An Operational Definition of ‘Abnormal Cognition’ to Optimize the Prediction of Progression to Dementia: What Are Optimal Cut-Off Points for Univariate and Multivariate Normative Comparisons?

Abstract

Background: In neuropsychology and neurology, there is no consensus on the definition of abnormal cognition. Objective: To operationally define ‘abnormal cognition’ for optimally predicting progression to dementia in a memory clinic sample, and to test whether multivariate profile analysis of cognitive test results improves this prediction compared to standard clinical evaluation. Methods: We used longitudinal data from 835 non-demented patients of the Amsterdam Dementia Cohort. For 10 cognitive measures at baseline, we determined which number of abnormal tests and which magnitude of score deviations best predicted progression. Results: Predictive ability for progression to dementia of one, two, and three abnormal test scores out of 10 is highly similar (Cox hazard ratios: 3.7–4.1) provided cut-off values are adapted appropriately. Cut-offs have to be less stringent if the number of abnormal tests required increases: the optimal cut-off is z < –1.45 when one deviating score is required, z < –1.15 when two abnormal tests are required, and z < –0.70 when three abnormal tests are required. The profile analysis has similar predictive ability at the cut-off of p < 0.22 (hazard ratio 3.8). A likelihood ratio test showed that this analysis improves prediction of progression to dementia when added to standard clinical evaluation (p < 0.001). Conclusion: Abnormal cognition may be defined as one, two, or three abnormal test scores out of 10 if the magnitude of score deviations is adapted accordingly. An abnormal score profile predicts decline to dementia equally well, and improves the prediction when used complimentary to standard clinical evaluation.