Cognitive resilience and severe Alzheimer’s disease neuropathology

Abstract

Cognitive resilience in Alzheimer’s disease (AD) can be defined as retention of high cognition despite presence of considerable cerebral AD lesions. We sought to identify factors associated with this phenomenon.

Data were obtained from National Alzheimer’s Coordinating Centre (NACC) dataset. Subjects with severe AD neuropathology, based on National Institute on Aging–Reagan (NIA-Reagan) criteria, no other primary neuropathology, and a ≤ 2-year interval between last follow-up and death were included. Mini-mental status examination score ≥ 24 was used as a proxy for normal cognition.

In total, 654 cases were included; 59 (9%) were cognitively resilient. Multivariable logistic regression model showed that resilient participants were more educated, had a lower body mass index (BMI), were more likely to be lifetime/recent smoker or use an anticoagulant/antiplatelet agent, compared with cognitively impaired subjects.

In addition to expected protective factors such as higher education and lower BMI, our results showed that smoking (especially recent smoking) and anticoagulant/antiplatelet consumption are associated with resilience to clinical cognitive expression of severe AD pathology. Pharmacological approaches using this information might be explored for clinical AD amelioration.

1. Introduction

Alzheimer’s disease (AD) is the most common neurodegenerative disease; its detrimental effects on millions of individuals worldwide, represents a global healthcare challenge [1]). With limited success in pharmacological interventions that prevent the onset of AD pathology, developing new therapeutic strategies with aims of ameliorating or delaying the clinical manifestations of AD is crucial. Furthermore, forestalling the onset of cognitive decline would result in significant public health savings [44].

The relationship between the level of AD pathology and clinical presentations is more complex than once assumed. Although there is a correlation between cognitive impairment and pathological burden, there are individuals who retain their cognitive abilities despite the presence of pathologic disease-defining lesions of AD in their brain: the term “resilience” refers to this phenomenon [4], [33], [39]. In a study by Bennet et al on 134 elderly patients with normal cognition, post-mortem assessment estimated that approximately 37.3 % of cases met the NIA-Reagan criteria for intermediate/severe AD pathology [6].

There are important gaps in our knowledge on biological mechanisms and factors that are associated with resilience to AD lesions. Several studies have reported education to be protective against clinical presentations of AD pathology, but other factors have not been properly investigated [2], [32], [34]. Most studies on resilience investigated a limited number of correlates and used neuroimaging or cerebrospinal fluid (CSF) biomarkers to estimate the level of AD pathology with only a few using post-mortem data [2], [31], [32], [34].

Another area of study in the field of AD pathology is the status of TAR DNA-binding protein of 43 kDa (TDP-43) in the resilient vs non– resilient cases. Previous studies have demonstrated that TDP-43 pathology is a major contributing factor towards the manifestation of the clinical features of AD, and TDP-positive patients are less likely to be cognitively resilient to AD pathologies [9], [19], [22].

Our aim is to expand previous studies by including a broad group of demographics, clinical, genetic, and pathological factors, and determine which are associated with cognitive resilience in subjects with severe AD pathology.

2. Materials and methods

2.1. Data source

Data for this retrospective cohort study were collected from the datasets developed by the National Alzheimer’s Coordinating Centre (NACC) between 2005 and 2017. The NACC foundation is funded by National Institute on Aging (NIA) and is responsible for developing and maintaining a database from 34 past and present Alzheimer's Disease Centers (ADCs). NACC datasets include Uniform Data Set (UDS) and Neuropathology (NP) Data Set. The UDS comprises the participants’ demographic and clinical information, including standardized clinical assessments and annual follow-ups for as long as they are able and willing to participate. This information is collected by trained clinicians and clinic personnel from participants and collateral history (usually a family member) [5].

To determine which factors are associated with cognitive resilience in AD, several demographic and clinical features at subjects’ last visit were obtained from UDS. These include age at last visit, sex, years of education, Mini-Mental State Exam (MMSE) score, recent smoking (defined as having smoked in the last 30 days prior to the assessment), pack-years of smoking, body mass index (BMI), history of lifetime smoking (defined as having smoked more than 100 cigarettes in life), alcohol abuse, hypertension, diabetes, thyroid disease, hypercholesterolemia, stroke, transient ischemic attack (TIA), congestive heart failure (CHF), heart attack/cardiac arrest, depression in the last two years and depression episodes more than two years ago. Current use of anticoagulant or antiplatelet agents, as well as antihypertensive, and diabetes medication were recorded.

