The Effect of Vascular Neuropathology on Late-life Cognition: Results from the SMART Project

Abstract

Background

Cerebral vascular pathology may contribute to cognitive decline experienced by some elderly near death. Given evidence for mixed neuropathologies in advanced age, preventing or reducing cerebrovascular burden in late life may be beneficial.

Objective

To correlate measures of cerebral vascular pathology with cognitive trajectories.

Setting

Observational study.

Participants

A cohort of 2,274 individuals who came to autopsy at a mean age of 89.3 years and 82 percent of whom had at least two cognitive assessments within the last six years of life was compiled from six centers conducting longitudinal studies.

Measurements

For each cognitive domain: immediate and delayed memory, language, and naming, three trajectories were examined: good, intermediate, and poor cognition. The probability of a participant belonging to each trajectory was associated with measures of cerebral vascular pathology after adjustment for demographics, APOE, and Alzheimer neuropathology.

Results

A large proportion of the cohort (72-94%) experienced good or intermediate cognition in the four domains examined. The presence of arteriolosclerosis and the presence of lacunar infarcts doubled the odds of belonging to the poor cognitive trajectory for language when compared to the good trajectory. The presence of lacunar infarcts increased the odds of an intermediate or poor trajectory for immediate and delayed recall while the presence of large artery infarcts increased the odds of poor trajectories for all four cognitive domains examined. Microinfarcts and cerebral amyloid angiopathy had little effect on the trajectories.

Conclusion

Indicators of cerebral vascular pathology act differently on late life cognition.

Introduction

Two common disease processes affecting the aging brain are Alzheimer's disease (AD) and cerebrovascular disease (due to cerebrovascular pathology (CeVP)). Multiple studies involving large autopsy cohorts show that most individuals in advanced old age harbor both AD and CeVP in the brain although the severity and subtype of pathologies are highly variable (1-4). CeVP and AD pathology appear to act independently on the brain (5-8). AD pathology is widely recognized to consist of amyloid plaques and neurofibrillary tangles (9). On the other hand, CeVP has less broadly accepted diagnostic categories with multiple pathologies, including infarcts, hemorrhages, atherosclerosis, arteriolosclerosis, and cerebral amyloid angiopathy (CAA). These pathologies often are comorbid and may affect different cognitive domains; therefore, studies of CeVP have more sources of variability. This is clinically relevant since the presence of multiple CeVP greatly increases the odds of dementia in the old (10).

The association between CeVP subtypes identified at autopsy and late life cognition is understudied. One approach is to correlate the presence of CeVP at autopsy with the last cognitive assessment proximal to death. Such studies have shown that the presence of microinfarcts, CAA, and lacunar infarcts correlate with measures of global cognition (4, 11-13) and the presence of macroscopic infarcts correlated with measures of episodic memory and working memory in cognitively intact elderly at baseline (15). In a longitudinal analysis, the presence of macroscopic infarcts correlated with a linear decline in cognition among cognitively intact elderly at baseline (16). Assigning a cerebrovascular parenchymal pathology score to each of 79 autopsies Chui et al. (17) showed that this CeVP process contributes to cognitive impairment. A longitudinal analysis showed that some elderly experience a change in the slope of the linear decline in cognition as they approach death and related this change to the presence of macroscopic infarcts (3). Recent work also found CAA to be related to cognitive decline (18). Hence, clarifying the effect of mixed CeVP on late-life cognition provides insight into avoiding the transition to dementia associated with poor cognition.

Although the pathology-based studies are invaluable, the episodic nature of CeVP is such that meaningful pathology often occurs proximal to death (i.e., after the final cognitive assessment). Thus, clinical studies – using neuroimaging as a surrogate for CeVP in live patients – are very important. For example, based on the magnetic resonance imaging and pathologic analysis in 116 patients with a mix of diagnoses (normal, AD, mixed dementia, and cardiovascular disease) Zheng et al. (7) conclude that global cognition is affected more by AD pathology than arteriosclerosis. Based on PET imaging in 393 normal controls from the Mayo Study of Aging Vemuri et al. (8) show that the presence of brain infarcts and white matter hyperintensities, each at a severe level, is additive and not synergistic when comparing the relative effect of AD pathology versus vascular pathology on cognitive decline.

