Research on artificial intelligence (AI) has emerged as a promising field that has the potential to improve patient outcomes, for example, by optimizing timing of antibiotic therapy in the intensive care unit (ICU) or by AI-based delirium management, as recently published in this journal [1, 2]. Despite its potential, we have to be aware that not all patients may equally benefit from such advancements;‘unfair’or ‘unequal’AI algorithms could reinforce systemic health disparities. For example, a recent study demonstrated consistent underdiagnosed chest X-ray pathologies by an AI algorithm in black and female patients [3]. In fact, even well-established ICU prediction models could be unfair. During the COVID-19 pandemic, Sequential Organ Failure Assessment (SOFA)-based allocation of ICU resources was proven to have racial inequality and could have induced disparities [4]. These results stress that especially future AI-based ICU interventions, or policies, should be fair and have a similar impact on all patients involved, irrespective of gender, ethnicity, and other protected personal characteristics as recently stated by the World Health Organization (WHO)[5].
One of the reasons AI research has skyrocketed in intensive care medicine [6] is the availability of large publicly available datasets, such as the Medical Information Mart for Intensive Care (MIMIC)[7]. These data are often collected at single site and as such could underrepresent