Table 1.

Main dimensions of effectiveness and burdens of artificial intelligence and machine learning systems

Concern	Example of failure
Effectiveness
Accuracy	Hospitalisation risk model fails to predict actual hospitalisations
Replicability	Feature selection for risk model cannot be replicated
Generalisability	Risk models works in one hospital, but not another
Explainability	Outputs of the risk model cannot be explained easily to a human user, limiting take-up rate by decision makers
Burden
Bias	Risk model performs well only on one demographic group on which it is trained
Privacy	Risk model uses and/or discloses sensitive information about individual
Due care/process	Individualised patient assessment is compromised owing to risk score