Addressing the ‘optimal cutoff’ Bias in Primary Care Testing

Le résumé fourni par la source

INTRODUCTION: Ordinal tests, an important source of information on mental disorders in primary care, are validated by their 'diagnostic accuracy' in relation to a binary reference standard. Their Sensitivities and Specificities are measured at each possible cutoff on the test scale and the 'optimal' cutoff determined as that which maximises the separation of healthy from unhealthy populations, usually by Youden's statistic. This gives equal weight to Sensitivity and Specificity, which, while valid from this population research perspective, has no relevance in clinical practice (screening, case-finding). Use of an 'optimal' cutoff breaches the requirement for the patient to be provided with the most individualised test output metrics possible, in order that they may bring their informed preferences to bear in the post-test decision making. METHOD: The magnitude of the optimality bias is investigated empirically using the DEPRESSD Individual Participant dataset-based meta-analyses for the Edinburgh Postnatal Depression Scale and the Patient Health Questionnaire 9. RESULT: An online Clinical Information Aid produced from these meta-analyses makes instantly accessible a full set of test metric outputs at all cutoffs and crucially that at the patient's precise score cutoff. CONCLUSION: The optimality bias can be both established and eliminated.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.