Questioning the Relevance of Sex Categories implemented in Medical Decision Support Systems - The Example of Pulmonary Function

Abstract

Medical decision support systems rely on a variety of data to provide advice and predictions that contribute to diagnoses. Data categorizations, which are often hidden and therefore invisible, play a major role in the statistical models implemented in these digital tools. Male/female bicategorization is a paradigmatic case that has been little studied in this context. This study analyzes its use and determinants in pulmonary function measurement as a case study. Using a corpus of sixty articles, carefully selected in the medical literature for their representativeness, and the reference equations on which decision support systems are based, we investigate the role of male/female categorization as used in everyday clinical practice, its origins, and the place of sex and gender in this issue. This research reveals a naturalization of men/women differences in favor of sex, underpinning a binary essentialism of biological sex. Naturalization of men/women differences is a major concern for the use of predictive artificial intelligence models and the development of decision algorithms, with a possible worsening of health disparities as a result of biases in training data.