expert-adjudicated stroke, CHD, and CVD event in REGARDS were defined as the gold standard. Logistic regression was used to develop algorithms including demographic data, ICD-9-CM diagnosis and procedure codes from Medicare claims for hospitalization and outpatient visits. We chose cut-points of predicted probability to define fatal stroke, CHD and CVD using these algorithms. Cut-points were chosen such that a similar proportion of participants had a fatal stroke, fatal CHD, and fatal CVD through REGARDS adjudication and the algorithms.
Results: We analyzed data for 2,675 deaths among REGARDS study participants; 142 fatal strokes, 479 fatal CHD events, and 608 fatal CVD events. The algorithms are described in the footnote to the Table. C-statistics for the algorithms were 0.966 (95%CI 0.948, 0.980) for fatal stroke, 0.861 (95%CI 0.843, 0.878) for fatal CHD and 0.883 (95%CI 0.868, 0.898) for fatal CVD. Sensitivity, specificity, and positive and negative predictive values using cut-points of 24%, 30%, and 34% predicted probabilities for fatal stroke, fatal CHD and fatal CVD, respectively, are provided in the Table.
Conclusion: The algorithms we developed can be used to identify Medicare beneficiaries with fatal CVD events.
Author Disclosures: F. Xie: None. L. Colantonio: None. J. Curtis: B. Research Grant; Significant; Amgen. G. Consultant/Advisory Board; Modest; Amgen. M. Kilgore: B. Research Grant; Significant; Amgen. G. Consultant/Advisory Board; Modest; Amgen. E. Levitan: B. Research Grant; Significant; Amgen. G. Consultant/Advisory Board; Modest; Amgen, Novartis. K. Monda: A. Employment; Significant; Amgen. M. Safford: B. Research Grant; Significant; Amgen. G. Consultant/Advisory Board; Modest; Amgen. B. Taylor: A. Employment; Significant; Amgen. M. Woodward: G. Consultant/Advisory Board; Modest; Amgen. P. Muntner: B. Research Grant; Significant; Amgen.
- © 2017 by American Heart Association, Inc.