Longitudinal assessment of cognitive decline and resilience in high-level Alzheimer disease neuropathologic change

Abstract

Background

Alzheimer disease neuropathologic change (ADNC) is the most common pathology underlying cognitive impairment and dementia in the aging population, but there is significant variation in outcome between affected individuals. Moreover, other common neurodegenerative processes are often concurrent and may significantly worsen cognition, but the degree to which these processes interact and affect the rate of cognitive decline remains unclear. Herein, we aim to investigate features influencing cognitive trajectories over the final 15 years of life in individuals with high-level ADNC.

Methods

We performed a cross-sectional cohort study of 586 participants from the National Alzheimer’s Coordinating Center (NACC) database, who were ≥ 65 years of age and displayed high-level ADNC at autopsy, and who had available longitudinal cognitive data and Clinical Dementia Rating (CDR) performed within the final 24 months of life. This cohort was subdivided into “resilient” individuals/those with minimal progression of cognitive decline (MinP; n = 75), intermediate/moderate progression of cognitive decline (ModP; n = 255), and rapid/maximal progression of cognitive decline (MaxP; n = 256) as determined by global cognitive performance and the rate of cognitive decline. Demographic, neuropathologic, genetic, and clinical features were evaluated using multivariable logistic regression analysis.

Results

Individuals with rapid progression were more likely to have at least one APOE ε4 allele (OR: 2.08 [95% CI: 1.16–3.74], p < 0.01), higher Braak stage (2.19 [1.20–3.98], p < 0.01), higher Thal phase (2.45 [1.24–4.83], p < 0.01), more severe white matter rarefaction (1.68 [1.21–2.35], p < 0.01), and in the final 24 months of life, more frequent untreated/undertreated hyperlipidemia (1.74 [1.35–5.56], p < 0.01) and less frequent untreated/undertreated depression (0.40 [0.17–0.91], p < 0.05). Conversely, resilient individuals harbored less frequent APOE ε4 alleles (0.17 [0.06–0.55], p < 0.01), lower Thal phase (0.33 [0.12–0.95], p < 0.05), lower CERAD neuritic plaque score (0.32 [0.11–0.97], p < 0.05), less frequent untreated/undertreated psychosis (0.06 [0.01–0.39], p < 0.01), and more frequent untreated/undertreated depression (8.73 [2.12–35.85], p < 0.01).

Conclusions

These data suggest that resilience and progression in ADNC are impacted by AD-relevant genetics and the severity of late-stage ADNC (even within the narrow range of values compatible with high-level ADNC), additional pathologic features, and potentially the clinical management of underlying systemic disorders.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13195-025-01904-6.

Introduction

Alzheimer disease (AD) was initially described by Alois Alzheimer in the early 20th century [1]. It constitutes a severe global health concern, affecting an estimated 7.2 million Americans aged 65 and older, and has been the 7th leading cause of death since 2020 [2]. Alzheimer disease neuropathologic change (ADNC), the pathologic substrate of AD, is thought to be the single most common pathology underlying clinical dementia [3]. ADNC is defined by the presence and extent of two interrelated proteins, hyperphosphorylated tau (p-tau) and β-amyloid (Aβ). The earliest deposition of p-tau is thought to precede Aβ, beginning in the brainstem before involving the entorhinal cortex, hippocampus and other limbic structures, and neocortex in six hierarchical Braak stages [4–6], although alternative p-tau progression patterns have been proposed, particularly in the medial temporal lobe in the absence of Aβ [7–11]. Aβ is first deposited in the neocortex, with progression through the entorhinal cortex, hippocampus, basal ganglia, brainstem, and cerebellum in five progressive Thal phases [12]. Additionally, the neocortical density of neuritic plaques (NP), composed of both Aβ and p-tau, is measured with CERAD criteria [13], and these three components are integrated to determine the overall level of ADNC [14, 15]. While cognitive decline and amnestic dementia are associated with this overall level of ADNC, studies have demonstrated that they correlate best with Braak stage and CERAD NP score [16, 17], although Aβ progression is not benign in the context of aging and AD [18, 19].

