OUP user menu

Small Vascular and Alzheimer Disease-Related Pathologic Determinants of Dementia in the Oldest-Old

Lidia Sinka MD, PhD, Enikö Kövari MD, Gabriel Gold MD, Patrick R. Hof MD, François R. Herrmann MD, MPH, Constantin Bouras MD, Panteleimon Giannakopoulos MD
DOI: http://dx.doi.org/10.1097/NEN.0b013e3181ffc3b9 1247-1255 First published online: 1 December 2010


The relative contributions of Alzheimer disease (AD) and vascular lesion burden to the occurrence of cognitive decline are more difficult to define in the oldest-old than they are in younger cohorts. To address this issue, we examined 93 prospectively documented autopsy cases from 90 to 103 years with various degrees of AD lesions, lacunes, and microvascular pathology. Cognitive assessment was performed prospectively using the Clinical Dementia Rating scale. Neuropathologic evaluation included the Braak neurofibrillary tangle (NFT) and ß-amyloid (Aß) protein deposition staging and bilateral semi-quantitative assessment of vascular lesions. Statistics included regression models and receiver operating characteristic analyses. Braak NFTs, Aß deposition, and cortical microinfarcts (CMIs) predicted 30% of Clinical Dementia Rating variability and 49% of the presence of dementia. Braak NFT and CMI thresholds yielded 0.82 sensitivity, 0.91 specificity, and 0.84 correct classification rates for dementia. Using these threshold values, we could distinguish 3 groups of demented cases and propose criteria for neuropathologic definition of mixed dementia, pure vascular dementia, and AD in very old age. Braak NFT staging and severity of CMI allow for defining most of demented cases in the oldest-old. Most importantly, single cutoff scores for these variables that could be used in the future to formulate neuropathologic criteria for mixed dementia in this age group were identified.

Key Words
  • Alzheimer disease
  • Brain aging
  • Mixed dementia
  • Oldest-old
  • Vascular dementia


Whether the accumulation of Alzheimer disease (AD) and vascular lesions is an unavoidable, consequence of normal brain aging remains among the most controversial debates in clinical research. The study of oldest-old people is particularly relevant in this context because it may help define the extent of changes in brain morphology that occur in the final years of life. Although some authors report significant associations between neurofibrillary tangle (NFT) burden in hippocampal subdivisions and dementia, thereby supporting the idea of a limbic dementia in the oldest-old (13), most recent contributions reveal a striking resistance to AD lesions and considerable overlap in NFT and β-amyloid (Aβ) burden between demented and nondemented individuals in this age group (413). The attenuation in the association between AD lesion burden and dementia with advancing age was also confirmed by the recent report of the Medical Research Council Cognitive Function and Ageing study (14), which challenged the utility of the Braak hierarchical schemes in this population (15); thus, additional pathologic factors may determine the clinical expression of dementia in the oldest-old. In particular, the accumulation of microvascular and small macrovascular lesions that are known to occur in most very old individuals may lower the burden of AD lesions necessary to produce dementia. This assumption parallels the widely disseminated idea that both pure AD and vascular dementia (VaD) are progressively substituted by mixed dementia in very old individuals (8, 1619). We previously documented that the assessment of Braak NFT staging, cortical microinfarcts (CMIs), and thalamic/basal ganglia could provide thresholds for the identification of cases with of mixed dementia in 156 elderly individuals younger than 90 years (15); however, this cohort included only 11% of nonagenarians and centenarians. To date, the relative contribution of AD and small vascular pathology in the occurrence of cognitive decline over age 90 is still a matter of debate. To address this issue, we investigated an independent series of 93 individuals from 90 to 103 years whose cognitive status was prospectively documented and who had various degrees of AD pathology, lacunes, and different types of microvascular pathology including CMI, diffuse and focal gliosis, periventricular, and deep white matter demyelination. Using systematic semiquantitative assessment of various types of vascular lesions and multivariate models that control for the possible confounding effect of age, we report here the identification of threshold values for AD and microvascular pathology that might provide the first operational definition of mixed dementia in the oldest-old.

