Quality of evidence is determined by study design, sampling method, and accuracy of diagnosis (Table ). Since AD is currently defined as a clinical-pathological entity, the ideal methodology for the question posed in this review is a population-based cohort, established in midlife and followed prospectively to autopsy. In this review, we group the literature related to vascular risk factors and AD into three methodological categories: 1) representative and non-representative prospective longitudinal cohort studies including autopsy (level A); 2) representative and non-representative prospective longitudinal studies without autopsy (level B); and 3) cross-sectional neuropathological studies (level C). Cross-sectional clinical studies without autopsy are not included in this review (level D).
Grading system for quality of evidence
The quality of ascertaining vascular risk factors, the criteria for diagnosis of dementia, and the method of rating distribution and severity of neuropathology vary considerably. Information about vascular risk factors ranges from self-report, proxy informants, medical records, to direct clinical and laboratory measures that are ascertained cross-sectionally or longitudinally. Differential misclassification of risk factors by self-report (for example, over-reported in AD group and under-reported in control group) will bias the associations between vascular risk factors and AD positively. In addition, temporality of risk factor assessment (for example, midlife versus late life) may affect the associations observed since neurodegenerative brain changes begin by middle age and exposure to risk factors changes over time. Besides the simple presence versus absence of a risk factor, the duration and magnitude of exposure to a risk factor should also be ideally considered. Epidemiological studies have diagnostic limitations since cases are usually defined based on a predetermined cutoff. Consequently, there is a possibility that 'mild dementia' cases are included in the normal group, which may attenuate the association between dementia and its risk factors.
A pathological diagnosis of AD, based on severity and distribution of NFTs and NPs [4
], remains the gold standard for diagnosis. Using the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) criteria as the reference standard, a clinical diagnosis of AD, based on typical clinical presentation and the exclusion of other causes of dementia [5
], showed 93% sensitivity but only 55% specificity [6
]. An evidenced-based review of the literature in 2002 [7
] showed small positive likelihood ratios for the diagnosis of dementia subtype (that is, 2 to 5) compared to the generic diagnosis of dementia (that is, >10). It is even more difficult to diagnose dementia due to mixed pathologies, yet in community-based studies, over half of older subjects with dementia harbor mixed pathologies at autopsy [8
], often showing infarcts up to half of which were not detected during life [11
]. In the absence of a true association between vascular risk factors and AD, misclassification of mixed AD/vascular dementia (VaD) cases as clinical AD would positively bias the associations between vascular risk factors and AD. Cerebrospinal fluid markers of amyloid-beta (Aβ) and phospho-tau and amyloid PET imaging have recently been incorporated into clinical research criteria [12
] and promise to improve specificity of the clinical diagnosis in the future.
While autopsy studies minimize errors in diagnosis, neuropathological data may be limited in many important ways. Neuropathological data are themselves cross-sectional (that is, can only be collected once); it is difficult, therefore, to determine sequence and causality based on neuropathological findings. At autopsy, severity of atherosclerosis, NFTs, NPs, and number of infarcts are often limited to semi-ordinal measures of severity, which may limit the power to detect small associations. Autopsy samples of convenience are likely to include significant selection bias (level of evidence = C). When a participant is selected from hospitals, memory clinics, or tertiary referral centers, the underlying population is usually not definable. Vascular pathology is relatively under-represented in autopsy studies drawn from memory clinics versus the community [13
]. Figure illustrates highly overlapping atherosclerosis, stroke, and AD pathologies in an older population. Figure also shows how retrospective selection of cases may lead to a comparison between AD cases and super healthy controls (CN1) that have neither AD nor vascular disease, leading to a spurious association between vascular risk factors and AD.
Figure 2 Sampling frame for convenience sample of Alzheimer's disease subjects and normal controls. Green shows two samples of normal controls (CN): CN1 is highly selected and CN2 is randomly sampled. In a cohort of Alzheimer's disease subjects (AD) and CN1, an (more ...)
A much stronger study design would be a population-based sample, followed prospectively from midlife to autopsy (level of evidence = A1). In such a cohort, the distribution of vascular risk factors of the cohort should be representative of the population (Figure ). Establishment of the cohort in midlife, well before the expression of disease, eliminates important aspects of selection bias. If the cohort is established later in life, when subjects at highest risk have already died, associations may be weakened and even reverse direction. If the cohort is assembled later in life and followed prospectively, comparable diagnostic thresholds can be used to first identify and remove prevalent cases and then prospectively identify incident cases (Figures and ). This prospective design strengthens the etiological significance of associations between risk factors and incident disease (level of evidence = B). Ideally, prospective longitudinal studies will include autopsy to confirm associations between risk factors and neuropathologically defined disease (level of evidence = A).
Figure 3 Sampling frame for a longitudinal aging cohort. An aging cohort study includes normal controls and subjects with Alzheimer's disease (AD) and stroke. The severity of arteriosclerosis and vascular risk factors are randomly sampled and representative for (more ...)
Sampling frame for a longitudinal aging cohort. At baseline examination, prevalent cases of stroke and Alzheimer's disease (AD) are identified and excluded from longitudinal follow-up.
Figure 5 Sampling frame for a longitudinal aging cohort. Prospective longitudinal follow-up allows estimation of new incident cases of stroke and Alzheimer's disease (AD). In ideal circumstances, it also allows the collection of a representative autopsy sample. (more ...)