The Mini-Cog: receiver operating characteristics with expert and naïve raters

Background: As elderly populations grow, dementia detection in the community is increasingly needed. Existing screens are largely unused because of time and training requirements. We developed the Mini-Cog, a brief dementia screen with high sensitivity, specificity, and acceptability. Here we describe the development of its scoring algorithm, its receiver operating characteristics (ROC), and the generalizability of its clock drawing scoring system.

Sample and methods: A total of 249 multi-lingual older adults were examined. Scores on the three-item recall task and the clock drawing task (CDT-CERAD version) were combined to create an optimal algorithm. Receiver operating characteristics for seven alternatives were compared with those of the MMSE and the CASI using expert raters. To assess the CDT scoring generalizability, 20 naïve raters, without explicit instructions or prior CDT exposure, scored 80 randomly selected clocks as "normal" or "abnormal" (20 from each of four CERAD categories).

Results: An algorithm maximizing sensitivity and correct diagnosis was defined. Its ROC compared favorably with those of the MMSE and CASI. CDT concordance between naïve and trained raters was >98% for normal, moderately and severely impaired clocks, but lower (60%) for mildly impaired clocks. Recalculation of the Mini-Cog's performance, assuming that naïve raters would score all mildly impaired CDTs in the full sample as normal, retained high sensitivity (97%) and specificity (95%).

Conclusion: The Mini-Cog algorithm performs well with simple clock scoring techniques. The results suggest that the Mini-Cog may be used successfully by relatively untrained raters as a first-stage dementia screen. Further research is needed to characterize the Mini-Cog's utility when population dementia prevalences are low.