Data Science for Decision Support in the HIV Care Cascade

Abstract HIV and AIDS continue to be significant public health issues, but with recent advances in treatment, technology, clinical and social support, the research and treatment agenda now explicitly and realistically includes bringing the decades-long pandemic to an end. The President’s Emergency Plan for AIDS Relief (PEPFAR) is an ongoing multi-billion investment to deliver antiviral therapy to those in low- and middle-income countries (LMIC), and has been regarded by many as the most successful public health intervention in modern history, having dramatically reduced both prevalence and incidence of HIV over the past two decades. With both clinical trials and observational studies conclusively demonstrating that immediate treatment with antiretroviral therapy (ART) is the mosteffective way to both treat HIV and prevent the transmission of new infections, retention in HIV care and suppression of viral load through compliance with ART are arguably the most effective methods available for bringing the pandemic to an end, and indeed are encoded in the UNAIDS 95-95-95 benchmarks of having 95% of cases diagnosed; 95% of diagnosed cases initiated and retained on ART; and 95% of treated individuals achieving viral suppression. Clinical decision support systems (CDSS) tailored to the requirements of LMICs have been shown to improve compliance with guidelines and quality of care by a range of healthcare staff. Use of machine learning algorithms allows the development of prediction models for clinical complications and outcomes, which can guide health care staff in early identification of problems and appropriate interventions. The Specific Aims of this proposal therefore are (1) to use a large electronic health record (EHR) database to develop and validate statistical machine learning models to identify patient at high risk for loss to follow up and viral failure; (2) to develop and field test implementation of clinical decision support tools based on these models that will be implemented at the point of care; and (3) to evaluate the efficacy of the decision support tools, in terms of improving patient retention and reducing viral failure, using a randomized comparison at the clinic level. Our project will be implemented at the Academic Model Providing Access to Healthcare (AMPATH), an HIV care program in western Kenya serving nearly 200,000 people with HIV.