Table 2 Equalized odds evaluation for COVID-19 prediction task (ethnicity mitigation) on prospective and external test sets, optimized to sensitivities of 0.9.

From: An adversarial training framework for mitigating algorithmic biases in clinical machine learning

 

Prospective Validation

External Validation

OUH (n = 22,857, prevalence = 8.80%)

PUH (n = 37,896, prevalence = 5.29%)

UHB (n = 10,293; prevalence = 4.27%)

BH (n = 1177; prevalence = 12.2%)

Basic

Adv

Basic

Adv

Basic

Adv

Basic

Adv

TP SD

0.0387

0.0224

0.0490

0.0480

0.0985

0.0975

0.0964

0.0954

FP SD

0.0574

0.0539

0.0361

0.0361

0.0510

0.0387

0.1039

0.0970

  1. Results reported as SD of true positive and false positive rates, across all ethnicity labels. Bolded values denote best scores.
Back to article page