The purpose of this manuscript is to report on the presence of five indicators of CeVP measured at autopsy – microinfarcts, large artery infarcts, lacunar infarcts, arteriolosclerosis, and CAA – on changes in late-life cognition. To maximize the chances of understanding the association between CeVP subtypes and cognitive changes, we evaluated a large autopsy cohort that was combined across different research centers (18). Further, we incorporated information from longitudinal cognitive studies to evaluate the cognitive decline during the last six years of life, rather than relying only on the final clinical examination.

Methods

Study Participants

Data were drawn from the Statistical Modeling of Aging and Risk of Transition (SMART) Study, which is a harmonized database of demographics, genetics, medical history, family history, neuropathology, and cognitive assessments of participants in 11 longitudinal studies of aging and dementia from six research centers. Details on the structure of this database have been reported previously (19). Briefly, data came from the Memory and Aging project at Washington University (WASHU); the Biologically Resilient Adults in Neurological Studies project at the University of Kentucky (UKY); the Nun Study (NUN); the Honolulu Asia Aging Study (HAAS); the Religious Orders Study and the Memory and Aging projects at Rush University (RUSH); and the Oregon Brain Aging Study, the African American Dementia and Aging Project and the Klamath Exceptional Aging Project at the Oregon Health & Science University (OHSU). This database contains records on a total of 11,541 participants in these projects with 3,001 of these participants coming to autopsy. For inclusion in the present study, participants had to undergo at least two cognitive assessments within six years prior to death (except for HAAS which had a greater than 2-year interval period between serial assessments). This yielded 2,274 autopsies for analysis, with 82 percent of autopsies having at least two cognitive assessments before death. All research procedures were approved by the Institutional Review Boards at the cohort's home research center, and all participants provided written informed consent.

Cognitive Evaluations

Measures of confrontation naming (the Boston Naming Test (BNT), 15-item version (20), language/verbal fluency (Animal Naming, (21)), immediate and delayed memory were derived from the longitudinal cognitive assessments at each of the six centers participating in the project. Raw scores on the neuropsychological exams were converted to z scores using the baseline assessments among cognitively intact participants at each center. For measures involving two or more tests, z scores were averaged across component instruments to estimate the measures of immediate and delayed recall. Immediate recall was based on the average of two z scores, one for Word List Total (20) and the other for Logical Memory Story A (Immediate Recall, (22)), whenever both test results were available. When only one test was available, the composite z score was set to the z score for the available test. Delayed recall was a similar composite of Word List Delayed Recall z score (20) and Logical Memory Story A (Delayed Recall) z score when available. Data for the BNT15 and immediate recall were not available from the WASHU site.

Neuropathology Measures

Indicators of AD pathology were Braak stage (stage III or higher; (23)), neuritic plaques (intermediate or frequent; (24)), and diffuse plaques (intermediate or frequent; (24)). Indicators of CeVP included the presence of large artery infarcts, lacunar infarcts, severe arteriolosclerosis (H & E stain) and CAA (Congo red stain). The latter assessment was missing on 53.3% of the autopsies, and when the analysis was limited to those autopsies where it was present, the results did not change since presence of CAA was not related to cognition. Hence, this measure was removed from the analyses. Large infarcts required a maximum diameter of 1 cm or greater and involvement of medium to large sized meningocerebral vessels. Lacunes had infarcts or hemorrhages with maximum diameter less than or equal to 1 cm attributed to small parenchymal vascular disease most commonly found in deep gray matter. A microinfarct is judged dichotomously (present/absent) and is defined as an area of infarction seen on microscopy using H&E stain that was too small to be observed grossly (25). Arteriolosclerosis is graded on a 0-3 global scale (none, mild, moderate, or severe), and is defined as concentric hyaline thickening of the media of arterioles, which may be accompanied by intimal fibrosis (26, 27).

Statistical Analysis

Descriptive statistics were produced for all variables for each center in the study and overall. Summary measures included mean and standard deviation for continuous variables and percent frequencies for categorical responses. A trajectory analysis was applied to the profiles of cognitive assessments within the last six years of life. Covariates in the analysis included age at death, gender, years of education, presence of at least one apoliprotein-ε4 allele (APOE-ε4), and indicators for AD pathology through the indicators for Braak stage (stage III or higher), neuritic plaques, and diffuse plaques; as well as an adjustment for research center based on indicator variables.

