Algorithmic Bias
Also called: model bias, health equity bias
Systematic differences in a model's performance across patient groups, usually inherited from unrepresentative training data or biased labels.
Algorithmic bias appears when a model performs measurably worse for some populations than others. Sources include training data that under-represents a group, labels reflecting historical care disparities, and proxy variables that encode access to care rather than disease.
The most cited example is a widely deployed risk model that used healthcare spending as a proxy for illness, systematically underestimating risk in Black patients who had historically received less care for the same conditions.
Where This Gets Misread
The failure was not in the mathematics ... the model predicted spending accurately. It was in the assumption that spending measured illness. Bias in medical AI usually enters through the choice of what to predict, not through the algorithm, and no amount of model auditing catches a badly chosen target.