Mission and objectives
The aim of this tutorial is to cover the basic principles of fairness and bias in health data science and highlight the ways in which bias can arise throughout the research pipeline. This will be an interactive session with a substantial amount of group discussion. Participants will be encouraged to contribute perspectives drawn from their own work so that learnings can be shared across disciplines.
Intended Audience
This tutorial is suitable for participants with some exposure to data science / statistical analysis. Deep technical skills are not required.
Expected Outcomes
After this tutorial, participants will be able to:
- Recognize how bias can emerge at various stages of a healthcare data analysis pipeline.
- Understand the consequences of biased datasets, particularly in contexts such as genetic data or underreported medical conditions in health data science.
- Gain a general understanding of methods for assessing algorithmic fairness.
Format and Schedule
| Time | Activity | Session |
|---|---|---|
| 10 min | Ice-breaker | |
| 30 min | 1 | What’s ‘fair’ anyway? An introduction to bias and fairness in health data science |
| 30 min | 2 | Case studies: Fairness in the real world |
| 15 min | 2 | Case study presentations |
| — | — | Break |
| 20 min | 3 | Where bias creeps in: A tour of the data analysis pipeline |
| 30 min | 4 | Breakout groups: Designing solutions for fairness across the pipeline |
| 15 min | 4 | Lightning pitches |
| 20 min | 5 | Looking ahead: the future of fairness in health data science (Rotating panel discussion) |
