STANDING Together: Standards for Data Diversity, Inclusivity and Generalisability in health datasets
Abstract
This is a report of the STANdards for Data Diversity, Inclusivity and Generalisability (STANDING Together) programme, international multi-institutional collaboration, to build recommendations towards best practices for transparent reporting and evaluation of bias as it affects datasets used to develop or evaluate AI health technologies. The programme was announced in November 2021 and used standard consensus methodology to build recommendations, comprising a multi-staged modified e-Delphi approach. It has a specific focus on datasets used for the development, evaluation and monitoring of AI health technologies, and responds to increasing international awareness of algorithmic bias, and the wider harms of under-representative, biased datasets. It builds upon previous work on the documentation of datasets within and beyond the medical domain. This Consensus Statement describes the STANDING Together Delphi study, from development of the candidate items to the final recommendations and accompanying explanatory text.