There is significant heterogeneity in individual clinical metrics, including overall cognition, the rate of cognitive decline, and the specific cognitive domains affected by this underlying pathology. Additionally, there are “resilient” individuals, who are able to maintain normal cognition despite significant amounts of ADNC, and “resistant” individuals, who have normal cognition and lack the degree of ADNC which would be expected at their respective age [20–23]. More recently, it has been demonstrated that contributors to the symptoms classically associated with clinical AD also include the presence of other neurodegenerative pathologies (including various forms of cerebrovascular disease [CVD], Lewy body disease [LBD], and limbic-predominant age-related TDP-43 encephalopathy [LATE]), which are frequently comorbid with ADNC and worsen overall cognition in these individuals [17, 24–31], although there are conflicting results as to the effect of co-pathologies on the rate of cognitive decline [32–34]. A degree of “resilience” to ADNC may also simply be due to “resistance” to some of these other comorbid neurodegenerative processes [35, 36].

To better resolve the clinical heterogeneity in ADNC, we selected 586 participants from the National Alzheimer’s Coordinating Center (NACC) database with high-level ADNC (Braak stage V-VI, Thal phase 4–5, and CERAD NP score “moderate” or “frequent”). These individuals were divided into those who could be termed as “resilient” with minimal progression of cognitive decline (MinP), those with moderate progression of cognitive decline (ModP), and those with rapid and maximal progression of cognitive decline (MaxP), irrespective of other neuropathologic or clinical features. Using this framework, we investigated the cognitive/neuropsychological, genetic, neuropathologic, and systemic factors which may influence the progression of cognitive symptoms in these individuals.

Methods

National Alzheimer’s Coordinating Center (NACC) dataset and case selection 

Data for this study were downloaded with permission from the NACC, established with funding from the National Institute on Aging (NIA) (U01 AG016976) (https://naccdata.org/), and sourced from 37 Alzheimer’s Disease Research Centers (ADRC) located across the United States. NACC data, downloaded on November 25, 2024 (Data Request Number 13611), included standardized Uniform Data Set (UDS) version 3 variables (https://naccdata.org/data-collection/forms-documentation/uds-3), Neuropathology (NP) Data Set version 11 variables (https://naccdata.org/data-collection/forms-documentation/np-11), and Genetic Data Set variables (https://files.alz.washington.edu/documentation/rdd-genetic-data.pdf), as previously described [17, 24, 32, 37, 38]. All participants provided written informed consent at each ARDC of origin, and study protocols were approved by the institutional review boards (IRBs) at each institution.

The criteria for case selection were high-level ADNC [14, 15], age at death ≥ 65 years, at least 3 cognitive exams during which global Clinical Dementia Rating (CDR) and CDR Sum of Boxes (CDR-SB) were assessed, final cognitive exam within the final 24 months of life, and global CDR ≤ 0.5 at first clinical exam (Supplemental Figure S1). 586 cases met these criteria and were further subdivided into three groups based on cognitive trajectories [39]; (1) resilient participants/those with minimal progression (MinP), defined as those with global CDR of 0–0.5.5 at their final clinical exam (n = 75), (2) participants with intermediate/moderate progression (ModP), defined as those with global CDR of 1–2 at their final clinical exam (n = 255), and (3) participants with maximal progression (MaxP), defined as those with global CDR of 3 at their final clinical exam (n = 256) (Fig. 1). We then sought to evaluate wide ranging differences between these three groups, using a systematic approach guided by data availability.

Fig. 1.

Cognitive trajectories for individual participants with minimal progression (MinP), moderate progression (ModP), and rapid/maximal progression (MaxP) for A Clinical Dementia Rating Sum of Boxes (CDR-SB) and B Mini-Mental State Examination (MMSE)

Demographic, genetic, and neuropathologic variables 

Demographic and genetic variables for each individual were extracted from the UDS and genetic data sets, and included participants’ sex (UDS variable SEX), age at cognitive exam (NACCAGE), age at death (NACCDAGE), education (EDUC), race (RACE), and APOE genotype (NACCAPOE). The time between each participant’s death and a given cognitive exam was calculated using variables for month of death (NACCMOD), year of death (NACCYOD), visit month (VISITMO), and visit year (VISITYR).