AD pathology severity was evaluated by Braak & Braak staging (Stages I-VI) and Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) scores (sparse, moderate, or frequent), which assess the density of neurofibrillary tangles (NFTs) and neocortical neuritic plaques (NPs), respectively [8], [29]. These data are included in the NP Data Set.

In addition to AD pathology, NP Data Set was also used for collecting the following pathology data: presence or absence of Infarcts/lacunes, microinfarcts, large hemorrhages, microbleeds, amyloid angiopathy, neocortical Lewy body pathologies, and hippocampal sclerosis. The Research Data Dictionary-Genetic Data set (RDD-Gen) was employed to extract information on participants’ APOE ε4 status.

2.2. Subject criteria

To ensure autopsy information was related to clinical data at last visit, all subjects who did not have their last visit in the 2-year interval prior to their death were excluded from this study.

The focus of current study was on resilience to severe AD pathology; therefore, based on NIA/Reagan criteria only subjects with both frequent NPs based on CERAD score and B&B stage V/VI were selected. All other cases with less than severe AD pathology were excluded from our analyses.

Within this subgroup, cases with impaired cognition were compared with cases with intact cognition (resilient cases). Mini-Mental State Exam (MMSE) score at their last visit was employed.

for cognitive assessment [30]. Based on current literature, cases with MMSE scores ≥ 24 were defined as being cognitively intact, and thus resilient [30].

All patients with the following primary diagnoses were excluded: prion disease, trinucleotide repeat disorder (e.g., Huntington disease, spinocerebellar ataxia, other), malformation of cortical development, white matter disease (i.e., multiple sclerosis or other demyelinating diseases), amyotrophic lateral sclerosis (ALS), neoplasm (primary and metastatic), frontotemporal lobar degeneration with tau pathology.

All participants and co-participants provided written informed consent. Research using the NACC database was approved by the St. Michael’s Hospital Research Ethics Board (REB).

2.3. Statistical analysis

All statistical Analyses were performed with (Statistical Package for Social Sciences) SPSS version 25. The distribution of all variables was first assessed with bivariate tests between resilient and non-resilient cases, using χ2 test for categorical data, and either independent samples t-test or Mann-Whitney U test (based on Kolmogorov-Smirnov test of normality results) for continuous data. Multiple comparison correction was conducted using False Discovery Rate (FDR) method to adjust the P values generated from bivariate tests [17]. Next, based on bivariate analyses results, to ensure predictors do not lose significance when combined with each other, we ran a multivariable logistic regression model for all the variables that had a p value lower than 0.25. The less stringent threshold of α = 0.25 was used as variable entry level to decrease the chance of important variables being dropped from the model [16]. To adjust the predictors for sex, this variable was also included in the final model. The results of logistic regression analyses were reported as odds ratio (OR) with associated 95 % confidence intervals (CI), as well as p value. The statistical significance for the final model was assessed using α = 0.05.

The comparison of variables using bivariate tests in the first step was repeated in two separate sensitivity analyses. In the first analysis, MMSE threshold for defining resilience was increased to 25. This was to address whether our results would be affected by individuals with mild cognitive impairment that were grouped as resilient. Furthermore, to explore if our findings would change if resilient participants developed cognitive impairment symptoms after their last visit, a second analysis was performed limiting the time interval between last appointment and death to 18 months. Because of the very small sample sizes of the resilient groups, we did not use regression model analysis in these sub-studies. We also explored the MMSE score of resilient subjects in their penultimate visit to investigate whether the scores were stable or showed a change over time.

3. Results

A total of 654 participants from NACC database met our criteria for analysis and were included in our study. Of these, 59 (9 %) were categorized as cognitively resilient (Fig. 1). Table 1 and Table 2 show demographic, clinical, pathological, and genetic information for resilient and non-resilient cases in both sub-groups.

Fig. 1.

