Project Libra: A New Analytic Tool for Comparative Effectiveness Analyses of Multipayer Claims Databases

Research Objective: The project aimed to develop a secure, interactive tool that will enable researchers to perform comparative effectiveness studies and other types of research on a multi-payer claims database with reduced need for complicated programming. Methods: A common data model, through which multiple data sources are standardized and linked via common data structures and vocabularies, was used to format five administrative databases: Medicare Chronic Condition Warehouse, Thomson Reuters MarketScan® Medicaid Multistate, Medicare Supplemental, and Commercial databases, and Healthcare Cost and Utilization Project National Inpatient Sample database. A web-based User-Interface was developed that captures the logic typically required by CER methods and capitalizes on the longitudinality of administrative data. Tools were developed to allow users to search taxonomies to select particular drugs, diagnoses, and procedures; apply inclusion/exclusion criteria; and create outcome measures. A comparison of outcomes of atrial fibrillation treatment with rate or rhythm control medications was used to pilot test the tool. Results: The tool allowed users to quickly define a study sample using criteria commonly used in claims-based CER studies. The flow diagram created in the tool graphically illustrated the attrition of the sample size as inclusion and exclusion criteria were applied. Embedded SAS procedures enabled analysis of comparison populations. They revealed a higher rate of coronary artery disease and heart failure prior to drug initiation among the amiodorone versus the calcium channel blocker population and a higher rate of post-drug initiation acquired hypothyroidism and pulmonary disease among the amiodorone population. Conclusions: Claims databases contain health care information on millions of patients in the United States. The strengths of Libra tool is that it allows a substantial portion of the research exploration, hypothesis generation; and statistical analysis to be performed in real-time using a web-based interface leveraging a large multi-payer database.