Materials and Methods


The initial autopsy series included 1355 patients who were autopsied at the Geriatric and Psychiatric Hospitals of the University of Geneva during the period 1993–2003. Three criteria were used to define the final sample. First, cases younger than 90 years at death were excluded (n = 1083). Second, a Clinical Dementia Rating Scale (CDR) (20) had to be performed at most 3 months before death (n = 179). The CDR score was chosen because it was the most consistently available measure for all cases during this short period. Third, cases with clinically evident CNS disorders (i.e. tumors [n = 13], neuroinflammatory disorders [n = 2], Parkinson disease or Lewy body dementia [n = 15], and stroke [n = 62]) and those with macroscopic vascular pathology (i.e. brain infarcts [n = 78], hemorrhage [n = 17], hippocampal sclerosis [n = 34]) and other neurodegenerative lesions (i.e. Pick bodies [n = 1], or Lewy bodies [n = 12]) in routine neuropathologic assessment were excluded. In 234 cases, 2 exclusion criteria were present. No cases with frontotemporal dementia with ubiquitin-positive inclusions, argyrophilic grain disease, or venous sinus thrombosis were present in this hospital-based series. The final sample included 93 patients aged 90 to 103 years. Among them, there were 71 cases with prospectively documented Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV), diagnosis of dementia, including 29 with AD, 18 with VaD, and 24 with mixed dementia. At the time of the diagnosis, a full neuropsychological battery was administrated in all of these cases including the WAIS-R (Wechsler Adult Intelligence Scale-Revised test); Code; Mini Mental State Examination orientation items; Mattis Dementia Rating Scale items; Digit Span Forward; Corsi Block-Tapping Test; Buschke Double Memory Test; 16-items Shapes test; Verbal Fluency Test; Trail Making Test; Boston Naming Test; tests of ideomotor, reflexive, and constructional praxis; and Ghent Overlapping Figure test (21). Sex and age distribution of the cases according to CDR score are listed in Table 1. The main causes of death were infectious disorders (35.9%), heart failure (49.9%), pulmonary embolism (8.7%), and cancer (5.5%). The study received formal approval from the Local Ethics Committee of the University of Geneva Hospitals.

View this table:

Demographic Data and Clinical Dementia Rating Scores

CDRNo. Cases (F/M)Age, Mean ± SD, yr
05 (4/1)96.8 ± 4.4
0.517 (10/7)94.8 ± 3.8
18 (4/4)93.5 ± 3.1
226 (17/9)94.6 ± 3.4
337 (27/10)94.8 ± 3.5
All cases93 (62/31)95.2 ± 3.7
  • CDR, clinical dementia rating; F, female; M, male.

Neuropathology Methods

Brains obtained at autopsy were fixed in 15% formaldehyde for at least 4 weeks and cut into 1-cm-thick coronal slices. To define Braak NFT and AA staging and identify Pick bodies, ubiquitin-positive inclusions of frontotemporal dementia, argyrophilic grains, and Lewy bodies, tissue blocks from hippocampus, temporal, frontal, parietal, and occipital cortex were embedded in paraffin, and 12-μm-thick sections were processed with highly specific and fully characterized antibodies to phosphorylation-dependent tau AT8 (1:1000; Immunogenetics, Ghent, Belgium) (22), core Aβ protein A4 4G8 (1:1000; Signet Laboratories, Dedham, MA) (23), α-synuclein (1:20,000 courtesy of Dr. Y. Charnay, Department of Psychiatry, University of Medicine, Geneva, Switzerland), and ubiquitin (1:100; Sigma, St Louis, MO). Tissues were incubated overnight at 4°C. After incubation, sections were processed by the peroxidase-antiperoxidase method using 3,3′-diaminobenzidine as chromogen (24).

