Fitting the epidemiology and neuropathology of the early stages of Alzheimer’s disease to prevent dementia

One of the main findings of this research is that one-half of the population older than 60 years of age in a developed country has some degree of Aβ deposition in the brain. This result agrees with those from population-based autopsy studies, such as the Medical Research Council Cognitive Function and Aging Study that showed 47% of nondemented subjects older than 70 years had mild, moderate or severe densities of neuritic plaques [29]. The enormous size of such a group of affected individuals is highlighted by the fact that it would encompass 15.2% of the total Spanish population (46,148,700 persons). Since the lifetime risk of AD at age 60 years is 19%, an intervention program targeting the 50% of subjects with Aβ deposits would include treating 68% of that population who would never reach the dementia stage [30]. With use of both neuropathological and clinicoepidemiological data, discrete event simulation modeling has provided an estimate of the time lapse between the earliest preclinical changes and the onset of dementia. This interval is usually considered to range from 10 to 20 years [5], but according to the model may be more than 30 years. The findings of both high prevalence and prolonged asymptomatic periods have important public health implications in the study of dementia, particularly in the prevention of AD. Salomon and colleagues have discussed the relevance for prevention plans based on two different concepts of AD onset [30]; that is, whether the starting point relies on pathological changes only [5-7] or on specific clinical manifestations [4]. Consistent with the definition of AD onset, the same preventive intervention can be labeled as primary when dealing with clinical symptoms only or as secondary if prior pathological changes are considered. In relation to our use of Aβ deposition in the model as the marker of the preclinical stage, we understand that primary prevention interventions should be applied before the onset of pathological changes by promoting initiation and maintenance of good health and eliminating or modifying potential risk conditions, if these are ever ascertained [30,31]. We believe secondary prevention would target subjects at the preclinical phase to avoid progression to overt clinical AD and this objective could be hypothetically achieved by risk-factor intervention or disease-modifying therapies. In turn, interventions in the prodromal stage would qualify as tertiary prevention to target the delay or avoidance of dementia onset in the presence of symptoms. In keeping with these definitions and our results, the size of the population targets would include the whole nondemented population for primary prevention, 23.3% of that population older than 40 years for secondary prevention, and 14.2% older than 60 years for tertiary programs. The size of these target populations constrains the viability of an ideal public health perspective to reduce AD prevalence based on a scalable approach, with interventions first targeting risk factors and then pharmacologic treatment in a highly selected group of individuals [9].

Much information has been gathered regarding the role of vascular risk factors and certain lifestyle conditions, such as low education, diet or sedentary behavior, and their association with dementia in general and AD in particular [32]. However, at present, no causal relationship has been demonstrated, and there is no evidence that the reduction of risk factors directly results in decreased incidence of either biomarker positivity or clinical AD [11]. Prevention strategies with anti-amyloid therapies have been proposed for asymptomatic individuals with the genetic and sporadic forms of AD [33]. However, our findings suggest that pharmacologic interventions in subjects with preclinical sporadic AD are hardly feasible. First, screening programs to detect amyloid based on positron emission tomography are so far unaffordable, and those based on lumbar puncture are invasive and need further standardization. Second, even if a more accessible biomarker becomes available and a therapy shows sufficient efficacy and safety, its application would target the 50% of subjects older than 60 years who have amyloid deposits, 62% of whom would never reach the dementia stage [34]. Overdiagnosis is a concept borrowed from cancer-screening evaluation that can also be applied to AD research; in this situation, overdiagnosis describes those individuals found to have AD by the presence of positive biomarkers but who will never have symptoms [35]. The risk of falling into the big prostate mistake has already been highlighted [36].

Tertiary programs might hypothetically treat symptomatic subjects with MCI due to AD to delay or prevent progression to the dementia stage. According to our data, it is still difficult to take a different approach for people with MCI. Subjects with MCI are at a significantly increased risk for the development of dementia associated with AD [37], and therefore represent a target population for preventive strategies. In our model, if the intervention in 2010 had been limited to subjects older than 60 years of age in the prodromal stage (MCI with positive AD biomarkers), the size of the target population would have been 14.2% (1,419,000 individuals). However, to make the diagnosis in this population, subjects with general MCI or MCI of any type should be screened with the appropriate AD markers, but they are not yet widely available. The size of the population with general MCI limits the generalized use of biomarkers or therapies and shows the need for more specificity in identification of the target population. Identifying those subjects with MCI who have a high probability of biomarker positivity would help to select those at high risk of rapid conversion to dementia. Improved predictive values at early diagnosis would result in fewer treated individuals who otherwise would never have had dementia. It is true that this approach can be only speculative before a disease-modifying treatment is available, but subjects with MCI and a positive biomarker could well be targeted in clinical trials with the already approved, cheaper cholinesterase inhibitors in an attempt to delay onset of dementia [38]. The definition of MCI from clinical studies has determined the super-selection of patient groups at high risk of dementia [39], but direct application of such data to population-based programs may not be completely appropriate. On one hand, the clinical course of MCI is not always progressive. On the other, not all subjects in whom dementia develops have previously fulfilled the diagnostic criteria for MCI [39].

One of the strengths of our work is the combination of the two competitive risks that determine the epidemiology of AD: time to onset of dementia and time until death by other causes. The inclusion of age in the mathematical functions and the thorough validation process (see Additional file 1) shows that our AD model has not ignored the ‘elephant in the room’ analogy described by Brayne in her seminal paper [40]. Our approach to AD is a view extending over the patients’ lifetime and incorporating the long preclinical stage that can only be characterized by progressive neuropathological changes. We also acknowledge some limitations in our approach. First, modeling findings should always be interpreted with caution, because they are based on a mathematical representation of the disease and not directly on empirical data. However, our representation of the natural history of AD is robust because it relies on an integrative framework that includes clinical, neuropathological, demographic and epidemiological elements. Second, we did not address the current multifactorial expression of the course of AD in late life in terms of cognitive impairment and dementia [31]. Defining mathematical functions to explore the role of vascular factors or other comorbidities in the development of the early stages of AD should be the object of further analysis. As an initial approach, we have reproduced a pure AD model based on Aβ deposition. However, we acknowledge that understanding the role of vascular and lifestyle-related factors is a key issue in the research agenda for dementia prevention and including them will be an advance in the mathematical representation of the natural history of AD. Third, estimating the prevalence of the preclinical and clinical stages from a model grounded in neuropathological data confronts the problem of the well-established lack of correlation between amyloid neuropathological findings and clinical manifestations [18,19]. However, our model is intended to provide epidemiological indicators to characterize the time from the onset of AD-related changes in the brain until the onset of clinical milestones, not to support or question the pathogenic or bystander role of amyloid. Simulation models can integrate existing partial knowledge of preclinical and prodromal phases of AD in joint representation of its natural history and assessment of the impact of possible interventions in terms of health benefit and economic burden. As has been noted in other studies with different approaches, results of simulation models show the lack of correlation between clinical and pathological classifications [18,19]. Aβ (Thal phase 1) is present in a significant percentage of the population 65 to 70 years of age; it seems to increase thereafter and plateau in the population aged 85 years or older. These differences support the hypothesis that Aβ accumulation is necessary, but insufficient alone, to cause dementia; other factors including vascular conditions, inflammation, brain reserve and cognitive reserve are also involved.