The trajectory analysis produces a small set of latent patterns that best summarize these profiles. The number of patterns may vary with neuropsychological instrument and choice of polynomials for trends over time but for consistency three trajectories were chosen: good, intermediate, or poor cognition. Each participant belongs to each latent trajectory, but with varying probability. Probabilities vary in accordance with the covariates (see list above) and study variables (the vascular pathology variables) after fitting a polytomous logistic model to the data. These models produce an odds ratio (OR) for each study variable adjusted for the covariates listed above that reflects the odds that the individual belongs to a given pattern when compared to a chosen base pattern. All computations were done using PROC TRAJ in PC SAS 9.3® (SAS Institute, Inc; Cary, NC).

Missing CEVP variables

Due to missing values on arteriolosclerosis and microinfarcts two analyses were conducted. The first analysis used data from 1,823 autopsies having all four CEVP variables while the second analysis used data from 2,274 autopsies but involved only the variables lacunar infarcts and large artery infarcts. The research sites HAAS and NUN accounted for 88% of the missing information on arteriolosclerosis and microinfarcts.

Results

Tables 1A and 1B list descriptive statistics for Analysis 1 and 2, respectively by study center. In Table 1B the mean number of cognitive assessments within the last six years of life was 3.8 ± 2.0 and varied by center. The higher means were from OHSU, UKY, and RUSH, which aimed to assess participants annually while NUN and WASHU averaged 18 months between assessments. HAAS sought to assess participants on a bi-annual basis. Deaths occurred at a mean age of 89.3 years, with 25% of the participants dying before age 85.5 and under 10% dying before age 82.0 years. The severity of AD pathology varied by center with RUSH having the highest proportion of autopsies with high Braak stages and intermediate or frequent neuritic plaques. The variability in CeVP indicators across centers was even larger, partially due to HAAS, which had the highest proportion of cases with lacunar infarcts and microinfarcts but the smallest proportion with arteriolosclerosis when compared to the other centers. The descriptive statistics are similar between the tables, except for HAAS where deaths occurred almost three years earlier in Table 1A and the NUN Study where fewer assessments were available for analysis in Table 1B.

Table 1.

A Descriptive statistics by research center for the first analysis (n = 1,823)
	N	# Assessments		Death Age		Years of education		Female	APOE4	BRAAK	NEUR	DIFF	LAC	LINF	ARTER	MICRO
		Mean	Std	Mean	Std	Mean	Std
HAAS	151	1.4	0.5	87.4	4.8	10.6	3.2	0.0%	21.2%	52.3%	28.5%	52.3%	56.3%	30.5%	11.3%	76.2%
NUN	247	4.1	0.9	90.8	4.8	18	3.3	100.0%	22.3%	34.0%	47.4%	81.0%	27.5%	14.2%	19.8%	14.6%
OHSU	139	4.9	1.5	94.2	4.3	13.4	3.2	58.3%	20.1%	54.0%	44.6%	25.9%	25.2%	15.1%	96.4%	39.6%
RUSH	937	4.7	1.8	88.1	6.6	16.4	3.6	63.4%	26.3%	53.3%	66.5%	68.6%	39.5%	20.3%	89.2%	16.5%
WASHU	85	3.2	1.6	92.8	6.7	14.3	3.4	64.7%	23.5%	38.8%	30.6%	68.2%	34.1%	8.2%	94.1%	35.3%
UKY	264	4.9	1.5	87.6	7.1	16	2.4	60.2%	27.3%	34.8%	57.6%	65.9%	33.7%	19.3%	54.9%	27.3%
All	1823	4.3	1.8	89	6.5	15.8	3.9	62.3%	24.8%	47.3%	56.1%	65.3%	37.1%	19.2%	69.2%	25.4%
Note: NEUR = neuritic plaques, DIFF = diffuse plaques, LAC = lacunar infarcts, LINF = large infarcts, ARTER = severe arteriolosclerosis, and MICRO = micro infarcts

