Using data from the ADNI, we have investigated acceleration in rates of whole-brain and hippocampal atrophy and ventricular enlargement, using 3-year follow-up data on controls and subjects with MCI and 2-year follow-up data for subjects with AD. We have assessed acceleration using all available data by using linear mixed models. The clinically defined groups of subjects—controls, MCI, and AD—showed the expected group differences in mean rates of atrophy at baseline: brain atrophy rates were ~2.4 times greater in the AD subjects than in the controls, ventricular enlargement was ~3 times greater, and hippocampal atrophy was ~4.3 times greater (). The MCI group rates fell almost exactly halfway between the AD and control means for each of the measures.
Hippocampal atrophy in subjects with MCI was estimated to accelerate by 0.22%/year2
= 0.037) on average. Furthermore, in the subjects with MCI who subsequently were given a diagnosis of AD, the acceleration was twice this at 0.50%/year2
= 0.003) and was very similar to that seen in the AD group (0.49%/year2
= 0.16). This suggests that the observed acceleration of hippocampal loss in MCI subjects is mainly driven by the MCI subjects that were observed to progress to clinical AD within 3 years. Although the estimated hippocampal acceleration in AD subjects was moderately large, there was no statistically significant evidence that it differed from zero. However, this may be because of reduced power due to the shorter period of follow-up (only 2 years as opposed to 3 for the MCI group) and/or greater heterogeneity in the AD group, e.g., a proportion of AD subjects may no longer be accelerating in hippocampal atrophy as suggested in the sigmoid model.5
As for ventricular expansion, the rates in MCI and AD subjects were estimated to accelerate at 0.27 and 0.88 mL/year2
Previous studies investigating the time course of hippocampal volume loss as a function of time from baseline have reported conflicting results.9,10,26
Using 0-, 6-, and 12-month magnetic resonance scans from the ADNI, hippocampal atrophy rates were estimated to accelerate by 26.5 mm3
(or 1.6%) in subjects with AD and 12.1 mm3
(or 0.6%) in subjects with MCI, but remained constant in controls.9
Another study found that hippocampal atrophy rate in controls (aged 49–85 years) accelerated from 30 mm3
/year to 124 mm3
/year in approximately 15 months,10
although the annual percentage change from the first to second follow-up was approximately 4%, which was similar to the rates in AD subjects reported by an ADNI study.9
Such acceleration may be driven by the older control subjects. In addition, accelerating hippocampal atrophy was found in the control, MCI, and AD subjects with AD-like CSF molecular profiles using 0-, 6-, and 12-month magnetic resonance scans from ADNI.11
However, using 0-, 6-, and 12-month magnetic resonance scans from the controls and MCI subjects in ADNI, the intercept of the regression line assuming no acceleration fitted to the volume loss of the hippocampus at 6 and 12 months was close to zero.26
This implies that the atrophy rates in controls and MCI subjects stay nearly the same over 1 year. Differences in results between studies regarding acceleration may be attributable to differences in the number of subjects, the number of time points, and the hippocampal segmentation protocol used. To address the latter, the “harmonization of hippocampal segmentation” project was recently set up to develop a harmonized protocol that aims to reduce the present heterogeneity in hippocampal segmentation protocols.27
Many studies have investigated the time course of brain tissue loss as a function of age or other characteristics.3,4,6–8,28,29
Using data from young adult familial AD patients, hippocampal atrophy was estimated to accelerate by 0.40%/year2
(95% CI 0.17, 0.63), which is similar to our estimate of 0.49%/year2
In the same study, the authors found evidence that whole-brain atrophy rates accelerated by 0.26%/year2
(95% CI 0.16, 0.37), an estimate that falls just outside the 95% CI we obtained in this study. This might be attributable to a more aggressive onset to the familial disease—perhaps consistent with an age at onset that is some 30 or 40 years earlier than the age at onset in late-onset sporadic AD in ADNI—or to the fact that the clinically defined AD (dementia thought to be due to AD) group in ADNI inevitably contains a proportion of subjects who do not have AD pathologically.
Recently it was hypothesized that the dynamics of cerebral tissue loss within individuals follows a nonlinear and sigmoidal trajectory.5
Several studies have attempted to fit sigmoidal models for the trajectory of cerebral tissue loss (e.g., hippocampus and cortical thickness) using cross-sectional data from ADNI.11,30,31
The sigmoidal model implies that atrophy rates within a subject initially increase from zero, remain constant for a period, and then eventually decrease to zero. In the current study, we modeled the trajectory of cerebral tissue loss in each disease group using longitudinal data from ADNI over a 2- or 3-year period. Our linear mixed model allowed for atrophy rates to vary between subjects, but assumed that any acceleration was the same for all subjects. Further analysis with longer follow-up will permit models to be fitted that allow acceleration (or deceleration) in rates to vary both within and between subjects, which may provide evidence regarding the validity of different models of progression.
Our analysis has a number of strengths and limitations. Statistical power to detect acceleration in atrophy rates is strongly affected by the length of time over which subjects are followed, which was relatively short, particularly for the AD subjects. Despite this, our estimates of acceleration have relatively narrow CIs. A proportion of subjects in each group had missing scans at one or more follow-up time points, causing a loss of information and potentially biasing our results. However, we have used a statistical modeling technique that accommodates such missingness under the missing at random assumption. We have used a consistent atrophy measurement technique that treats each time point equally, which avoids bias toward any particular time point. Our analysis has used data from a large number of subjects with standardized and consistent protocols used for imaging and diagnosis.
Overall, we have found evidence of acceleration in hippocampal atrophy rates in subjects with MCI in the ADNI. This acceleration seems to be driven by those subjects with MCI who subsequently received a clinical diagnosis of AD dementia within 3 years. The rate of acceleration of hippocampal atrophy is relatively slow compared with the difference in atrophy rates in controls and clinical AD. This suggests a long period of transition to the pathologic losses seen in clinical AD, provided acceleration in rates is constant. Rates of ventricular enlargement were found to increase over time in both MCI and AD subjects. Our findings give further motivation for future longitudinal studies that will enable investigation of how atrophy rates evolve throughout the course of AD.