Lacunes defined as small definitive ischemic necrosis, ranging from 1 mm to 1.5 cm, located in the white matter or basal ganglia and thalamus were identified on macroscopic examination and verified on LuxolYvan GiesonYstained histologic slides. To visualize CMI, focal cortical and white matter gliosis and cortical cerebral amyloid angiopathy (CAA), tissue blocks from the anterior hippocampus, inferior temporal cortex (area 20), frontal cortex (area 9), parietal cortex (area 40), and occipital cortex (areas 17 and 18) bilaterally were cut into 20-Hm-thick sections and were stained with the Globus silver impregnation technique (24). To assess diffuse white matter and periventricular demyelination, whole coronal slices at the level of anterior commissure were embedded in paraffin, cut into 20-Km-thick sections, and stained with LuxolYvan Gieson.

Neuropathologic Diagnosis

Patients were classified as having VaD, AD, or mixed dementia. All AD cases were confirmed using the National Institute on AgingYReagan criteria for high likelihood AD (25). These cases did not show any of the above-mentioned vascular lesions. In the absence of widely accepted criteria for this age group, the neuropathologic diagnosis of VaD was based on the presence of a lacunar state in the basal ganglia and thalamus or CMIs affecting at least 3 cortical areas (i.e. at least 1 CMI/area; 5 serial sections/area), excluding the primary and secondary visual cortex. No or minimal NFTs and AA pathology were found in cases of pure VaD, all of which corresponded to NIA-Reagan criteria of low-likelihood AD. Cases that satisfied both neuropathologic criteria for AD (NIA-Reagan high-likelihood AD) and our criteria for VaD were classified as having mixed dementia.

Semiquantitative Assessment of AD LesionYRelated and Vascular Burden

All cases were classified according to Braak NFT staging system (26). Aβ protein deposition staging was performed according to the amyloid nomenclature proposed by Thal et al (27). Lacunes, CMIs, and focal cortical gliosis were assessed semiquantitatively in 10 sections per area using the following scores: 0 = absence of such lesions, 1 = fewer than 3 lesions per slide, 2 = 3 to 5 lesions per slide, and 3 = more than 5 lesions per slide. The severity of subcortical gliosis was defined semiquantitatively by 2 experienced neuropathologist depending on global estimates of the number of reactive astrocytes in the subcortical white matter in the same number of sections stained with Globus silver impregnation, using the following rating scale: 0 = absent, 1 = mild, 2 = moderate, and 3 = severe. Although we cannot exclude that additional pathology may be present mainly in neocortical areas, the use of a high number of sections limits this possibility. For each of these lesions, a total score was obtained by adding the scores of each area. The severity of diffuse white matter and periventricular demyelination in each hemisphere was estimated in LuxolYvan Gieson-stained sections using the same semiquantitative scale. Scores for each hemisphere were added to obtain a total score. The same semiquantitative assessment of lacunes and microvascular pathology was used in our previous studies with a high interrater reliability (28, 29).

Braak NFTs and Aβ protein deposition staging and vascular pathology scoring were performed by 2 independent investigators (E.K. and C.B.), who were blinded to the clinical findings. There was a high interrater reliability with κ values ranging from 0.90 to 0.94. The remaining cases were resolved by consensus meeting.

