Improving the Use of Patient Registries for Comparative Effectiveness
A key challenge in prospective registries is that patient-reported outcomes (PROs) and physicians' ratings of effectiveness and toxicity may be collected at structured intervals that do not coincide with the starting and/or stopping of treatment. Information on treatment starts and stops may be collected only periodically, so that the precise dates of these events may not be known. We propose the following aims to address the challenges presented by these uncertainties. First, we will examine the effect of varying methodological choices for dealing with PROs and treatment exposures under different assumptions in a prospective registry of rheumatoid arthritis (RA). Since few RA treatments have been compared in head-to-head RCTs, there is little comparative effectiveness data. The investigators have access to a prospective RA patient registry (BRASS) that includes 10 years of follow-up for over 1,100 patients. In addition to semiannual patient and physician questionnaires, the investigators have access to electronic medical records; this information can serve as a reference standard. We will conduct focus groups with RA patients to understand which PROs are most important. These results will inform simulation studies to evaluate potential strategies for dealing with the methodological challenges of prospective registries under different assumptions on various outcomes and exposures. Second, we will test the effects of the simulation results in outcomes models using the actual data from the relevant RA prospective registry. Among patients who have failed treatment with a TNF antagonist to (a) compare the effectiveness of two different biologic treatments (abatacept versus tocilizumab) and (b) compare the risk of infection associated with abatacept versus tocilizumab. About 50% of patients with RA will eventually receive a biologic agent. Almost all patients receive a TNF antagonist as first-line biologic, but 50% of patients will not respond adequately or cannot tolerate them, requiring a biologic agent with a different mechanism of action, such as abatacept and tocilizumab. There is almost no evidence-basis to choose between agents. Data from BRASS will be analyzed to study these questions using the different assumptions tested in the simulations from Aim 1. Results of the models will be compared with other available information to determine the best analytic strategies. Finally, we will survey patients and providers about the acceptability of different intervals for reporting PROs and physician-reported outcomes. With information from Aims 1 and 2 on how different assumptions on reporting PROs and physician-rated outcomes bias analyses, we will go back to patients and providers involved in the BRASS registry to understand how the burden of reporting affects their involvement.