Secondary Analysis of Cognitive Data from the Multi-Center AIDS Cohort Study (MACS) to Identify Longitudinal Change Phenotypes in HIV+ Individuals

点击翻译按钮获取中文摘要

PROJECT SUMMARY Antiretroviral therapies (ART) have modified Human Immunodeficiency Virus (HIV) infection from a nearly universal fatal disease to a manageable chronic condition, yet associated symptoms like cognitive impairments (CI) persist at a higher rate than in comparable uninfected control populations. These impairments are heterogeneous in terms of presentation (e.g. cognitive domains) and trajectory. Given that domain-specific impairments are likely due to different pathological changes, studies that investigate CI based on a global metric of impairment are likely to miss this. The application of advanced analysis methods, including machine learning to interrogate complex diseases and biological processes, is beginning to transform our understanding of the underlying mechanisms that contribute to complex diseases, including CI in the setting of HIV infection. In order to identify and characterize factors that contribute to nonlinear patterns of cognitive change, we propose to capitalize on over 20 years of data from the Multi-Center AIDS Cohort Study (MACS). MACS is one of the largest and longest running studies of men infected with HIV, that includes a substudy with biannual neurocognitive testing that began in 1988. A complete battery of longitudinal neurocognitive testing data is available to us through this cohort, as well as concurrent datasets that include demographic, clinical, psychiatric, lifestyle and biochemical data. The data set consists of over 3500 subjects with over 500,000 data points. In Aim 1, we propose to use data-driven methods (k-means clustering for joint longitudinal trajectories and dynamic time warping) to identify groups of individuals with distinct domain-specific patterns of cognitive change. In Aim 2, we will use the vast amount of data collected on these subjects to identify subtype-specific variables that contribute or predict group membership. By using advanced machine learning methods that are unconstrained by preset statistical or biological assumptions, we are uniquely positioned to identify factors that contribute to longitudinal patterns of change in specific cognitive functions.