Flow chart showing the number of cases with severe AD pathology, and the breakdown into resilient and non-resilient groups.

Table 1.

Demographics and clinical correlates of subjects with and without cognitive resilience to severe AD pathology.

Variable	Braak stage V/VI and frequent neuritic plaques
	Normal cognition		Impaired cognition
	n = 59 (9 %)		n = 595 (91 %)
	N/mean	%/SD	N/mean	%/SD	p (χ2)	p (χ2) *
Age at last visit	81.4	9.8	77.7	10.7	0.005	0.04
Male	34	57.6	343	57.6	0.99	0.99
Education (y)	16.2	2.9	15.1	3.2	0.01	0.04
Lifetime smoking history	38	66.7	259	45.7	0.002	0.04
Pack-years (among smokers)	66.13	58.78	54.65	40.96	0.47	NA
Smoked recently (among smokers)	5	13.2	12	4.6	0.03	NA
Alcohol abuse
Absent	53	93	534	92.1	0.89	0.93
Recent	0	0	2	0.3
Remote	4	7	44	7.6
BMI (kg/m2)	23.8	4.4	25.6	4.5	0.008	0.04
Hypertension
Absent	24	42.1	248	42.5	0.15	0.35
Recent	32	56.1	284	48.7
Remote	1	1.8	51	8.7
Diabetes
Absent	52	91.2	505	86.6	0.54	0.7
Recent	4	7	69	11.8
Remote	1	1.8	9	1.5
Thyroid disease
Absent	40	70.2	449	76.9	0.32	0.67
Recent	12	21.1	108	18.5
Remote	5	8.8	27	4.6
Hypercholesterolemia
Absent	22	39.3	272	46.8	0.43	0.69
Recent	29	51.8	249	42.9
Remote	5	8.9	60	10.3

Variable	Braak stage V/VI and frequent neuritic plaques
	Normal cognition		Impaired cognition
	n=59 (9%)		n=595 (91%)
	N/mean	%/SD	N/mean	%/SD	p (χ2)	p (χ2)*
Stroke
Absent	51	89.5	525	90.4	0.41	0.69
Recent	0	0	12	2.1
Remote	6	10.5	44	7.6
Transient ischemic attack (TIA)
Absent	53	93	539	93.6	0.7	0.77
Recent	0	0	5	0.9
Remote	4	7	32	5.6
Congestive heart failure (CHF)
Absent	52	91.2	548	94	0.52	0.7
Recent	3	5.3	26	4.5
Remote	2	3.5	9	1.5
Heart attack/cardiac arrest
Absent	52	91.2	520	88.9	0.65	0.75
Recent	0	0	8	1.4
Remote	5	8.8	57	9.7
Depression in the last two years (Active depression)	25	44.6	293	50.3	0.42	0.69
Depression episodes more than two years ago (Remote depression)	10	17.5	170	29.5	0.05	0.15
Use of an anticoagulant or antiplatelet agent at last visit	32	55.2	221	38.2	0.01	0.04
Use of any type of an antihypertensive medication at last visit	37	63.8	312	53.9	0.14	0.35
Use of a diabetes medication at last visit	4	6.9	52	9	0.59	0.72

Table 2.

Non-AD pathological and genetic correlates of subjects with and without cognitive resilience to severe AD pathology.

Variable	Braak stage V/VI and frequent neuritic plaques
	Normal cognition		Impaired cognition
	n = 59 (9 %)		n = 595 (91 %)
	N/mean	%/SD	N/mean	%/SD	p (χ2)	p (χ2)*
Neocortical Lewy body pathology	7	11.9	76	12.8	0.83	0.91
Infarct and lacunes	12	20.3	109	18.4	0.71	0.87
Microinfarct	10	16.9	111	18.7	0.74	0.78
Hemorrhages and microbleeds	5	8.5	32	5.5	0.34	0.72
Hippocampal sclerosis	2	10.5	15	10.9	0.96	0.96
Amyloid angiopathy	41	70.7	457	80	0.09	0.54
APOE ε4 carrier	34	60.7	336	63	0.73	0.81

Mean age at last visit and years of education were both higher for resilient than non-resilient cases (p value = 0.01). Conversely, distribution of gender was similar between resilient groups and their non-resilient counterparts (Table 1 and Table 2).

