Related Experiment Video
Updated: Mar 16, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
Interpreting "I don't know" use by persons living with dementia in Mini-Mental State Examinations
Ashley M Hesson1, Heike Pichler2
1Department of Obstetrics and Gynecology, University of Michigan, 1500 E. Medical Center Drive, L3211 Women's, SPC 5276, Ann Arbor, MI 48109-5276, USA.
Objective:
We investigate dementia patients' use of "I don't know" (IDK) in Mini-Mental State Exams (MMSEs) using objective linguistic indicators to differentiate IDK signalling lack of knowledge (LOK) from IDK used to hedge responses, affect exam progression etc. We hypothesize that increased proportional use of LOK-IDK correlates with worsening dementia severity.
Methods:
189 IDK tokens were extracted from 72 MMSE interactions and coded for linguistic/social characteristics. A data-driven, discourse position/relation-based functional taxonomy for IDK in MMSE was developed and the resulting functional distribution was subjected to multiple logistic regression.
Results:
Use of LOK-IDK (vs. non-LOK-IDK) is significantly correlated (p=0.01) with clinicians' subjective ratings of patients' dementia as 'severe' vs. 'mild'/'moderate', indicating that objective sociolinguistic criteria approximate physician judgments. 92% of 'severe' patients' IDKs signalled LOK, compared to only 68% of 'mild' patients', suggesting that uncritical interpretation of IDK as signalling LOK would result in 8-32% of IDK responses being mis-scored.
Conclusion:
LOK and non-LOK uses distinguished on the basis of reliable, objective usage patterns are differentially distributed among dementia severity groups.
Practice Implications:
LOK-IDK serves as a supplemental indicator of dementia severity. Correct interpretation may improve diagnostic accuracy and allow clinicians to respond supportively during cognitive assessment.
