# Healthsheet: Development of a Transparency Artifact for Health Datasets

**Authors:** Negar Rostamzadeh, Diana Mincu, Subhrajit Roy, Andrew Smart, Lauren Wilcox, Mahima Pushkarna, Jessica Schrouff, Razvan Amironesei, Nyalleng Moorosi, Katherine Heller  
**arXiv:** [2202.13028](https://arxiv.org/abs/2202.13028)  
**Date:** 2022-02-26  
**Summary page:** https://www.summarizepaper.com/en/paper/2202.13028v1/

## Key points

- Machine learning (ML) has shown great potential in healthcare applications
- Ethical concerns stem from structural inequalities in data collection, usage, and handling
- Developing guidelines for creating, using, and maintaining ML healthcare datasets is essential
- Healthsheet introduced as a contextualized adaptation of the datasheet questionnaire for health-related applications
- Contextualizing datasheets for healthcare is necessary
- Inconsistency in broader use of accountability practices like datasheets within the ML for health community
- Lack of incentives for creating datasheets but long-term benefits recognized
- Suggestions to change incentives to prioritize data sharing over publications and standardize reporting regulations
- Healthsheets facilitate efficient dataset usage and long-term follow-up through detailed documentation

## Summary

, , , , 

Machine learning (ML) has shown great potential in various healthcare applications, with data playing a crucial role in the development of ML-based systems. However, ethical concerns surrounding its use often stem from structural inequalities in data collection, usage, and handling. To address this issue, developing guidelines for creating, using, and maintaining ML healthcare datasets is essential. In this study, Healthsheet was introduced as a contextualized adaptation of the datasheet questionnaire specifically designed for health-related applications. Through semi-structured interviews and case studies on publicly-available healthcare datasets such as Electronic Health Records (EHR) and clinical trial study data, it was found that contextualizing datasheets for healthcare is necessary. Despite incentives to adopt accountability practices like datasheets, there is inconsistency in their broader use within the ML for health community. Participants acknowledged the lack of incentives for creating datasheets but recognized the long-term benefits of reduced overhead from investing in documentation upfront. Suggestions were made to change incentives within the community to prioritize data sharing over publications and standardize reporting regulations. Overall, Healthsheets serve as a valuable transparency artifact for health datasets by facilitating efficient dataset usage and long-term follow-up through detailed documentation that aids researchers in understanding data origins and potential usages while securely storing it.