We investigated three smoking-related variables: having a lifetime smoking history (defined as having smoked more than 100 cigarettes in life), pack-years of smoking, and recent smoking history (defined as having smoked in the last 30 days). A higher percentage of smokers were found to be present in the asymptomatic group in comparison to their non-resilient counterparts (66.7 % versus 45.7 %, p value=0.002). The two other smoking variables, number of pack-years and history of recent smoking, were compared among participants with a lifetime smoking history only. Our results indicate that the number of participants with a recent smoking history was found to be significantly higher in the resilient than the non-resilient group (13.2 % versus 4.6 %, p value=0.003). Packs-year numbers were similar between the two groups (p value=0.47).

In terms of medications, resilient individuals were more likely to have consumed an anticoagulant or antiplatelet agent at last visit, compared to the cognitively impaired cohort (Table 1 and Table 2). BMI was significantly lower in resilient subjects (23.8 (kg/m2) versus 25.6 (kg/m2), p value=0.006).

Next, based on the univariable analysis results, we developed a multivariable model to estimate the risk of being cognitively resilient while controlling for multiple predictors. Results are shown in Tables 3 and Fig. 2. Based on our final model, odds of being cognitively resilient to frequent NPs and B&B stage V or VI were increased significantly for subjects who had more education, had a lifetime or recent smoking history, had lower BMI, and used an anticoagulant or antiplatelet agent at their last visit (Table 3).

Table 3.

Multivariable logistic regression model for odds of being cognitively resilient to Braak stage V/VI and frequent neuritic plaques pathologies adjusted for demographics (n = 467).

Predictor	OR (95 % CI)	p Value
Age at last visit (yrs)	1.03 (0.99–1.07)	0.054
Sex (male)	1.35 (0.67–2.68)	0.39
Education (yrs)	1.16 (1.03–1.3)	0.005
BMI (kg/m2)	0.91 (0.84–0.99)	0.02
Lifetime smoker	2.35 (1.2–4.6)	0.01
Recent smoking	4.76 (1.12–20.14)	0.03
Use of an anticoagulant or antiplatelet agent at last visit	2.15 (1.13–4.9)	0.02
Depression episodes more than two years ago	0.53 (0.24–1.19)	0.12

Fig. 2.

Odds ratio plot of factors associated with cognitive resilience to severe AD pathology based on the binary logistic regression model.

Out of 59 resilient cases, 42 subjects had more than one visit. MMSE scores of the last two visits in this group are shown in Fig. 3. Twenty-nine (69 %) subjects had MMSE scores that were 0–2 points lower in their last visit compared to the previous visit. In addition, 4 (9.5 %) subjects showed a decline of more than 2 points between the two visits. In contrast, nine (21.4 %) participants demonstrated higher MMSE scores in their last visit, compared to the penultimate follow up.

Fig. 3.

MMSE scores in the last two follow-up visits in the resilient cases with more than one visit. Each floating bar demonstrates the two MMSE scores in the last two visits with the length of each bar representing the change between the said visits. The bars with patterns represent increased scores in the last visit compared to the previous visit.

The findings from the two separate sensitivity analyses yielded similar results in most circumstances, supporting the likelihood that these are robust findings. However, use of an anticoagulant/antiplatelet agent at last visit and recent smoking were each significantly associated with resilience in one of the two sensitivity analyses (i.e., mutually exclusive; Supplementary Table 1, Table 2, Table 3).

4. Discussion

We found a much lower prevalence (9 %) of cognitive resilience than Bennet et al. (37.3 %), likely because of our focus on severe Alzheimer’s disease pathology only [6]. Using the literature supported, but arbitrary cut-off of MMSE 24, our analysis indicates that higher education, lower BMI, Lifetime smoking history, recent smoking, and use of an anticoagulant or antiplatelet agent at last visit were independently associated with higher odds of being resilient to severe AD lesion load. These results mostly held up the challenge of two separate sensitivity analysis using more stringent criteria.

