Retention in care is essential to HIV treatment and prevention, yet less than half of people living with HIV in the U.S. are retained in medical care. Effective retention interventions, such as intensive case management and patient navigation, are highly resource intensive. With diminishing resources for HIV care, better approaches are needed to identify patients at highest risk for retention failure who would most benefit from retention resources. A predictive model may quantify a specific patient's risk of future retention-in-care failure based on his/her unique characteristics. Such a predictive model based on electronic health data and supplemental social factor informed data could be automated to generate risk prediction in real time. Instead of attempting to locate and re-engage patients who are “lost to follow-up” as is the current practice, a predictive model would allow case managers to identify at risk clients and intervene to prevent retention failure before it occurs. I have a strong background in clinical informatics, biostatistics, and epidemiology. Through this K23, I will further develop my skills in longitudinal data analysis and advanced data analytics and create a predictive model of retention in care. In Aim 1, I will create a predictive model of retention in care using EHR data from a large clinical data research network spanning 11 healthcare systems in Chicago, utilizing mixed effects logistic regression and random forest. Through Aim 2, I will evaluate whether the addition of supplemental social factor informed electronic data sources into the predictive model enhances its performance (e.g., unstructured text of EHR notes, geospatial data, social media data). Finally, in Aim 3, I will explore the feasibility of using the model in real time to increase retention efforts for at-risk patients. I will complete this project under the supervision of my mentor (Dr. John Schneider), co-mentor (Dr. David Meltzer), and my advisory team (Dr. Robert Gibbons, Rayid Ghani, and Dr. C. Hendricks Brown). Together, this multidisciplinary team brings nationally renowned expertise in HIV research, EHR research, longitudinal data analysis, natural language processing, social media data, implementation science, and ethics. In addition, they serve as Directors of the Chicago Center for HIV Elimination (Schneider), Center for Health and the Social Sciences (Meltzer), Center for Data Science and Public Policy (Ghani), Center for Health Statistics (Gibbons), and Center for Prevention Implementation Methodology for Drug Abuse and HIV (Brown). An integrated program of coursework, seminars, structured mentorship, research activities, and conferences will provide me with the skills necessary to complete the proposed research and transition to independence. My long-term career goal is to become an independent investigator utilizing HIV informatics to develop prediction models and tools to inform HIV prevention and treatment across the HIV care continuum. The mentorship and training that I will receive through this K23 award will provide me with the foundation necessary to pursue that goal and this proposal will form the basis for future R01 proposals.

Retention in care is essential for HIV treatment and prevention, yet less than half of people living with HIV in the United States are retained in medical care. Electronic data, including advanced electronic health records, geospatial data, and social media profile data, can be used to create an automated predictive model to identify individuals at highest risk for retention in care failure. Instead of attempting to locate and re-engage patients who are no longer retained in care as is the current practice, a predictive model will allow case managers to identify at risk individuals and provide personalized retention resources based on individual risk factors before patients disengage from care.