Each neurodegenerative pathology was defined using NACC NP dataset variables, as previously described in detail [17, 24]. ADNC level was determined from the variable NPADNC and confirmed using Braak stage (NACCBRAA), Thal phase (NPTHAL), and CERAD NP score (NACCNEUR). LBD stage was assessed using the variables NACCLEWY and NPLBOD (no Lewy body pathology, brainstem-predominant, limbic [transitional], or diffuse neocortical). LATE stage was derived from the variables NPTDPB, NPTDPC, and NPTDPE (as well as NPFTDTDP and NPALSMND), using updated recommendations for distinguishing LATE from other pathologies with TDP-43 (no LATE neuropathologic change, amygdala only, +hippocampus, or + middle frontal gyrus) [31, 40]. Hippocampal sclerosis was determined with the variable NPHIPSCL (none or unilateral/bilateral/laterality not assessed). Aging-related tau astrogliopathy (ARTAG) was determined with NPARTAG (absent or present). Various measures of cerebrovascular disease were evaluated with the NACC NP variables for infarcts/lacunes (NACCINF) (absent or present), microinfarcts (NACCMICR) (absent or present), hemorrhages/microbleeds (NACCHEM) (absent or present), arteriolosclerosis (NACCARTE) (none, mild, moderate, or severe), and white matter rarefaction (NPWMR) (none, mild, moderate, or severe) [17, 24, 36]. Cerebral amyloid angiopathy (CAA) was determined with the NACC NP dataset variable NACCAMY (none, mild, moderate, or severe).

Cognitive and neuropsychological variables 

Longitudinal cognitive and neuropsychological testing was assessed using UDS variables. These included global CDR (CDRGLOB), CDR Sum of Boxes (CDR-SB) (UDS variable CDRSUM), Mini-Mental State Examination (MMSE) (UDS variable NACCMMSE), logical memory immediate recall (LMI) (UDS variable LOGIMEM), logical memory delayed recall (LMD) (UDS variable MEMUNITS), time elapsed since LMI (UDS variable MEMTIME), digit span forward (DSF) (UDS variable DIGIF), digit span backward (DSB) (UDS variable DIGIB), Trail Making Test Part A (TMT-A) (UDS variable TRAILA), Trail Making Test Part B (TMT-B) (UDS variable TRAILB), Wechsler Adult Intelligence Scale Digit Symbol Substitution Test (WAIS DS) (UDS variable WAIS), animal naming fluency (ANIMALS), vegetable naming fluency (VEG), and Boston Naming Test, 30 odd items (BNT) (UDS variable BOSTON), as previously described in detail [17, 24, 32, 41–48].

Psychiatric and behavioral variables

 Psychiatric and behavioral features were derived from UDS variables of the Neuropsychiatric Inventory-Questionnaire (NPI-Q) and included depression or dysphoria (DEPD) and active depression within the last two years (DEP2YRS), delusions (DEL), hallucinations (HALL), apathy or indifference (APA), disinhibition (DISN), nighttime behaviors (NITE), anxiety (ANX), agitation or aggression (AGIT), irritability or lability (IRR), appetite and eating problems (APP), and motor disturbance (MOT) [24].

Clinician-assessed medical conditions and medication variables

 Systemic medical conditions and medications were assessed from UDS variables, including presence of dementia (DEMENTED), multidomain amnestic dementia (AMNDEM), primary progressive aphasia (PPA) (UDS variable NACCPPA), posterior cortical atrophy (PCA), history of cancer in the previous 12 months (CANCER), diabetes (DIABET), hypertension (HYPERT), hyperlipidemia (HYPCHOL), vitamin B12 deficiency (B12DEF), thyroid disease (THYDIS), myocardial infarction (MYOINF), congestive heart failure (CONGHRT), atrial fibrillation (AFIBRILL), total number of medications (NACCAMD), antihypertensives (NACCAHTN), angiotensin converting enzyme inhibitors (ACE-I) (UDS variable NACCACEI), beta-blockers (NACCBETA), calcium channel blockers (NACCCCBS), diabetes medications (NACCDBMD), diuretics (NACCDIUR), angiotensin II inhibitors (NACCANGI), lipid lowering medications (NACCLIPL), nonsteroidal anti-inflammatory drugs (NSAIDs) (UDS variable NACCNSD), anticoagulants (NACCAC), antidepressants (NACCADEP), antipsychotics (NACCAPSY), AD symptomatic therapies (NACCADMD), antiparkinsonian agents (NACCPDMD), and diabetes medications (NACCDBMD). Unmedicated systemic conditions were defined as the presence of a particular condition in the absence of therapy (e.g., positive for hyperlipidemia and negative for lipid-lowering therapy).

