How were these recommendations developed?
STANDING Together is based on a multi-step process that incorporates the views of diverse stakeholders (including patients and public representatives), a systematic review of existing health data standards, and a modified Delphi study.
The STANDING Together recommendations aim to be widely applicable and meet the needs of as many users as possible. Our green paper and this website openly reports the findings so far, and provides an opportunity for them to be ‘sense-checked’ and to be refined in the light of further feedback.
Blue boxes = initial item generation
Yellow boxes = compiled versions of recommendations
Pink boxes = delphi study activities
Green boxes = stakeholder involvement & feedback.
Who are these recommendations for, and how should they be used?
The STANDING Together recommendations aim to reduce bias and exacerbation of health inequalities caused by healthcare AI applications, by providing standards that can support dataset documentation and dataset usage. To achieve this, our standards are split into two parts:
Aimed at dataset curators, to ensure transparent documentation of healthcare datasets*.
Allows users of the data to understand the limitations and potential biases within datasets.
Not intended to be prescriptive about dataset content
Aimed at dataset users, to enable informed decision making regarding dataset use
Encourages users, evaluators, regulators and other stakeholders to ensure datasets appropriately align with the intended purpose of the health technology under development.
Recommendations proposed are aligned with requirements of medical device regulations.
Each part can be used as a standalone tool, or they may be used together. Evaluators and regulators of AI and digital health technologies may find value in both parts of the standards.
* Please refer to Healthsheet and other similar artefacts for more detailed guidance about the structure and content of dataset documentation.
Read about our recommendations:
Click the links below to read each part of our recommendations: