In this study, we present three lines of evidence indicating that AD proceeds more aggressively among younger elderly than among older elderly individuals, leading to a blurring of the distinction between AD and HC among the oldest old. First, we found that annual brain atrophy rates for AD and MCI individuals decrease with increasing baseline age, while atrophy rates for clinically normal older individuals remain constant or exhibit a slight increase with age. Second, we found that baseline CSF biomarker levels indicated greater disease burden in younger than in older MCI and AD patients, while disease burden increased with age among HCs. Third, we found that AD patients showed reduced rates of cognitive decline with increasing baseline age, despite uniformity of cognitive impairment at study entry. These findings have important implications for detection of AD among the oldest old, for the design of clinical trials of potentially disease-modifying therapies, and for biomarker and clinical disease trajectories.
All ROIs examined showed a decrease in longitudinal atrophy rate for MCI and AD individuals with increasing age at study entry. This decrease was most apparent in association cortices that tend to be affected later in the disease process, such as the inferior parietal lobule, middle temporal gyrus, and the retrosplenial cortex. The decrease in rate of atrophy with age in patient groups, combined with a trend toward increasing atrophy rates with age for HCs, resulted in a pattern of convergence of atrophy rates in association cortices across all diagnostic groups, with close convergence around 85 years of age, suggesting that in these brain regions neuronal number in patients is increasingly preserved with advancing baseline age 
. However, the ROIs impacted by NFTs in earliest stages of the disorder (Braak stages I–II, transentorhinal, and stage III, limbic 
) continued to show significant differences in rates of decline for patients and older controls at more advanced ages, although differences were smaller than those observed at younger ages. Thus the power to discriminate MCI or AD individuals from HCs is differentially retained across ROIs, with the medial temporal ROIs that are particularly vulnerable to early neurofibrillary pathology retaining the strongest discriminative ability with advancing age.
In cross-sectional analyses, levels of disease burden as measured by CSF biomarkers decreased with increased age at study entry for MCI and AD participants, but increased with age for HC participants. The difference in the age-dependence of these biomarkers between patients and controls was particularly strong for ptau, leading to full convergence of ptau levels across patient groups at 85 years. The increase in ptau levels with age in HCs likely reflects increased burden of AD-specific neurofibrillary pathology 
, and may underlie the increase with age in rate of clinical decline 
and atrophy rate for the entorhinal cortex and hippocampus 
observed here for HCs.
Consistent with the structural MRI results, rates of longitudinal cognitive decline also showed convergence between patient and HC groups with increasing age at study entry. AD patients showed significant decrease in rate of decline on MMSE and a trend towards reduced rates of decline on ADAS-Cog with increased age, decreasing the difference in rates of decline between AD and HCs with increasing age. Due to uniform enrollment criteria across age, however, baseline cognitive measures did not differ with age within diagnostic groups, and showed constant separation among groups across age. This indicates that within each diagnostic group, individuals were at a uniform cognitive stage irrespective of baseline age. The differences observed in baseline CSF biomarkers and in rates of cognitive and structural change with age therefore do not stem from differences in clinical severity with age at the time of enrollment, but instead appear to reflect a decrease in rate of disease progression with baseline age. Within-cohort rates of decline for CDR-SB, however, were independent of age. CDR-SB measures global clinical or functional change, and may therefore not be as sensitive as ADAS-Cog to small declines in cognition 
. A similar result was found in a recent study of age and rate of cognitive decline in AD 
, which also observed that older age at baseline was significantly associated with a slower rate of decline in ADAS-Cog 11, ADAS-memory, and MMSE. This decrease in rate of cognitive decline is consistent with our results showing decrease in 3-year rates of progression from MCI to dementia with advancing age: close to 50% of MCI participants aged 65 years developed dementia, whereas only about 25% of those aged 85 years did.
These findings showing a decrease with advancing age in multiple measures of neurodegeneration between AD patient cohorts and HCs are in agreement with earlier research showing that the correlation between dementia severity and NFT burden decreases with advancing age 
, and a more recent study assessing prevalence of moderate or severe pathological lesions in participants aged 69 to 103 years, which found attenuation with advancing age in the association between dementia and the densities of both neuritic plaques and NFTs in all examined ROIs 
. In contrast, however, the latter study also found that dichotomized measures of neocortical and hippocampal atrophy allowed for distinguishing individuals with dementia from those without dementia, regardless of age, in agreement with the retained discriminative ability of hippocampus and entorhinal cortex measures shown here. Our findings are also in agreement with a recent cross-sectional analysis showing that the AD-related cognitive and structural MRI changes seen in AD patients aged 60–75 years are less salient in patients aged 80–91 years 
, and a recent longitudinal MRI analysis showing that rates of whole brain atrophy were greater in younger than older participants with amnestic MCI 
Our results showing smaller baseline structure size in older as compared with younger participants, modeled using linear fits with age, are in broad agreement with earlier manual and automated volumetry analyses 
. In the latter study 
, covering the seventh through tenth decades in age, a generalized additive model (GAM) 
was used, and nonlinear trends in baseline structure size as a function of age were reported. In the later decades, in particular, the cross-sectionally assessed atrophy accelerated or decelerated with age, depending on ROI and diagnostic group. It should be noted that cross-sectional analyses are not the most accurate approach to estimating longitudinal rates of change. It is also important to note that with GAMs the uncertainty in estimated structure size as a function of age can grow substantially from the middle of the age range toward the younger and older extremes in the age range modeled–which is often where the estimated nonlinearities tend to be most pronounced. This increased uncertainty may help account for discrepancies with other findings. For example, cross-sectional analysis comparing manual with voxel-based morphometry measures of hippocampal volume in healthy individuals 
indicates acceleration of hippocampal atrophy with age in HCs. This is in agreement with our longitudinal results and in broad agreement with earlier results from manual longitudinal volumetry 
, but is at variance with the deceleration of hippocampal atrophy with age in HCs resulting from the GAM analysis of cross-sectional data 
Our findings of reduced rates of clinical and morphometric decline with age in patients with AD may help shed light on the contradictory results that have been reported for incidence of AD dementia for the oldest old, aged 90 years and older. For example, the Cache County Study 
and an autopsy study 
found that incidence rates of AD decrease, beginning in the early 90 s. In contrast, the 90+ Study examining all-cause dementia, including 60% of participants diagnosed with AD, had a larger contingent of oldest participants, and found that dementia incidence continued to increase with age 
. Although our results do not directly address the incidence of AD with age, our finding that rate of cognitive decline decreases with baseline age suggests that for the oldest old, clinical detection of AD, which relies on progressive decline in cognitive ability, may be more difficult. Reduction in CSF biomarkers of disease burden, and in atrophic changes in brain structure, with age suggests that the newly revised criteria for diagnosis of AD in research settings, which recommends incorporation of these biomarkers 
, may not help overcome this problem. It should also be noted, more broadly, that since incidence is the number of new cases in a population in a given time period, e.g. a year, non-constant rates of decline with age are likely to play a role in the incidence at different ages of clinically diagnosed AD.
The blurring of the distinction between HCs and patients with MCI or AD with advanced age has important implications for clinical trial design. As shown in and , all potential outcome measures evaluated here provide significant power for detecting disease-modifying therapeutic effects for MCI and mild AD participants aged 65–75, with the previously reported advantage for structural measures, such as entorhinal cortex and hippocampus, over clinical measures 
. However, power is dramatically reduced with increasing age of the study sample. Due to the exponential rise in prevalence of AD with age (through approximately 85 years 
), older individuals will increasingly be more available for clinical trials as compared with younger individuals. Yet, detecting a decrease in atrophy rate, or in rate of clinical decline, due to a disease-modifying therapy becomes increasingly difficult the older the study cohort. Thus, the extent to which older individuals are represented in the study sample could profoundly affect the power for detecting a therapeutic effect. In particular, a small but significant disease-modifying effect, which could significantly reduce the global burden of the disease 
, might be found in younger cohorts but would likely not be found in older cohorts. Given demographic trends, there is an urgent need to develop disease-modifying therapies to avert what will otherwise be an AD epidemic 
. Since the power to detect therapeutic effects can be reduced dramatically in older cohorts, it is of immediate importance to consider fully cohort age in drug discovery.
With regard to disease trajectories, the observed reduction in rates of decline with advancing age allows for two distinct, plausible scenarios. In the first scenario, a cognitively healthy individual begins to experience a slightly elevated rate of decline at an early point in life, so that over a prolonged period, tissue loss gradually accumulates and cognitive function gradually declines. This slow course continues into advanced ages. At some point, despite the gradual nature of change, enough tissue loss and cognitive impairment has ensued to enable a physician to diagnosis the disorder.
In the second scenario, a cognitively healthy older individual initially experiences the same rates of structural and cognitive change as those of other cognitively healthy individuals of the same age, but the individual begins to experience a more rapid course of decline as symptoms develop and worsen. Rate of decline then slows at some point, so that at an advanced age, the rate of decline again approaches the rate observed for healthy older individuals of the same age. In this scenario, the biomarker trajectory follows a sigmoidal course. Despite the popularity of sigmoidal curves for describing biomarker trajectories in AD 
, and the development of mathematical models of neuron death kinetics where neuron death is considered not to arise from cumulative damage 
, there is no conclusive evidence demonstrating a slowing in rate of brain atrophy over time within individual AD patients. Furthermore, slowing in rate of neuronal loss from age and disease progression might not be biologically plausible: arguments from protein homeostasis indicate instead that cell death is more likely to accelerate than to slow with disease progression 
A recent investigation 
of the shapes of AD biomarker trajectories examined whether, as suggested by sigmoidal models, hippocampal atrophy rate slowed with increasing disease severity in four patient cohorts. Employing a liberal significance threshold (p-value
0.1), and restricting the sample to amyloid-positive participants, evidence of slowing of the rate of volume loss with increasing disease severity, as assessed with MMSE score, was found in three cohorts when cross-sectional data were analyzed but not when longitudinal data were analyzed. In the fourth cohort the reverse occurred: evidence of slowing of the rate of volume loss with increasing disease severity was found when longitudinal data were analyzed but not when cross-sectional data were analyzed. These inconsistencies do not strongly support the sigmoidal model for AD-related atrophy. Furthermore, when the sample was not restricted to amyloid-positive participants, cross-sectional analyses consistently showed an increase in hippocampal atrophy with increasing disease severity, and longitudinal analyses consistently showed an increase in hippocampal atrophy rate with increasing disease severity–results that do not support the contention that atrophy rates eventually decelerate with advancing disease severity.
Another possible explanation of differences in rates of decline with age observed here is differential dropout, in which older patients with more severe decline dropped out of the study at a higher rate than younger individuals with more severe decline. However, evidence suggests that such differential dropout is unlikely to have affected the observed results. Older participants were not more likely than younger participants to drop out after the baseline session, and there was no difference in symptom severity between those who dropped out after baseline versus those who completed one or more follow-ups. However, differential recruitment into the study may contribute to the observed results. Participation in a trial such as ADNI requires a large commitment of time and effort. It is possible that for the oldest old patients with AD, if they were experiencing a rapid course of decline, they or their caregivers may have been less willing to participate in a burdensome, non-treatment study than those with milder decline. Such differential enrollment, where younger but not older rapidly declining individuals enrolled in the study seems unlikely given the systematic decrease in rates of decline with increasing age for both MCI and AD cohorts.
Limitations of this study include the highly select nature of the study sample: individuals were required to be generally healthy with no evidence of comorbid disorders that could affect study participation, such as depression or vascular dementia. A second limitation is the lack of histopathological verification of AD pathology. It is possible that many participants suffer from subclinical cerebrovascular disease, and the contribution of cerebrovascular disease to dementia symptoms may be greater among the very old. A recent neuropathological study 
, in addition to finding attenuation in the association between dementia and NFT burden in nonagenarians, also found an increase in vascular burden in mixed pathologies in people over 90 years. This increases the possibility of misdiagnosis, particularly since microvascular pathology and AD are synergistic in the development of dementia in old age 
. Furthermore, and somewhat circularly, the convergent patterns in rates of decline might themselves contribute to increased potential for misdiagnosis at older ages. The small number of AD individuals over the age of 85 years is a further limitation of the study.
Despite these limitations, this study suggests that the association between neuropathological markers and dementia attenuates with age and needs to be taken into account in models of AD. The degree of attenuation is remarkably uniform across diverse markers–CSF AD-related protein densities, atrophy rates in multiple brain regions, and rates of clinical decline–all showing strong convergence patterns across diagnostic groups at older ages, though it is significant that atrophy rates in the hippocampus and particularly the entorhinal cortex indicate slower convergence. As a result, methods for early disease detection and assessment of therapeutic interventions cannot be applied uniformly across the entire elderly age spectrum. Given demographic trends, in particular the rapid growth in the proportion of very old individuals, greater emphasis needs to be placed on further elucidating the effects of age on the disease process to better prepare for the diagnosis, care, and treatment of the oldest old.