Statistical Analysis

Maximal likelihood ordered logistic regression with proportional odds was used to evaluate the association between CDR scores (the dependent variable) and neuropathologic parameters (Braak NFT staging, Aβ protein deposition staging, lacunes, and microvascular pathology scores) in a univariate model. Subsequently, the same method was applied in a multiple model to take into account the effect of age and interactions between the neuropathologic variables. In addition, the series was dichotomized according to the presence or absence of a DSM-IV diagnosis of dementia to built logistic regression models exploring the impact of lacunes and microvascular pathology on the presence of dementia. Braak NFT and Aβ staging were entered as dummy variables in all regression models. Using sensitivity analysis, we constructed corresponding receiver operating characteristic (ROC) curves and identified the best threshold for neuropathologic variables to differentiate demented from nondemented cases and then reconstructed the ROC curves with the best thresholds. An ROC curve graphs the sensitivity and the false-positive rate (1 – specificity) over a range of possible threshold values; an area under the curve (AUC) of 1.0 corresponds to a perfect prediction whereas a value of 0.5 to a useless model. Binomial confidence intervals were computed for the AUC. Statistical analyses were performed using the Stata software package, release 11.0 (College Station, TX).



There were 48 cases with AD, 21 with VaD, and 17 with mixed dementia based on routine neuropathologic analysis. Among cases with DSM-IV dementia, there were 7 (9.9%) without NFTs and 10 (14.1%) without AA deposits; all were diagnosed clinically and pathologically as VaD cases. Only 4 demented cases were free from vascular lesions. There was no case with significant NFT formation in the absence of amyloid pathology that could be classified as tangle-dominant dementia according to the criteria of Jellinger and Attems (30). There were 18% of nondemented cases and 12.7% of cases with DSM-IV dementia that corresponded to Braak stage I. Among demented cases, 32.5% had Braak stages lower than IV. Approximately 25% of cases displayed no or minimal Aβ deposits (Aβ stage 1; 46% in nondemented cases, 11.3% in cases with DSM-IV dementia) in neocortical areas. The most frequently observed microvascular changes were diffuse white matter (77.2% in both cases with DSM-IV dementia and nondemented cases), periventricular demyelination (62.7% in nondemented cases, 45.5% in cases with DSM-IV dementia), and white matter gliosis (81.8% in nondemented cases, 60.6% in cases with DSM-IV dementia). Focal cortical gliosis was present in 45.6% of nondemented cases and 25.3% of cases with DSM-IV dementia. The presence of CMI most closely correlated with dementia (18.2% in nondemented cases, 54.9% in cases with DSM-IV dementia). Lacunes were preferentially located in white matter (13.7% in nondemented cases, 27.2% in cases with DSM-IV dementia) or basal ganglia (13.7% in nondemented cases, 24% in cases with DSM-IV dementia). They coexisted with CMI in 27.5% of demented cases. Cortical CAA was very rare: 3.2% in nondemented cases and 13.9% in cases with DSM-IV dementia. This unexpectedly low percentage may be due to the use of Globus silver impregnation because this method allows for the detection only of cases with the more severe form of CAA that are frequently associated with vascular lesions (scores 3 and 4 according to Olichney et al [31]).

Table 2 summarizes the frequencies of AD lesion staging and vascular lesions as a function of the CDR score. As expected, the prevalence of Braak stages higher than III increased steadily between CDR 2 and 3 cases. This was also the case for Aβ staging III and IV. Among vascular lesions, the frequency of CMIs increased steadily as a function of the CDR score (<5% in CDR 0 to 1 cases, but 16% in CDR 2 cases and 33% in CDR 3 cases). White matter lacunes and diffuse white matter demyelination were also observed more frequently in severely demented cases.

View this table:

