035 The cutoff biases and their consequences

Placing cutoffs on ordinal tests validated against a binary reference standard leads to two distinct but reinforcing biases that jeopardise optimal care decisions for the individual patient-as-person. The pre-emption bias arises from the embedding of tradeoff preferences through the prescription of an ‘optimal’ cutoff. The imprecision bias arises from the use of parameters at a cutoff score rather than those at the patient’s precise test score as input into the clinical decision. Logical and empirical reasons suggest both biases contribute to population-level overdiagnosis and overtreatment, though this is not an important reason why cutoffs should be abandoned. The empirical argument is based on analysis of datasets for the Edinburgh Postnatal Depression Scale, the Patient Health Questionnaire 9 and the Montreal Cognitive Assessment instruments. The mutually supporting research and practice drivers behind cutoff-based testing are explored, along with a proposed alternative approach.