Related Experiment Video
Updated: Mar 7, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Comparing Approaches to Optimize Cut-off Scores for Short Cognitive Screening Instruments in Mild Cognitive
Rónán O'Caoimh1,2, Yang Gao1, Anton Svendovski3
1Centre for Gerontology and Rehabilitation, University College Cork, St Finbarrs Hospital, Cork City, Ireland.
Background:
Although required to improve the usability of cognitive screening instruments (CSIs), the use of cut-off scores is controversial yet poorly researched.
Objective:
To explore cut-off scores for two short CSIs: the Standardized Mini-Mental State Examination (SMMSE) and Quick Mild Cognitive Impairment (Qmci) screen, describing adjustments in scores for diagnosis (MCI or dementia), age (≤, >75 years), and education (<, ≥12 years), comparing two methods: the maximal accuracy approach, derived from receiver operating characteristic curves, and Youden's Index.
Methods:
Pooled analysis of assessments from patients attending memory clinics in Canada between 1999-2010 : 766 with mild cognitive impairment (MCI) and 1,746 with dementia, and 875 normal controls.
Results:
The Qmci was more accurate than the SMMSE in differentiating controls from MCI or cognitive impairment (MCI and dementia). Employing the maximal accuracy approach, the optimal SMMSE cut-off for cognitive impairment was <28/30 (AUC 0.86, sensitivity 74%, specificity 88%) versus <63/100 for the Qmci (AUC 0.93, sensitivity 85%, specificity 85%). Using Youden's Index, the optimal SMMSE cut-off remained <28/30 but fell slightly to <62/100 for the Qmci (sensitivity 83%, specificity 87%). The optimal cut-off for MCI was <29/30 for the SMMSE and <67/100 for the Qmci, irrespective of technique. The maximal accuracy approach generally produced higher Qmci cut-offs than Youden's Index, both requiring adjustment for age and education. There were no clinically meaningful differences in SMMSE cut-off scores by age and education or method employed.
Conclusion:
Caution should be exercised selecting cut-offs as these differ by age, education, and method of derivation, with the extent of adjustment varying between CSIs.