Distribution of Neuropathologic Findings as a Function of the CDR Score

CDR Score
Neuropathologic Findingsn (%)00.5123
Braak NFT I–II33 (35.5)4 (4.3)12 (12.9)3 (3.2)9 (9.7)5 (5.4)
Braak NFT III–IV38 (40.8)10 (1.1)5 (5.4)5 (5.4)12 (12.9)6 (16.1)
Braak NFT V–VI22 (23.7)0005.4 (5.8)18.3 (19.68)
Aβ I–II42 (45.2)5.4 (5.8)14.0 (15.05)7.5 (8.06)11.8 (12.69)6.5 (6.99)
Aβ III–IV51 (54.8)04.3 (4.62)1.1 (1.18)16.1 (17.31)33.3 (35.81)
CMIs43 (46.2)1.1 (1.18)3.2 (3.44)3.2 (3.44)14.0 (15.05)24.7 (26.56)
Thalamic lacunes6 (6.5)0003.2 (3.44)3.2 (3.44)
Basal ganglia lacunes20 (21.5)1.1 (1.18)2.1 (2.26)1.1 (1.18)7.5 (8.06)9.7 (10.43)
White matter lacunes23 (24.7)1.1 (1.18)2.1 (2.26)05.4 (5.81)16.1 (17.31)
Periventricular demyelination47 (50.5)3.2 (3.44)11.8 (12.69)6.5 (6.99)16.1 (17.31)12.9 (13.87)
Diffuse white matter demyelination72 (77.4)3.2 (3.44)15.1 (16.24)8.6 (9.25)21.5 (23.12)29.0 (31.18)
Focal cortical gliosis28 (30.1)2.1 (2.26)8.6 (9.25)2.1 (2.26)10.8 (11.61)6.5 (6.99)
White matter gliosis61 (65.6)4.3 (4.62)15.1 (16.24)5.4 (5.81)17.2 (18.49)23.7 (25.48)
Cortical CAA7 (7.5)1.1 (1.18)002.1 (2.26)4.3 (4.62)
  • Percentages were calculated with reference to the total sample (n = 93).

  • Aβ, β-amyloid; CDR, Clinical Dementia Rating; NFT, neurofibrillary tangle.

There was no significant association between age and AA staging or between age and vascular scores. There was only a modest association between age and Braak NFT score (Spearman ρ = 0.276, p = 0.0073, n = 93). As expected, there was a strong correlation between Braak NFT score and AA staging, (Spearman ρ = 0.575, p < 0.0001, n = 93). There was no correlation between CMI and either NFT score (Spearman ρ = 0.00, p = 0.96) or Aβ staging (Spearman ρ = 0.16, p = 0.13).

Relationship of CDR Scores and Clinical Diagnosis

By univariate analyses, there were 4 independent variables that were significantly related to CDR scores. These included age (p < 0.05), Braak NFT staging (p < 0.001), Aβ deposition staging (p < 0.001), and CMI (p < 0.05). In contrast, CAA, thalamic, basal ganglia and white matter lacunes, periventricular and diffuse white matter demyelination, and focal and diffuse cortical gliosis were not significantly related to CDR scores. We then tested a multiple model including all 4 variables that proved significant in the univariate approach. Three of the variables (Braak NFT staging, Aβ deposition staging, and CMI) remained significant predictors of cognitive status (Table 3). Their concomitant assessment allowed for predicting 30% of the CDR variability.

View this table:

Multivariate Analysis of Alzheimer Disease and Vascular Pathology Impact on Clinical Dementia Rating Scores (Maximum Likelihood Ordered Logistic Regression)

NFT BraakI1.00--
Aβ staging11.00--
  • Braak staging scores 4 or higher, β-amyloid (Aβ) staging scores 3 or higher, and CMI scores significantly increase the risk for higher CDR scores.

  • CI, confidence interval; OR, odds ratio.

We then evaluated the relationship between the most important clinical outcome (presence or absence of DSM-IV diagnosis of dementia) and neuropathologic parameters. In univariate analyses, Braak NFT staging, Aβ deposition staging, CMI score, and age were significant predictors of the presence of clinical dementia. A multiple model that included all of these variables revealed that age was no longer a significant predictor and that the 3 remaining neuropathologic scores explained 48.9% of the presence of dementia. In this model, Braak NFT staging explained 24%, CMI score 18%, and Aβ deposition staging 6.9% of the presence of dementia. Likelihood ratios were highly significant for Braak NFT and CMI (LR test, p < 0.0001) but not for AA (LR test, p = 0.100).

