The Fairness principle states that medical AI tools should maintain the same performance across individuals and groups of individuals. AI-driven medical care should be provided equally for all citizens, independently of their sex, gender, ethnicity, age, socio-economic status and (dis)abilities, among other attributes. Fair medical AI tools should be developed such that potential AI biases are minimized as much as possible, or identified and reported.
| Recommendations | Operations | Examples |
|---|---|---|
| Define any potential sources of bias (fairness 1) | Engage relevant stakeholders to define the sources of bias | Patients, clinicians, epidemiologists, ethicists, social carers |
| Define standard attributes that might affect the AI tool’s fairness | Sex, age, socioeconomic status | |
| Identify application specific sources of bias beyond standard attributes | Skin colour for skin cancer detection, breast density for breast cancer detection | |
| Identify all possible human biases | Data labelling, data curation | |
| Collect information on individuals’ and data attributes (fairness 2) | Request approval for collecting data on personal attributes | Sex, age, ethnicity, socioeconomic status |
| Collect information on standard attributes of the individuals | Sex, age, nationality, education | |
| Include application specific information relevant for fairness analysis | Skin colour, breast density, presence of implants, comorbidity | |
| Estimate data distributions across subgroups | Male v female, across ethnic groups | |
| Evaluate fairness and bias correction measures (fairness 3) | Select attributes and factors for fairness evaluation | Sex, age, skin colour, comorbidity |
| Define fairness metrics and criteria | Statistical parity difference defined fairness between −0.1 and 0.1 | |
| Evaluate fairness and identify biases | Fair with respect to age, biased with respect to sex | |
| Evaluate bias mitigation measures | Training data resampling, equalised odds postprocessing | |
| Evaluate impact of mitigation measures on model performance | Data resampling removed sex bias but reduced model performance | |
| Report identified and uncorrected biases | In AI information leaflet and technical documentation |