Developing Evidence to Inform Decisions About Effectiveness Research Network: Developing and Evaluating Methods for Record Linkage and Reducing Bias in Patient Registries

Non-experimental comparative effectiveness research (CER) and patient-centered outcomes research (PCOR) provide important information about “real world” combinations of interventions for heterogeneous patient populations. To overcome the limitations of single databases, data from sources such as administrative claims can be linked securely and confidentially with data from patient registries. Registry linkage can facilitate exposure and control group identification, improve measurement of risk factors and outcomes, or allow researchers to monitor events without contacting participants directly. However, at least four major challenges persist. First, unique personal identifiers for linkage are not always available. Second, non-experimental studies are subject to validity threats such as confounding by indication, selection bias, and misclassification of exposures or outcomes. Third, linkage errors can systematically bias estimates of treatment effectiveness. Fourth, although researchers have continued to develop new methods for deterministic and probabilistic linkage, these methods have not been tested thoroughly, and comprehensive guidelines are lacking. This project addresses these challenges by pursuing three major objectives: (1) Develop a framework and guidance for researchers on record linkage of registries to other data sources; (2) Develop and test new methods to improve confounding control and generalizability in CER/PCOR using linked data; and (3) Develop a method to improve instrumental variables analysis using linked registry data.

Principal Investigator: Task Order Pi Til Sturmer,PhD Task PI Alan Brookhart, PhD

Funding: AHRQ

Total Project Period: 9/27/12-7/24/14