Alzheimer's disease: rapid and slow progression

Abstract

The variability in the progression of Alzheimer's disease (AD) across patients has made identification of disease-delaying treatments difficult. Quantitative analysis of this variability has important implications in understanding the pathophysiology of AD and identifying disease-delaying treatments. The functional assessment staging (FAST) procedure characterizes seven stages in the course of AD from normal ageing to severe dementia. The present study applied statistical methods to analyse FAST stage durations from a dataset of 648 AD patients. These methods uncovered two distinct types of disease progression, characterized by different mean progression rates. We identified two separate distributions of FAST stage progression times differing by up to 2 years in mean duration within each stage. These results further indicate that if a patient progresses rapidly through a given FAST stage, then their further progression is also likely to be rapid. These findings support the hypothesis that progression of AD can occur via two different pathophysiological mechanisms that lead to distinct average rates of decline.

Figure 1.

Some statistics of the AD patient dataset. (a) A histogram of the numbers of patients with two, three and more assessments. (b) A histogram of intervisit times in the patient sample.

Figure 2.

A histogram of transition times for the AD patients. For each transition class i → j, the times elapsed between the two visits for patients first diagnosed with FAST stage i and subsequently with FAST stage j are shown. Each histogram includes (i) only first and last assessments (white) and (ii) all assessments (black).

Figure 3.

Illustration using genetic algorithms to classify six patients (numbers 1–6) into rapid (R) or slow (S) progressors. The initial number of partitions (organisms), N_I, is set to be a large number (in this case, the four partitions shown in boxes a, b, c, d). The fitness, f, for each partition is then computed. The fitness computation in our algorithm is described in the text and is derived from the FAST staging data of the patients. In the example in this figure, we used arbitrary fitness values for illustration purposes. The partitions with the highest fitness are more likely to be selected as the parents who will mate (second row of figure). In this example, the parents who will mate are partitions (a and b) and (a and c). If both parents have classified a given patient into the same progression rate group, then that patient's class remains unchanged in the progeny (circled numbers). To finish producing the progeny, the remaining patients for each category, R and S, are chosen randomly from the corresponding categories of the two parents. The progeny created in this way will constitute the next generation. The process ends when all, or nearly all, patients are classified the same by all the organisms. The final set of progeny is those with the highest fitness. For each FAST stage, this results in rapid and slow progressor groups with optimal separation of their durations.

Figure 4.

Modelling rapid and slow progressors based on the simulated underlying probability distribution of FAST stage durations. The separation algorithm was applied to artificial datasets, where it was a priori assumed that each patient was either a rapid or a slow progressor, and for both types the lengths of the four stages of the ‘disease’ were drawn from the corresponding distributions. The separation algorithm was run 200 times. For each run, each patient was classified as a ‘rapid’ or a ‘slow’ progressor. From these runs, the probability of each patient being a rapid progressor, p(rapid), was computed. The results are presented as histograms that show the probability of being a rapid progressor on the x-axis (each bar has a probability range of 0.04), and the number of such cases on the y-axis. The black and grey horizontal bars at the top of each panel represent the simulated overlap of the probability distributions of FAST stage durations between rapid and slow progressors.

Figure 5.

Result of sorting the FAST staging dataset using our separation algorithm. The same algorithm as in figure 4 was applied to a longitudinal dataset of 648 AD patients. The diagram shows that most of the patients are classified with certainty as rapid or slow progressors (only the data for transition classes 4 → 6, 4 → 7 and 5 → 7 were included in the histogram; other transitions will also be classified as rapid or slow with certainty). Inset: Using the empirical probabilities p(rapid) obtained by the separation algorithm, we computed the spread of the mean stage durations for FAST stages 4–7; this result demonstrates the statistical significance of the separation result.