Background. Medicare is one of the world's largest health insurance programs. It provides health insurance to nearly 44 million beneficiaries whose entitlements are based on age, disability, or end-stage renal disease (ESRD). Data of these ESRD beneficiaries are collected in the US Renal Data System (USRDS), which includes comorbidity information entered at the time of dialysis initiation (medical evidence data), and are used to shape health care policy. One limitation of USRDS data is the lack of validation of these medical evidence comorbidities against other comorbidity data sources, such as medical claims data. Methods. We examined the potential for discordance between USRDS Medical Evidence and medical claims data for 11 comorbid conditions amongst Medicare beneficiaries in 2011-2013 via sensitivity, specificity, kappa and hierarchical logistic regression. Results. Among 61,280 patients, most comorbid conditions recorded on the Medical Evidence forms showed high specificity (> 0.9), compared to prior medical claims as reference standard. However, both sensitivity and kappa statistics varied greatly and tended to be low (most < 0.5). Only diabetes appeared accurate, whereas tobacco use and drug dependence showed the poorest quality (sensitivity and kappa < 0.1). Insti- tutionalization and patient region of residency were associated with data discordance for six and five comorbidities out of 11, respectively, after conservative adjustment of multiple testing. Discordance appeared to be non-informative for congestive heart failure but was most varied for drug dependence. Conclusions. We conclude that there is no improvement in comorbidity data quality in incident ESRD patients over the last two decades. Since these data are used in case-mix adjustment for outcome and quality of care metrics, the findings in this study should press regulators to implement measures to improve the accuracy of comorbidity data collection.
ASJC Scopus subject areas
- Biochemistry, Genetics and Molecular Biology(all)
- Agricultural and Biological Sciences(all)