Table 5.
University hospital – fairness ratios (false negative rate and false discovery rate) relative to the reference groups, female for sex and 30-40 years for age, are assessed using the 80% rule, where values within (0.8-1.25) are fair. Median false negative rate and false discovery rate ratio scores and 90% CI (5th and 95th percentiles): Sex – the artificial intelligence model is fair – achieves both equal opportunity (false negative rate ratio) and predictive parity (false discovery rate ratio). Age range - the artificial intelligence model shows fairness infringements for patients aged 40-120 y for predictive parity (false discovery rate ratio) and no fairness infringements for equal opportunity (false negative rate ratio).
| Feature and group | FNRa ratio (90% CI) | FDRb ratio (90% CI) | ||
| Sex |
|
|
||
|
|
Male | 0.957 (0.937-0.983) | 1.005 (0.969-1.038) | |
|
|
Femalec | 1.000 (1.000-1.000) | 1.000 (1.000-1.000) | |
| Age range (y) |
|
|
||
|
|
10-20 | 0.849 (0.556-1.093) | 0.000 (0.000-0.000) | |
|
|
20-30 | 0.964 (0.846-1.080) | 0.857 (0.304-1.396) | |
|
|
30-40c | 1.000 (1.000-1.000) | 1.000 (1.000-1.000) | |
|
|
40-50 | 0.946 (0.882-1.031) | 1.273 (0.654-1.527) | |
|
|
50-60 | 0.899 (0.890-0.976) | 1.421 (0.959-1.544) | |
|
|
60-70 | 0.862 (0.816-0.898) | 1.440 (0.961-1.607) | |
|
|
70-80 | 0.843 (0.780-0.873) | 1.448 (0.931-1.686) | |
|
|
80-90 | 0.830 (0.771-0.849) | 1.454 (0.947-1.630) | |
|
|
90-100 | 0.806 (0.755-0.839) | 1.391 (0.957-1.585) | |
|
|
100-110 | 0.821 (0.781-0.900) | 1.366 (0.954-1.582) | |
|
|
110-120 | 1.132 (1.126-1.185) | 2.000 (1.319-2.280) | |
aFNR: false negative rate.
bFDR: false discovery rate.
cReference Group.