Objectives: To study how well the scoring on each item of the MMSE relates to the sum-score when the purpose is to identify persons with cognitive impairment, and to identify an equally effective subset of MMSE items for predicting cognitive impairment.
Design: Retrospective survey of MMSE data for 850 elderly.
Setting: A variety of clinical settings.
Participants: Mean age 82 years (range 54 to 99), 74% women. The subjects were of three different categories: geriatric in-patients, patients living under supervision, and elderly people living independently at home.
Results: Five of the binomial ("State," "Town," "Name a pencil," "Name a watch," "Read and obey") and one of the polychotomous MMSE variables ("Learn three words and repeat immediately") had low sensitivity and gave high percentages of misclassifications versus the sumscore dichotomized at the cut-point 23/24. Univariate logistic regression indicated that the three remaining polychotomous variables ("Spell backwards," "Recall three words," and "Three-stage command") can be scored binomially. Two factors were identified on factor analysis. Logistic regression analysis showed that 12 of the original 20 items predicted the sumscore dichotomized at 23/24 with only 3% misclassifications. Validation against the psychogeriatrician's diagnosis showed that this 12-items MMSE derivative performs as well as the full MMSE.
Conclusions: Six of the 20 MMSE variables perform poorly regarding sensitivity and misclassifications versus the sumscore at cut-point 23/24. Two additional items did not contribute to the prediction of a low/high sumscore. The remaining 12 MMSE items can all be scored binomially and produce a sumscore which is equally as effective as the sumscore of the full MMSE when the purpose is to identify elderly patients with cognitive impairment.