B. Descriptive statistics by research center for the second analysis (n = 2,274)
	N	# Assessments		Death Age		Years of education		Female	APOE4	BRAAK	NEUR	DIFF	LAC	LINF
		Mean	Std	Mean	Std	Mean	Std
HAAS	387	1.4	0.6	90.1	4.7	11	3.2	0.0%	21.4%	39.0%	34.6%	53.0%	63.8%	26.6%
NUN	441	3.3	1.4	90.1	5.3	17.5	3.7	100.0%	24.3%	38.3%	45.4%	82.3%	31.3%	17.5%
OHSU	139	4.9	1.5	94.2	4.3	13.4	3.2	58.3%	20.1%	54.0%	44.6%	25.9%	25.2%	15.1%
RUSH	949	4.6	1.8	88	6.6	16.4	3.6	63.2%	26.2%	53.1%	66.4%	68.6%	39.3%	20.1%
WASHU	89	3.2	1.7	92.5	6.7	14.3	3.3	65.2%	23.6%	38.2%	30.3%	68.5%	33.7%	7.9%
UKY	269	4.9	1.5	87.3	7.4	16	2.4	60.2%	27.1%	34.2%	56.9%	65.8%	33.5%	19.0%
All	2274	3.8	1.9	89.3	6.4	15.4	4.1	59.0%	24.7%	45.1%	53.0%	65.7%	40.1%	19.8%

Note: NEUR = neuritic plaques, DIFF = diffuse plaques, LAC = lacunar infarcts, LINF = large infarcts

The trajectory analysis was conducted twice; the first used all four CeVP variables but was based on a reduced sample size of 1,823 autopsies due to missing data, while the second used only lacunar and large artery infarcts and was based on all 2,274 autopsies (Table 1). Both analyses produced similar results based on three latent trajectories for each cognitive domain. The first trajectory shows a pattern of a subset of individuals who retain their cognition at about one standard deviation above the mean for baseline cognitively intact participants (“normal”) throughout the last six years of life (Figure 1). We label this group as having good cognition near death. The second latent trajectory contains individuals who began with below normal test scores and declined slightly over the six years before death. This group is labeled intermediate cognition before death. The third latent trajectory belonged to individuals who began well below normal six years prior to death and then proceeded to decline even further as death approached. This is especially true of the 15-item BNT which declines from 4 standard deviations below normal to almost 10 standard deviations below normal during the six-year period in this group. This group is labeled poor cognition near death.

Figure 1.

Plot of three trajectories: good (solid), intermediate (large dashed) and poor cognition (short dashed) by cognitive domain. The probability that a subject follows one of these trajectories are in Panel A (n= 1,823): naming (0.747, 0.104, 0.089), language (0.188, 0.531, 0.281), immediate recall (0.323,0.485, 0.192), and delayed recall (0.366, 0.433, 0.201) and in Panel B (n=2,274): naming (0.744, 0.198, 0.059), language (0.172, 0.534, 0.295), immediate recall (0.417, 0.480, 0.103), and delayed recall (0.361, 0.464, 0.175). In parentheses are the probabilities of following a good, intermediate, and poor trajectory in that order

The contribution of the CeVP variables to the odds of group membership after adjustment for demographics, APOE-ε4, and AD neuropathology is summarized in Table 2. In the first analysis, (Table 2A) presence of microinfarcts did not contribute significantly to the odds of group membership for any of the cognitive domains considered. Presence of arteriolosclerosis contributed modestly, in that for verbal fluency, the presence of arteriolosclerosis almost doubled the odds of membership in the poor cognition trajectory when compared to the good cognition trajectory (OR = 1.95, 95% CI: 1.15-3.29). The contributions of large artery infarcts and lacunar infarcts to group membership will be summarized in the second analysis since it is based on a much larger sample size.

Table 2.