Education, as a protective factor against AD pathology, has been a widely accepted resilience element in the current literature, and our results confirm a similar association [2], [7], [11], [13]. It is hypothesized that people with higher education have the ability to use their brain networks in a more efficient way (cognitive reserve theory) [38].). As a result, a higher level of education increases the threshold for dementia in people with AD pathology.

Our findings indicate that cognitively normal subjects’ age was not significantly different from that of cases with impaired cognition after adjusting for other predictors in the model. Although increasing age is known as one of the strongest risk factors for AD, our finding is in line with previous studies that did not find any significant association between age and cognitive resilience [2], [32].

In the multivariable model, odds of being resilient was 2.4 times higher for smokers and 4.7 times higher for recent smokers but pack-years was not significantly different between the two groups. Furthermore, recent smoking was also the strongest independent predictor amongst all factors studied in the model, which may suggest a short-term effect of smoking on cognition. Although literature indicates that smoking is related to an increased risk for AD [35], [42]), previous studies did not suggest any association between smoking and cognitive resilience [2], [41]. It is now well documented in numerous studies that there is a robust inverse association between risk of Parkinson’s disease (PD) and smoking duration and intensity [14], [28]). However, the chemical compounds in tobacco responsible for this effect have yet to be defined. Several substances in smoke have been hypothesized to have a role in reducing PD risk, and it is possible the same component might be involved in protection against the cognitive manifestations of AD pathology [14], [37]. Considering cholinesterase inhibitors have been used to manage symptoms of AD, an interesting possibility is that nicotine itself is exerting the neuroprotective effect through its cholinomimetic nature. Further studies on a larger scale that include more diverse populations are needed to confirm this association. Based on our analysis, it may be reasonable to suggest that the effect of smoking on cognitive resilience is non-cumulative-dose dependant; having smoked in the last 30 days prior to the assessment can increase the chance of being resilient, regardless of the number of pack-years. Smoking is a major risk factor for many non-communicable diseases and multiple neoplasms [36]. Overall, the protective effect of tobacco will not exceed its risks, and it should never be recommended as a strategy for increasing the likelihood of resilience to AD. However, elucidating the nature of this relationship may provide insight into the biological mechanisms of resilience in AD and contribute to developing potential pharmacological targets for both treatment and preventive measures.

Finally, the possibility that subjects would change their smoking behaviour due to dementia must be considered. However, changes in smoking behaviour in this age group are uncommon, and smoking cessation secondary to cognitive decline has not been documented.

Our results suggest that people with higher BMI were at greater risk of impaired cognition. Subjects with impaired cognition had a mean BMI of 25.6 (kg/m2), which is above the standard normal range according to global guidelines, and BMI in this group is significantly higher than in the resilient group which had a mean BMI in the normal range (23.8) [18]. This could be related to the association of increased risk of AD with lipid dysregulation and chronic inflammation [25]. Obesity especially in mid-life has been proposed by several studies as an independent risk factor for developing dementia in older age, but its role in resilience is far from clear [15], [20], [26]. Larger cohort studies are still needed to understand the mechanisms underlying this association.

In our analysis there was no association between non-AD pathologies and resilience. Consistent with our findings, Bowles et al. showed that, in their 2019 study on participants from the Adult Changes in Thought (ACT) study, non-AD pathologies decreased the chance of resilience in subjects with intermediate or high AD pathology, but less so in high AD pathology-only cases [2]. In a study by Launer et al. on the Honolulu Asia Aging Autopsy Study (HAAS), brain microinfarcts were independently associated with cognitive function [24]. However, this association was much stronger in the non-demented group than in patients with dementia. In another study by Esiri et al. on the relationship between Braak stages of AD and subcortical lacunes, the association of cerebrovascular pathology with impaired cognition was robust at the early stages of AD only [12]. It is likely that non-AD pathologies’ role in cognition is much more significant early in the development of AD lesions than in the presence of a heavy burden of plaques and tangles. These discrepancies with previous studies could also be attributed to different methods in pathologic interpretation of lesions. Our database has recorded these pathologies as present or absent without indicating the severity, which may have obscured a true relationship. Interestingly in the study by Monsell in 2013 using NACC database, the association between non-AD pathologies and resilience was not significant, which is similar to results of the current study [31]. The TDP-43 proteinopathy variable was added recently to the NACC database and as a result, the majority of cases were lacking the data for this variable, so it was not included in the analyses. However, given the fact that a significant association between cognitive impairment in AD patients and TDP-43 pathology has been previously demonstrated, future investigation of the role of TDP-43 pathology in the phenomenon of cognitive resilience to severe AD pathology is warranted [9], [19], [22].