The ROC curves were constructed using the combination of Braak NFTs (AUC = 90%, 95% confidence interval [CI], 82%–95%) and AA staging with CMI scores (AUC = 83%, 95% CI, 77%–92%) to determine the threshold value with the best sensitivities and specificities. This corresponded to cutoff scores of 4 or greater for Braak NFT staging and 2 or greater for the CMI score. The corresponding cutoff scores for the combination of Aβ staging and CMI scores were 2 and 1 or greater, respectively. The performances of these models were as follows: sensitivity 0.82, specificity 0.91, positive predictive value 0.97, and negative predictive value 0.61 (Braak NFTs, CMI) and sensitivity 0.94, specificity 0.36, positive predictive value 0.83, and negative predictive value 0.67 (Aβ staging, CMI). The areas under the ROC curve were 0.88 and 0.77, respectively (Fig. 1). When the threshold scores were applied, 84% (for Braak NFT staging combination) and 81% (for Aβ staging combination) of the demented cases were correctly classified.


Receiver operating characteristic curves were used to compare the performances of 2 scoring systems. (A, B) The first takes into account Braak neurofibrillary tangle (NFT) staging and cortical microinfarcts (CMIs) (A) and the second β-amyloid (Aβ) staging and CMI (B). Curve A contains the best-performing threshold score (red point), as close as possible to the theoretically ideal of a sensitivity of 1 and a specificity of 1 (1 − specificity = 0). This point corresponds to a Braak NFT stage IV and CMI score of 2. Also, note that the area under the curve (AUC, shaded area) is significantly larger in A compared with that in B. CI indicates confidence interval.

In scatter plot diagrams, almost all nondemented cases corresponded to Braak NFT stage 4 or lower and to CMI score of 2 or lower. The same was true for most nondemented cases when Aβ staging was considered (Fig. 2). Most cases with DSM-IV diagnosis of AD corresponded to the combination of Braak NFT stage 4 or higher and CMI score of 2 or lower. Most VaD cases displayed Braak NFT stage 4 or lower. The distinction between the clinical subtypes of dementia was less easy when the combination of Aβ staging and CMI scores was considered. Not surprisingly, the poorest association between clinical diagnosis and neuropathologic variables was found in mixed dementia cases (Fig. 3).


Scatter plots of demented (black circle) and nondemented (blue triangle) cases according to the best cutoff values for Braak neurofibrillary tangle (NFT) stage (2.1) and β-amyloid (Aβ) staging (2.2) and cortical microinfarct (CMI) score. Black circles represent demented cases with high degenerative scores and low vascular scores in area A (pure Alzheimer disease), high vascular scores and low degenerative scores in area D (pure vascular dementia), and high vascular and degenerative scores in area B (mixed dementia). Several cases present with dementia, although both vascular and neurodegenerative scores are low (black circles in area C), suggesting that the coexistence of subthreshold degenerative and vascular lesions may have a synergistic effect. To avoid superimposition of multiple cases, a 10% random noise was applied to all points; brackets on the y axis indicate the magnitude of the jitter.


Scatter plots with cases coded according to their clinical DSM-IV diagnosis (nondemented, blue triangle; Alzheimer disease [AD], green circle; vascular dementia, red square; mixed dementia, red cross) for the combination of Braak neurofibrillary tangle (NFT) stage and cortical microinfarct (CMI) score (3.1) as well as A-amyloid (Aβ) staging and CMI score (3.2). Note the general agreement between clinical diagnosis and neuropathologic definition of AD but not for vascular or mixed cases. To avoid superimposition of multiple cases, a 10% random noise was applied to all points; brackets on the y axis indicate the magnitude of the jitter.


The present findings reveal that, although diffuse white matter and periventricular demyelination as well as white matter gliosis are present in most nonagenarians and centenarians, they did not negatively impact cognition. Less frequently seen thalamic and basal ganglia lacunes are also vascular epiphenomena without major cognitive repercussions in these very old people. The 2 key independent determinants of dementia in the cohort were Braak NFT staging and severity of CMI. These 2 variables were sufficient to identify most demented cases with a high level of accuracy, as demonstrated by our ROC analysis.

