Tracking Dementia by the IMC and the MMSE

To the Editor: Galasko and colleagues1 review interesting data from several reports monitoring progression in Alzheimer's disease (AD) patients. The rates of change for three of these scales are of particular relevance to pharmacological treatment studies as well as for monitoring illness progression: the Blessed Information Memory Concentration Test (IMC), the Folstein Mini-Mental State Examination (MMSE), and the Alzheimer's Disease Assessment Scale (ADAS). Galasko and colleagues, however, chose to characterize the size of the variances (ie, the standard deviations) on both the IMC and the MMSE as “fairly large, roughly equal to the means [change] indicating substantial individual variability.” They, correctly, list several reasons for this variability, including the fact that neither AD nor performance tests progress linearly, and that test-retest variability adds to the uncertainty (ie, variance) of determining the true annual rate of change. The purpose of this letter is to point out that a standard deviation of a change score that is roughly equal to the mean change is a rather “small,” standard deviation, and indicates a rather large effect for a measure. The accompanying table, adapted from the authors' Table 2, provides estimates of effect sizes, d's, for the three tests, expressing annualized mean rate of change as a proportion of its standard deviation (Table 1). The effect sizes for the IMC and the MMSE from the several studies were then combined to obtain an overall effect size and a 95% confidence interval for each test. Even without the statistical treatment, it is apparent that both the effect sizes and associated correlation coefficients for the rates of change of the IMC and the ADAS are markedly greater than for the MMSE. Expressed in other statistical terms, the IMC accounts for, or “explains,” about 3.3 times the variance accounted for by the MMSE. No matter the actual sources of variability, however, for many practical purposes, such as assessing change in a clinical trial, the variances of these measures are fairly small in comparison to the observed magnitude of the change. For example, using the overall effect sizes calculated in the table, if a placebo-controlled clinical trial were planned to assess a drug that is expected to halve the yearly rate of decline of AD patients, then only 96 patients would need to be randomly assigned to the drug or placebo group in order to have an 80% chance of detecting such a change if the IMC were used as the outcome instrument.2 By comparison, if the MMSE were used to measure outcome under the same assumptions, then 376 patients would need to be randomly assigned. The differences in sensitivity between the IMC and the MMSE are dramatic. Similarly, a practicing clinician, using the IMC, would begin to appreciate significant clinical changes with as few as nine AD patients in his or her clinical practice over the course of a year. All other things being equal, it appears that the IMC, because of its empirically observed characteristics, might be more sensitive to change and more useful as an outcome measure than the more commonly applied MMSE. Versions of the IMC have been used in European dementia clinical trials. These studies, for example, a placebo-controlled trial of deprenyl,3 show surprisingly large effects when the IMC is used to assess change. An issue, in clinical trials, is whether physicians would value a two-to-four point change on the IMC as sufficiently clinically meaningful. Editor's note:—The above letter was referred to the authors of the original paper and their reply follows.