Methodology for linking Ryan White HIV/AIDS Program Services Report (RSR) client level data over multiple years

BACKGROUND: The Health Resources and Services Administration’s (HRSA), HIV/AIDS Bureau (HAB) is responsible for leading the nation’s efforts to provide health care, medications, and support services to low-income people living with HIV through the Ryan White HIV/AIDS Program (RWHAP). The RWHAP funds and coordinates with cities, states, and local community-based organizations to deliver efficient and effective HIV care, treatment, and support services for over half a million vulnerable people living with HIV (PLWH) and their families in the United States. The annual RWHAP Services Report (RSR) is an important source of information for monitoring RWHAP’s progress towards National HIV/AIDS Strategy goals. Since 2010, HRSA HAB has used the annual client-level RSR data to monitor program-related outcomes, conduct program evaluations, understand service provision, and conduct extensive analysis on disparities in viral suppression and retention in HIV care. HRSA HAB receives annual RSR submissions from RWHAP recipients and sub-recipients. However, the de-identified nature of the data limits HRSA HAB’s ability to expand beyond year-to-year analyses and conduct additional analyses to evaluate outcomes for clients who are seen in multiple years. The current paper describes the development and validation of a method to link RSR client-level records across multiple data years. METHODS AND FINDINGS: Using seven RSR reporting years of data (2010 to 2016), we applied a Fellegi-Sunter (F-S) linkage model that used client demographic characteristics and their providers’ geographic locations to calculate matching weights for each record pair based on estimated agreement and disagreement conditional probabilities across RSR years. To validate our methodology, we conducted an internal sample review and external validation to assess the level of accuracy of the linkage, and the extent to which the linked data set corresponds accurately to clinical records of individual clients. The linkage result yielded 70 to 80 percent year-to-year client carry-over rate over seven years of the RSR data; 96 percent linkage ratio from the internal sample review and 79.9 to 94.2 percent of provider network client carry- over rate per year from the external validation. CONCLUSIONS: This methodology addresses a gap in data analysis capabilities by allowing HRSA HAB to link RWHAP clients across reporting years. Despite weak identifying information and lack of continuity of service reporting, the longitudinal linkage improves HRSA HAB’s ability to evaluate the patterns of viral suppression and monitor service utilization over time for individuals who receive services in multiple years. These analyses will support future analytic activities in understanding the impact and outcomes of the RWHAP, and will assist HRSA HAB in monitoring progress toward meeting National HIV/AIDS Strategy goals. For those looking for ways to assess health services data, the F-S unsupervised method combining weak identifying attributes and geographic proximity offers practical solutions to the problem of linking de-identified information about individuals across multiple years and improving longitudinal research.