Integrating data systems to improve HIV care engagement in King County, WA

Continuous engagement in HIV care and treatment is crucial for the health of persons living with HIV (PLWH) and for preventing HIV transmission to others. However, in the United States (US), care engagement, or retention in care, represents the biggest drop off in the HIV care continuum, which maps out the care process from HIV testing and diagnosis, linkage to and retention in HIV care, and ultimately achievement of viral suppression. Many health departments in the US use HIV surveillance data to facilitate HIV care engagement activities, a process known as data to care. While Data to Care programs have had some success, their effectiveness is hindered by the completeness and timeliness of HIV surveillance data. A novel approach to Data to Care uses real-time data exchange between HIV surveillance with external data sources, such as emergency department (ED) and inpatient (IP) hospitalization data and jail booking rosters, to improve the signal of Data to Care investigations, and provide a setting and an opportunity to re-engage PLWH in HIV care. Since real-time data exchange involves linking data sources that don’t often have a shared unique person identifier, these programs should also consider the accuracy of the record linkage algorithms they utilize, in order to maximize their reach and efficiency. We investigated the effect of the use of real-time data exchange on HIV care engagement outcomes in two settings: emergency department and inpatient hospitals and in jails. First, we evaluated the impact of an existing ED/hospital-based health information exchange on HIV care outcomes. We compared the proportion of patients that had a viral load test in the 3 months and viral suppression in the 6 months after an alert-eligible ED visit/inpatient admission in the pre-intervention (01/20/13-01/20/15) and post-intervention (07/20/15-07/20/17) periods. To assess whether our pre/post results could be due to secular trends, we compared the difference between patients with an alert-eligible ED visit/IP admission to patients who had a visit outside of the alert window in both the pre-intervention and post-intervention periods. Next, we developed a new automated, real-time data exchange between public health HIV surveillance and county jail data to identify incarcerated PLWH and facilitate post-incarceration HIV care engagement efforts. A team of public health relinkage specialists and jail release planners used this data exchange to guide case conferences about patients who were virally unsuppressed or out-of-care and jointly developed a plan for re-engagement in care and treatment. We compared viral load testing within 3 months and viral suppression within 6 months after release from jail among PLWH released in the post-intervention period (04/01/18-11/01/18) to those released in the pre-intervention period (10/01/16-10/01/17) using Cox proportional hazards models. Finally, we compared the performance of record linkage algorithms commonly used by data exchanges commonly used in public health practice. We compared five deterministic algorithms and two probabilistic record linkage algorithms using simulations and a real-world scenario. We simulated pairs of datasets while varying the number of erroneous fields per record and overlap between these datasets. We matched datasets using each algorithm and calculated their recall (sensitivity) and precision (positive predictive value). In a real-world scenario, HIV and STD surveillance data from King County, WA were matched to identify PLWH who had a syphilis diagnosis. We used manual review to define a gold standard and calculate recall and precision. In our evaluation of an ED/hospital-based health information exchange, patients in the post-intervention period were 1.08 times more likely to have a viral load test within 3 months after an ED visit/IP admission (95% CI: 0.97, 1.20) and 1.50 times more likely to achieve viral suppression within 6 months after an ED visit/IP admission (95% CI: 1.27, 1.76). However, there was a similar pre/post increase in both HIV care engagement (DID: 1.00, 95% CI: 0.84, 1.18) and viral suppression (DID: 1.01, 95% CI: 0.84, 1.20) among patients with visits outside of the alert window. After implementation of a real-time data exchange between HIV surveillance and jail booking data coupled with HIV care coordination between health department and jail release planners, viral load testing within 3 months after release from jail increased by 35% (95% CI: 0.84, 2.18) and viral suppression within 6 months after release from jail increased by 37% (95% CI: 0.82, 2.30), but these differences were not statistically significant. In our simulation study, we found that probabilistic algorithms maintained a high recall at nearly all data quality levels, while being comparable to deterministic algorithms in terms of precision. Deterministic algorithms typically failed to identify matches in scenarios with low data quality. In the real-world scenario, probabilistic algorithms had the lowest trade-off between recall and precision. The results of this dissertation indicate that ED/hospital-based data exchange provides substantial opportunities to interact with PLWH who are poorly engaged in HIV care. However, the observed increase in HIV re-engagement and viral suppression after implementation of this data exchange may reflect secular trends resulting from diverse interventions of which this program was only one. Real-time health information exchange with emergency departments and hospitals can identify PLWH who are inadequately engaged with care and facilitate D2C efforts, but more efforts are needed to improve the effectiveness of reengagement interventions linked to real-time D2C. Implementation of a real-time data exchange between HIV surveillance and jail booking rosters resulted in a trend towards improved post-incarceration HIV care outcomes for incarcerated PLWH who are virally unsuppressed/out-of-care in King County. Real-time data exchange between health departments and county jails is a promising strategy for identifying incarcerated PLWH to support care coordination and improving post-incarceration HIV care engagement. Finally, in our simulation study on record linkage algorithms, we found that probabilistic algorithms maximize the number of true matches identified, while still maintaining high precision. Public health activities that rely on the integration of multiple data sources to target intervention delivery should utilize probabilistic algorithms to reduce gaps in the coverage of interventions and maximize their reach.