A. Adjusted odds ratios# for each type of cerebral vascular pathology when determining group membership in the first trajectory analysis (n = 1,823)
Vascular Pathology	Group membership (vs. good cognition)	Naming OR (95% CI)	Language OR (95% CI)	Immediate Recall OR (95% CI)	Delayed Recall OR (95% CI)
Lacunar infarcts	moderate cognition	1.11 (0.79–1.54)	1.36 (0.97–1.91)	1.48* (1.10–1.99)	1.26 (0.94–1.69)
	poor cognition	1.23 (0.77–1.97)	1.92ˆ (1.31–2.80)	1.54* (1.05–2.27)	1.56* (1.08–2.27)
Large artery infarcts	moderate cognition	1.70ˆ (1.18–2.47)	1.08 (0.70–1.65)	1.65* (1.14–2.40)	1.74ˆ (1.21–2.52)
	poor cognition	1.57 (0.91–2.69)	1.69* (1.07–2.68)	1.62* (1.01–2.60)	1.49 (0.93–2.40)
Arteriolosclerosis	moderate cognition	1.08 (0.68–1.69)	0.96 (0.63–1.46)	0.88 (0.59–1.31)	1.04 (0.70–1.54)
	poor cognition	1.82 (0.88–3.74)	1.95ˆ (1.15–3.29)	1.33 (0.79–2.23)	1.55 (0.93–2.59)
Microinfarcts	moderate cognition	1.45 (0.99–2.13)	1.02 (0.69–1.52)	1.13 (0.79–1.62)	1.37 (0.96–1.94)
	poor cognition	1.21 (0.68–2.17)	1.49 (0.96–2.31)	1.43 (0.90–2.27)	1.40 (0.88–2.22)

B. Adjusted odds ratios#* for lacunar and large artery infarcts when determining group membership in the second trajectory analysis (n = 2,274)
Vascular Pathology	Group membership (vs. good cognition)	Naming OR (95% CI)	Language OR (95% CI)	Immediate Recall OR (95% CI)	Delayed Recall OR (95% CI)
Lacunar infarcts	moderate cognition	1.13 (0.85–1.51)	1.29 (0.94–1.76)	1.48ˆ(1.15–1.90)	1.38* (1.06–1.79)
	poor cognition	1.38 (0.91–2.11)	2.01ˆ (1.43–2.82)	1.65 *(1.08–2.52)	1.65* (1.15–2.35)
Large artery infarcts	moderate cognition	1.78ˆ (1.28–2.47)	1.06 (0.71–1.59)	1.72ˆ (1.26–2.34)	1.78ˆ (1.29–2.47)
	poor cognition	1.63* (1.01–2.62)	1.87ˆ (1.23–2.82)	1.81*(1.10–3.00)	1.60* (1.03–2.48)

#Odds ratios adjusted for age of death, gender, years of education, APOE-ε4 carrier status, study center, and AD pathology: severe Braak stage, neuritic plaques and diffuse plaques.

^*denotes P < 0.05 while

^ˆdenotes P < 0.01.

In the second analysis (Table 2B), for language, presence of lacunar infarcts doubled the odds of belonging to the poor cognitive trajectory when compared to the good trajectory. For both immediate recall and delayed recall, the presence of lacunar infarcts significantly increased the odds of belonging to either the intermediate or the poor cognitive trajectory when compared to the good cognitive trajectory. These ORs ranged from 1.38 to 1.65 for the memory domains. On the other hand, for three of the four cognitive domains (naming, immediate and delayed recall), large artery infarcts significantly increased the odds of membership in either the intermediate or the poor cognitive trajectory when compared to the good cognitive trajectory. For the memory domains, these ORs ranged from 1.60 to 1.81. For language, large artery infarcts significantly increased the odds of membership in the poor cognition trajectory when compared to the good cognition trajectory.

Discussion

In these analyses we investigated the associations between multiple subtypes of CeVP with late-life cognition, defined as the last six years of life. This was based on a large autopsy series obtained by combining longitudinal cognition data across six different study centers. Results showed that despite the advanced age of death in these combined cohorts (mean age 89.3), 72-94% of these deaths are preceded by either an intermediate decline or very little decline in the cognitive domains: immediate or delayed recall, language, or naming. Results also showed that the presence of large infarcts and lacunar infarcts had a much stronger association with cognitive decline in late life than the presence of microinfarcts, severe arteriolosclerosis, or CAA (data not shown). Thus, indicators of CeVP appear to act differentially on different domains of late-life cognition after controlling for demographics, APOE-ε4, and markers of AD neuropathology.

