Health disparities in AI-based medical triage and contact tracing during the COVID-19 pandemic

The COVID-19 pandemic has boosted the AI systems and algorithms developed for early detection and prognosis in risk patients in order to support clinicians to make quick decisions when healthcare resources are limited. However, such technological dependence, mainly based on quantitative and biological data of patients, may perpetuate some biases that condition the results obtained. Our hypothesis is that medical triage supported by automatized algorithms may have the risk to overshadow health disparities exacerbated by a lack of qualitative data. In this contribution, we aim to address what are the biases in AI systems and algorithms developed for medical triage and contact tracing regarding COVID-19 and their relationship with social determinants of health. We wonder whether or not, and how, health disparities are taken into account when AI systems applied to COVID-19 are developed. We carried out a scoping review in Medline, Pubmed, Scopus, Cinahl, WOS, Wiley Online Library and Arxiv.x with a search strategy focused on 1) identifying bias in AI systems developed for triage, risk prediction or contact tracing for COVID-19, and 2) identifying if health disparities are embraced in the corresponding literature or, on the contrary, they are overlooked. We found that concepts such as “social determinants of health” or “health disparities” are not explicitly mentioned in most of the articles examined. Likewise, factors like ageism, poverty, ethnicity or gender do not seem to be considered either, which skews the sample of the population selected as reference for automatized learning and AI decision-making criteria. We conclude that it would be appropriate to incorporate more qualitative data and social disparities concerns in future development of AI systems on medical triage or contact tracing for COVID-19.