A larger portion of asymptomatic cases used an anticoagulant/antiplatelet at their last visit. To our knowledge, no other studies have investigated the medication use in asymptomatic groups with AD pathology. A possible explanation might be that many of the medications used in this group, including aspirin, also have anti-inflammatory effects, and it has been suggested by previous studies that use of anti-inflammatory agents may protect subjects against the clinical onset of AD [3], [27], [40].). This protection possibly involves their ability of interfering with the chronic inflammatory response that commences decades before onset of dementia and continues through it [27].

The difference in proportion between symptomatic and asymptomatic groups who had ≥ 1 APOE ε4 allele, was not statistically significant. APOE is a well-established genetic risk factor for developing AD but its role in resilience is not consistent in literature [2], [23], [31]. It is reasonable to suggest APOE ε4 is associated with higher neuropathological load but does not have a major effect on resilience [23], [32]. Although the number of remote depression episodes were lower in resilient cases, the difference did not remain statistically significant after being combined with other predictors in the model. This could possibly be explained by the relatively small sample size of resilient cases. Several studies have suggested a link between clinical onset of AD and depressive symptoms; depression may hasten the clinical progression of AD [10], [43].

Before drawing conclusions from the data, limitations of the study must be addressed. We used the arbitrary MMSE score cut-off of 24 for separating asymptomatic from symptomatic subjects. Although cognition is inherently a continuous quantity and dividing it is not ideal, this cut-off has been used in previous studies and seems clinically appropriate [21]. Nevertheless, using this cut-off might not detect subjects with mild cognitive impairments (MCI) [30]. It will be helpful for future studies to incorporate other cognitive tests and have more extensive examinations to exclude cases with MCI. NACC database gathers information from multiple sites across USA. However, individuals enrolled in our study are mainly Caucasian and have high levels of education. This points out to another limitation to all studies using NACC data which is lack of complete randomization. Further studies on more diverse sample groups are needed to establish our findings.

On the other hand, the main strength of our resilience study is that we used autopsy data for neuropathological diagnosis of AD. To our knowledge, there have been only a few studies on resilience that utilized post-mortem data. Our inclusion requirement of advanced CREAD and Braak’s stages focuses the study on resilience to severe Alzheimer’s disease. Also, we examined a large group of demographic, clinical and pathological correlates to identify which factor could be protective against severe AD lesions.

Of note, using post-mortem data limits our ability to follow up on these patients, and determine if the resilient cases continue to have normal cognition, or are in the “pre-clinical” stage of AD. However, based on the MMSE scores at the last two visits for resilient subjects, the vast majority showed either a small decrease or an increase in their scores, whereas in less than 10 % was the decline large enough to suggest an early stage of cognitive collapse. Considering that the mean time interval between last visit and death in our resilient cases is less than a year (10.1 months, to be precise), it is reasonable to suggest that it would take years for most of the resilient cases to reach MMSE scores of significantly impaired cognition.

Also, given that resilience to severe AD pathology is observed rather uncommonly, it would be helpful to re-investigate the resilience data in the future, as more cognitively resilient AD cases are added to the NACC database. The consequent sample size increase can augment power of future studies to detect statistically significant differences between resilient and non-resilient groups.

In conclusion, the protective effect of smoking, and especially recent smoking, suggests that developing pharmacological mimics could be explored as a symptomatic treatment in the future.

Verification

We confirm that this manuscript has not been published elsewhere and is not under consideration by any other journal.

Disclosure statement

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

CRediT authorship contribution statement

Narges Ahangari: Writing – original draft, Methodology, Formal analysis, Writing – review & editing, Visualization. Corinne E. Fischer: Supervision, Writing – review & editing, Project administration. Tom A. Schweizer: Supervision, Writing – review & editing, Project administration. David G. Munoz: Conceptualization, Supervision, Writing – review & editing, Project administration.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.