# Ethical Machine Learning in Health

**Authors:** Irene Y. Chen, Emma Pierson, Sherri Rose, Shalmali Joshi, Kadija Ferryman, Marzyeh Ghassemi  
**arXiv:** [2009.10576](https://arxiv.org/abs/2009.10576)  
**Date:** 2020-09-22  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2009.10576v1/

## Key points

- Potential to exacerbate health inequities
- Importance of ethical considerations for equitable ML in healthcare
- Focus on social justice
- Importance of problem selection in ML applications
- Critical factors: global health injustice, racial injustice, gender injustice, diversity in scientific workforce
- Heterogeneous data losses
- Population-specific data losses
- Clinical diagnosis and healthcare costs
- Understanding confounding variables
- Selecting appropriate features
- Tuning parameters effectively
- Defining performance metrics
- Incorporating group fairness definitions
- Quantifying impact
- Assessing model generalizability
- Documenting model and data processes
- Overall goal: Promoting equitable access to quality healthcare services and mitigating biases through ethical considerations throughout the ML model development pipeline.

## Summary

In the field of healthcare, the use of machine learning (ML) has sparked ethical concerns due to its potential to exacerbate existing health inequities. This has prompted a closer examination of the ethical considerations surrounding equitable ML in the advancement of healthcare, particularly through the lens of social justice. One key aspect that has been highlighted is the importance of problem selection in ML applications within healthcare. Issues such as global health injustice, racial injustice, gender injustice, and diversity within the scientific workforce have all been identified as critical factors that must be considered when developing ML models for healthcare applications. Another crucial stage in the process is data collection, where challenges such as heterogeneous data losses and population-specific data losses need to be addressed to ensure that the resulting models are fair and unbiased. Furthermore, accurately defining outcomes in terms of clinical diagnosis and healthcare costs is essential for developing effective ML algorithms. Algorithm development also plays a significant role in ensuring ethical considerations are met. Understanding confounding variables, selecting appropriate features, tuning parameters effectively, and defining performance metrics are all important steps in this process. Additionally, incorporating group fairness definitions into algorithm development is crucial for promoting equity in healthcare outcomes. Post-deployment considerations are equally important in maintaining ethical standards in ML applications for healthcare. Quantifying impact, assessing model generalizability, and documenting both model and data processes are essential steps to ensure transparency and accountability. Overall, addressing these ethical considerations throughout the entire pipeline of developing ML models for healthcare applications is essential for promoting equitable access to quality healthcare services and mitigating potential biases that could perpetuate existing health disparities. By following these guidelines and recommendations outlined by experts in the field , we can strive towards a more just and inclusive healthcare system powered by machine learning technologies.