50.11 Reported Discrimination Predicts Biomarkers of Adverse Medical Outcomes Using Machine Learning Methodology

Abstract

Within the past decade, there has been increased attention to the relationship between stress, its physical manifestations, and complex psychosocial, interpersonal interactions. One such stressor that has been shown to significantly impact is the experience of perceived discrimination for members of African American (AA) communities. Racial trauma (or race-based stress response) is defined as the traumatic response to negative race-related experiences that are collectively characterized as racism, including acts of prejudice, discrimination, or violence against a subordinate racial group. Consistently, it has been shown that African Americans are among the hardest hit by this sociopolitical phenomenon, racism, in the United States. For African Americans, the experience of racism has been linked to earlier onset (and greater prevalence) of disability, reduced life expectancy, and various cardiometabolic, neurological, or mental health disorders than any other racial or ethnic minority group in the United States. In this study, we are interested in utilizing machine-based learning methodologies to investigate if such intrinsic factors, such as perceived prejudice or discrimination in childhood/adolescence, are predictive of poorer health outcomes in adulthood. Utilizing the National Longitudinal Study of Adolescent to Adult Health (Add Health) dataset, we utilized a recursive feature elimination regression analysis to investigate the relationship between reported discrimination and potential biomarkers of adverse health outcomes in adulthood. We found that reported discrimination was predictive of elevated levels of high-sensitivity C-reactive protein (hsCRP) (R2 = 0.2579), increased BMI, in adulthood. Our findings are proof of concept that machine learning methodology can be applied to large datasets to understand the potential health consequences of racial disparities in a hypothesis-free manner.