Some studies of late-life cognition focus on the single cognitive assessment proximal to death, but this makes it impossible to stratify individuals who are clearly declining in cognition near death from those who are not. In a study involving 835 autopsies selected from the National Alzheimer's Coordinating Center database, Serrano-Pozo et al. (28) showed that severe vessel disease, severe amyloid angiopathy, and hippocampal sclerosis were associated with worse Clinical Dementia Rating (CDR) sum of boxes scores recorded within two years of death. Differences in findings might be due to sample sizes, earlier age of death in the NACC study (mean age 82.5), a focus on the terminal score versus the change in score near death, or the different constructs being measured by cognitive domains studied here and the CDR, which considers both cognitive and functional impairments. However, both studies confirm that it is important to account for CeVP in addition to AD neuropathology when examining late-life cognition.

In an earlier study of 437 deceased participants in HAAS, it was found that microinfarct pathology is a significant and independent factor contributing to poorer global cognition proximal to death (13). A similar finding was recorded using a database of 425 autopsies from the Religious Orders Study (12). Although both of these studies overlap with the database in this manuscript, the emphasis on the last assessment before death highlights the role of microinfarcts, whereas focusing on decline less proximal to death emphasizes the role of lacunar infarcts and large infarcts. These results are consistent with the “lacunar hypothesis” that argues that strategically placed lacunar infarcts in the frontal-subcortical loops may lead to abrupt changes in cognition and behavior (29). They are consistent with recent studies (30, 31) that conclude small vessel diseases, which includes lacunar infarcts and microinfarcts, are emerging as important mechanisms leading to cognitive decline.

Terminal cognitive declines, which refer to the relatively precipitous drop in cognitive function occurring in the years preceding death, have been investigated extensively previously (32-35). However, these studies rarely examined pathological correlates with the cognitive trajectories. Those with autopsy data are often limited to examinations of cognitive trajectories just before death. In contrast, Yu et al. (3) recently showed that after adjusting for AD pathology and some CeVP, a more social lifestyle (engagement in cognitive activities, strong social networks, and a sense of purpose in life) avoids cognitive decline proximal to death. CeVP included the presence of macroscopic infarcts, Lewy body disease, and hippocampal sclerosis; the latter two factors were rarely reported in the current study. They derived latent trajectories to summarize end of life cognitive decline using a global measure of cognition. This resulted in four latent factors: no decline, intermediate decline, severe decline, and large fluctuations in cognition. The first three of these are in general agreement with the three trajectories examined here.

Despite its large sample, this manuscript has some potential limitations. Participants included here are typically well educated and survived well into their late 80s and beyond, so the results may not be generalizable to the broader population of community-dwelling older adults in the United States. In support of the latter, some studies (36, 37) show that in late onset dementias (age 85 plus) vascular involvement is common with pure AD a minority. Not all participants at the six study sites came to autopsy, contributing further to the potential for selection bias. The choice of six years as the length of time to study cognition proximal to death could affect the results. The choice was based on capturing at least two data points to estimate cognitive changes prior to death in all participating cohorts, which meant having at least six years to accommodate the planned periodic assessment schedule that varied from one to three years across cohorts. The choice of cognitive domains to examine was dictated by the overlap in cognitive tests performed at the study sites. As a result, this study does not relate CeVP to executive function/perceptual speed. Missing data could have affected results since there are over 3,001 autopsy reports in the SMART database, but only 2,274 were used here due to fewer than two cognitive assessments late in life. Finally, we intended to study five indicators of CeVP but CAA was recorded on less than half the eligible autopsies, while both microinfarcts and arteriolosclerosis were recorded on 80% of the 2,274 cases examined here. In addition, while plaque and tangle measurements are often summarized into one recognized AD pathology severity scale (none, low, intermediate, or high AD pathology), to the best of our knowledge no similar index has been proposed for combining information across multiple CeVP measurements into one vascular severity index. Finally, while this analysis demonstrates that CeVP variables add to the effect of AD pathology on late-life cognitive decline, there is a need to conduct further research into determining how much dementia incidence could be prevented by reducing CeVP in the population.

In summary, few studies focus on the effect of vascular pathology on cognition in the very old (38). This study shows the presence of large infarcts and lacunes outweighs the effect of microinfarcts, arteriolosclerosis, and CAA in the cognition experienced by some of the old during the last six years of life. These findings would support targeting cerebrovascular factors in order to prevent late life cognitive dysfunction.