What is test accuracy? Comparing unitary accuracy metrics for cognitive screening instruments.

Abstract

Aim: To examine four different accuracy metrics for assessment of commonly used cognitive screening instruments: correct classification accuracy, area under the receiver operating characteristic curve, F measure (F) or F1 score and Matthews correlation coefficient (MCC). Methods: Raw data were extracted from test accuracy studies of Mini-Mental State Examination. Montreal Cognitive Assessment, Mini-Addenbrooke's Cognitive Examination, Six-item Cognitive Impairment Test, informant AD8 and Free-Cog, and used to calculate the accuracy measures. Results: Each metric resulted in similar ordering of the screening instruments for diagnosis of both dementia and mild cognitive impairment. Area under the receiver operating characteristic curve gave the highest (most optimistic) and MCC the lowest (most pessimistic) accuracy value for each test examined, with correct classification accuracy and F falling between. Conclusion: All the accuracy measures examined have potential shortcomings. None can be recommended as the definitive unitary outcome measure for test accuracy studies. However, MCC has theoretical advantages and might be more widely adopted.