Translating intersectionality to fair machine learning in health sciences

Collins and Bilge² articulate six core ideas for intersectionality (Table 1). For illustration, we consider the hypothetical task of predicting cardiovascular events among a cohort of US hospital patients inclusive of Black transgender women. The first two ideas — social inequality and intersecting power relations — are best understood together. In relation to our task, the social inequalities in access to routine, high-quality primary care and health insurance for Black transgender individuals are due, in part, to intersecting oppressive power systems such as racism³ and transphobia⁴….