At first glance, our results agree with the notion that the number of cases with no or minimal AD lesions decreases after 90 years. In fact, the percentage of cases with Braak NFT staging I and II that correspond to the initial development of tau pathology within the hippocampal formation was 35.5% versus more than 50% reported in younger cohorts. To a lesser extent, the same was true for Aβ staging (12, 32). However, it is noteworthy that several cases displayed no NFT or Aβ deposits even when there was clinically overt dementia, there-by supporting the idea that the occurrence of AD-related pathologic changes is not an inevitable concomitant of brain aging (5, 912).

In agreement with the recent autopsy data in very old individuals who participated in the Baltimore Longitudinal Study of Ageing (33), our data show that Braak NFT staging remains a significant predictor of cognitive status even in oldest-old people; even Braak IV staging is associated with a more than 5-fold increase of the risk for higher CDR scores. This risk is increased 11-fold for cases with Braak NFT stages V and VI. However, the strength of the association between NFT burden and dementia severity clearly decreases when compared with our previous findings in mixed cases with a mean age of 85 years (34). This progressive decline in the ability to predict the severity of dementia on the basis of NFT burden is mainly due to the unusually high percentage of demented cases that displayed Braak NFT staging lower than IV (24% [26]). Taken together, these observations suggest that, although cognitively deleterious, the hierarchical NFT progression within the cerebral cortex is a less powerful determinant of dementia severity in patients older than 90 years. Aβ staging is also associated with cognitive decline in nonagenarians and centenarians, but its predictive power remains modest (14, 28, 29, 34, 35).

In contrast to the numerous investigations of AD lesion evolution, the role of lacunes and microvascular lesions in very old age has been overlooked. These pathologic changes were present in all but 4 cases, thereby stressing the importance of the vascular burden in mixed pathologies in people over 90 years (8, 1618, 36). Overall, the data of our multivariate analyses (using both CDR scores and DSM-IV dementia as dependent variables) are in line with the well-established synergistic effect of AD and microvascular pathology on the development of dementia in old age (32, 37). By analyzing each type of lesion separately, we have identified 3 main patterns of clinicopathologic correlations. The first concerns frequent but cognitively benign changes such as diffuse white matter and periventricular demyelination. This observation is of particular importance for these 2 types of lesions as they have been proposed to be at the origin of mixed dementia on the basis of earlier neuroimaging studies (3842). As in younger cases, our data imply that these neuropathologic changes have no cognitive repercussion in mixed cases. One should consider, however, that both types of demyelination were assessed only at the level of anterior commissure in this analysis. We cannot formally exclude the possibility that the total volume of white matter lesions may correlate better with cognitive measures. The second pattern refers to less frequent and cognitively innocuous changes such as cortical and white matter gliosis and lacunes. For these lesions, our findings disagree with previous observations in pure vascular and mixed cases of younger age that stressed the deleterious cognitive effect of thalamic and basal ganglia lacunes (28, 4346). It is possible that the development of these lesions and the subsequent disruption of subcortical frontal circuits are associated with higher mortality rates in younger ages and that, as a consequence, there is overrepresentation of milder cases in the ninth and tenth decades. Alternatively, individuals who survive over age 90 may be able to compensate the negative impact of subcortical lacunes by activating synaptic remodeling, as recently proposed (47). The third pattern is related to CMI, which was the only microvascular lesion clearly associated with cognitive outcome. As in our younger cases (34), a 1-point increase of the CMI score confers a 1.2-fold increase of the risk for higher CDR scores. Most importantly, this parameter alone predicted 18% of the variability in the presence of dementia (versus 12% in younger cases). These observations complement previous neuropathologic data demonstrating that the progressive accumulation of CMI in the course of brain aging is a deleterious phenomenon both for cognition and emotional regulation (28, 29, 32, 48). In the only neuropathologic analysis that addressed the cognitive repercussions of vascular burden in a community-based cohort of oldest-old individuals, Savva et al reported attenuation of the association between the presence of lacunes as well as small vessel disease and dementia over 90 years (14, 28, 29, 34, 35). These lesions still had a modest negative impact on cognition even in the oldest-old, however. No distinction was made between the different types of microvascular lesions in this study. Moreover, we cannot exclude that AD and vascular burden may have differential contributions to cognitive outcome in their community-based series and in the present hospital-based cohort.

