On the Importance of Longitudinal Research in Alzheimer's Disease

The paper "Monitoring Progression in Alzheimer's Disease" by Drs. Galasko, Corey-Brown, and Thal1 in the current issue of this journal brings together many prominent issues in longitudinal study of Alzheimer's Disease (AD). Such longitudinal studies are important for several reasons including the increasing age of the population, the rising number of people with Alzheimer's disease, the need to document prognosis, and we may hope, the need to monitor treatment effects. In addition, AD imposes substantial economic costs upon families and society.2 Despite the fact that researchers have studied AD for a long period of time, longitudinal research has only recently begun to proliferate. Galasko et al provide a useful compendium of these relevant studies and, perhaps more importantly, point out several limitations of research in this area. Although cross-sectional studies can provide valuable information, it is only through monitoring the changes of Alzheimer's patients that an accurate picture of the course of the disease may be obtained. Longitudinal investigations are vital for the elucidation of the nature of AD. One of the advantages of longitudinal work is that it should eventually permit researchers to distinguish between stages and subtypes of AD. In a cross-sectional study in which measurements are obtained at a single point in time, one cannot know whether any observed relations result from the inclusion of patients who are at different stages of their disease (an almost certain occurrence in any large-scale study) or from characteristics of different subtypes of the disease. For example, if aphasia develops late in the course of the disease, comparing early and late cases might give rise to the false idea that there were aphasic and non-aphasic subtypes of AD, when really one was looking at preaphasic and aphasic stages of the disease. Indeed, it is the rate at which abilities deteriorate and the points at which deterioration begins that allow subtypes to be defined. One illustration of the different answers that may be obtained from cross-sectional and longitudinal approaches was provided by Tinklenberg et al,3 who examined the patterns of relations among the items of the Mini-Mental State Examination (MMSE)4 in both cross-sectional and longitudinal data. Specifically, Tinklenberg et al replicated a previously proposed cross-sectional factor structure of the MMSE5 but found that the factor structure of the average rates of change of the items was quite different. The cross-sectional data were best described with a two-factor solution whereas the longitudinal data were best described by a five-factor solution. The five underlying factors of the average rates of change were related to orientation, following commands, language expression, language repetition, and recall. Examining the rates of decline of the factors provides a clearer picture of rate of progression than merely examining the rate of change of the total MMSE scores. The patients in Tinklenberg et als study exhibited an annual decline on the MMSE of approximately 10%, which is in agreement with other studies.6–8 However, as illustrated in Figure 1, the average rates of decline varied for the different factors. . The average rate of decline per year, expressed as a percentage of each factor's maximum possible score, for patients in the Tinklenberg et al3 study. The endpoints of the lines represent the average decline of the patients across all of their years of participation in the study. The lines have been placed at the same origin for illustrative purposes. The findings of Tinklenberg et al bear on another point made by Galasko et al: most studies of AD use "global" measures, such as the MMSE, which provide rather coarse estimates of the overall rate of decline. Fractionating decline among the factors proposed by Tinklenberg et al does not provide the definitive picture of AD because the MMSE was designed to be a brief screening test and not a thorough neuropsychological battery. As Galasko et al suggest, decline in AD would be better assessed with more precise measures. As Galasko et al point out, perhaps the most important problem facing longitudinal research relates to floor effects which occur when decline is no longer detected with the measurement instrument. When floor effects occur, there are two possible explanations: (1) no decline is occurring and/or (2) the scale employed is not sensitive enough to detect decline. Regardless of the source of floor effects, their presence can have dramatic consequences for estimations of the rates of decline. Inclusion of such flat data results in the underestimation of the rate of decline. The extent of this problem varies from patient to patient, depending upon whether a floor effect is present and the extent of time over which it occurs. We have recently developed an analytic model that allows researchers to account for floor effects as well as ceiling effects.9 In this model, which is called the trilinear model (see Figure 2), decline in AD is described as proceeding through three periods: an initial period of stability in which no perceptible decline occurs, a period of decline, and a final period of stability. One advantage of using the trilinear model to conceptualize data is that it excludes both ceiling and floor effects and thus allows a more accurate estimate of the rate of decline. Thus, if patients are at different stages of AD, decline can still be assessed without possible contamination from floor effects. It is important to realize that, although the trilinear model may not be the "definitive" model of decline in Alzheimer's disease, it is definitely superior to the popular approach of assessing the average rate of decline across all data collected. . An illustration of the trilinear model. Although the model allows estimation of other parameters, only the period of decline is designated in this graph. Using these analytic techniques we may soon be able to quantify many important phenomena relating to decline in AD. For example, much controversy recently surrounded the evaluation of tetrahydroaminoacridine (THA; Tacrine) by the Food and Drug Administration in terms of its ability to alter the rate of decline of AD patients. It was suggested that the drug might "give 6 more months of useful life" to patients. But how does one quantify useful life? Such quantification requires knowledge of the stage of disease that a patient is in. If the result were to prolong the overall duration of the disease (see Figure 3a), it may primarily prolong the end-stage, extremely debilitated phase of AD; if, on the other hand, it were to prolong periods where function was still high, and hopefully above the level requiring nursing home care, the drug might have both clinical and economic benefits (see Figure 3b). Unfortunately, the available longitudinal data pertaining to THA do not permit one to distinguish between these two possibilities. . An illustration of two potential drug effects on rate of decline of AD. Treatment A affects overall decline and prolongs length of illness. Treatment B does not prolong illness but increases time at higher levels of functional ability. Ultimately, longitudinal research in AD requires better ways of quantifying decline. One course of action is to "step down" to easier versions of tests as a patient's abilities deteriorate. For example, it has been suggested that aphasia could be measured through the course of Alzheimer's disease by developing versions of aphasia tests of varying degrees of difficulty. However, serious problems can arise when one attempts to chart the course of decline across scores obtained from different tests. One likely path suggested by Galasko et al is to pursue the development of new tests that allow the assessment of abilities throughout the course of AD. Surely we will see more work along these lines in the future.