Using a simple semiquantitative model including only the 2 main independent predictors (Braak NFT staging and CMI scores), we were able to identify most of the demented cases. The area under ROC curve reached 90%, an unexpectedly high percentage in light of the neuropathologic overlap between demented and nondemented cases in nonagenarians and centenarians (413). Using a single cutoff of 4 for Braak NFT staging and 2 for CMI score, we obtained extremely high sensitivity (0.82) and specificity (0.91) values. Importantly, specificity values were much higher than that obtained in our previous study of younger cases (0.79), with an acceptable loss of sensitivity (0.90). The combination of Aβ staging and CMI score provided higher sensitivity values (0.94) but suffered from an evident lack of specificity (0.36). Not surprisingly, however, the association between the DSM-IV clinical types of dementia and these cut-offs was not optimal mainly for vascular and mixed cases. In fact, the combination of high (≥4) Braak NFT staging and low (≥2) CMI score was present in most cases with DSM-IV diagnosis of AD. In contrast, the distribution of lesion scores was much more heterogeneous in vascular and mixed cases. This observation parallels earlier sets of clinicopathologic observations pointing to the low validity of currently available criteria for VaD (49).

The interpretation of such data needs some additional considerations. First, this hospital-based cohort is not representative of the whole spectrum of nonagenarians and centenarians (50). In particular, the inclusion of a high number of severely demented cases and the relative rarity of CDR1 cases might lead to an overestimation of the sensitivity and specificity values for the proposed cutoff scores. Second, the cases were divided clinically according to the presence or absence of DSM-IV dementia. Given the long period of autopsy selection, it was not possible to use specific criteria to identify mild cognitive impairment among our CDR 0.5 cases. Thus, we cannot exclude that the performance of the proposed neuropathologic hallmarks may change in this particular population. Third, although we assessed microvascular changes in several neocortical association areas bilaterally, the sensitivity and specificity values depend on this sampling strategy and should be tested in other neuropathologic settings. Fourth, the confidence intervals of odds ratio in our multivariate analyses were very wide reflecting the relative rarity of CDR 0 and 1 cases. In the same line, both the percentage of cognitive variability explained by the neuropathologic parameters and negative predictive value of our cutoff scores remain suboptimal reflecting the determinant role of additional factors not assessed in this study such as degree of cortical atrophy, neuronal loss, axodendritic pruning, reduced synaptic density, or failure of microvascular integrity (51). In this perspective, a recent study by Jellinger and Attems (30) suggested that subcortical arteriosclerotic changes and hippocampal sclerosis are frequent pathologic substrates of mixed dementia in the oldest-old. Future validation studies in large community-based cohorts of oldest-old individuals using the single cutoff scores for Braak NFT staging and CMI scores proposed here, in combination with a detailed analysis of lesional and non-lesional parameters, are necessary to explore further the structural determinants of cognitive decline in this age group.



  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
  29. 29.
  30. 30.
  31. 31.
  32. 32.
  33. 33.
  34. 34.
  35. 35.
  36. 36.
  37. 37.
  38. 38.
  39. 39.
  40. 40.
  41. 41.
  42. 42.
  43. 43.
  44. 44.
  45. 45.
  46. 46.
  47. 47.
  48. 48.
  49. 49.
  50. 50.
  51. 51.
View Abstract