Statistics and data analysis 

Mixed-effects regression modeling with random intercepts and random slopes for time per participant was used to compare neuropsychological test trajectories over time between MinP, ModP, and MaxP groups. Cognitive and neuropsychological tests were considered dependent variables, while age and group were included as fixed effects to test for overall differences between groups. To test for slope differences, an interaction group*time was included in the model, as previously described [32, 48, 49]. In R, the “readxl” package was used to import data, the “dplyr” package was used for data wrangling, the “lme4” package was used to fit mixed-effects models that account for repeated measurements from the same participant over time, the “lmerTest” package was used to provide significance testing for the mixed effects models, the “emmeans” package was used for post-hoc analysis, and the “broom.mixed” package was used for output formatting. The “lstrends” function from the “emmeans” package was used to compare pairwise slopes and the “broom.mixed” package was used for output formatting. Pairwise comparisons of slopes between outcome groups were obtained using the emtrends and pairsfunctions from the “emmeans” package, with p-values adjusted for multiple comparisons using Tukey’s method.

All univariate analysis was performed using GraphPad Prism version 10 (GraphPad Software, Inc., La Jolla, CA, USA). Differences in categorical variables (sex, race, APOE status, frequency of neuropathologic and clinical features, etc.) between MinP, ModP, and MaxP groups were calculated using Chi-squared test. Differences in continuous variables (age, years of education, etc.) between the three groups were evaluated using multiple t-tests. P-values for these tests were adjusted for multiple testing using false discovery rate (FDR) correction with two-stage linear step-up procedure of Benjamini, Krieger, and Yekutieli method. Multivariable logistic regression analysis was performed with MedCalc (MedCalc Software Ltd, Ostend, Belgium) [17, 48], with outcome measures (odds ratios of resilience [MinP] or rapid progression [MaxP]) and predictive factors (including demographic, genetic, neuropathologic, and systemic disease states/medication).

Results

Demographic features of the included cohort

 Of the 586 participants with high-level ADNC, 296 were male (50.5%) and 290 were female (49.5%). 557 of the participants were white (95.2%). The overall cohort had a mean education level of 16.3 ± 0.2 years and mean age of death of 82.9 ± 0.3 years. The cohort averaged 7.3 ± 0.1 cognitive exams per participant, with an average time of 7.4 ± 0.1 years between the first cognitive exam and participant death, and 0.8 ± 0.02 years between final cognitive exam and participant death. There were differences in the time between cognitive evaluations and death between the three groups, although the MinP group had the longest mean time between the initial evaluation and death and the most number of total cognitive evaluations. While the MinP group had the longest mean time between the final cognitive evaluation and death, this difference amounted to less than 3 months (Table 1).

Table 1.

Demographic, genetic, and neuropathologic features

	Minimal Progression	Moderate Progression	Maximal Progression	p-value^ (MinP vs. ModP)		p-value^ (MinP vs. MaxP)		p-value^ (ModP vs. MaxP)
Demographic Features
n	75	255	256	-		-		-
Age	87.9 ± 0.8	83.3 ± 0.5	81.0 ± 0.5	< 0.0001		< 0.0001		0.0012
Sex (M|F)	46.7% (35)|53.3% (40)	57.6% (147)|42.4% (108)	44.5% (114)|55.5% (142)	0.0928		0.7437		0.0030
Education Level (years)	16.5 ± 0.4	16.1 ± 0.4	16.5 ± 0.4	0.6035		> 0.9999		0.4798
Race (% white)*	96.0% (72/75)	95.7% (244/255)	94.1% (241/256)	0.9057		0.5323		0.4266
Total cognitive evaluations	7.9 ± 0.4	6.8 ± 0.2	7.6 ± 0.2	0.0029		0.3602		0.0009
Time between first evaluation and death (years)	8.0 ± 0.4	6.6 ± 0.2	7.9 ± 0.2	0.0017		0.7927		< 0.0001
Time between cognitive impairment conversion and death (years)	4.5 ± 0.1†	3.1 ± 0.1	5.2 ± 0.2	< 0.0001		0.0003		< 0.0001
Time between final evaluation and death (years)	0.99 ± 0.06	0.77 ± 0.03	0.78 ± 0.03	0.0007		0.0012		0.8138
ADNC Pathology
Braak Stage (5|6)	70.7% (53)|29.3% (22)	46.3% (118)|53.7% (137)	28.9% (74)|71.1% (182)	0.0002		< 0.0001		< 0.0001
Thal Phase (4|5)	49.3% (37)|50.7% (38)	30.2% (77)|69.8% (178)	17.6% (45)|82.4% (211)	0.0022		< 0.0001		0.0008
CERAD (Moderate|Frequent)	38.7% (29)|61.3% (46)	21.6% (55)|78.4% (200)	18.4% (47)|81.6% (209)	0.0028		0.0002		0.3641
CAA (0|1|2|3)	24.3% (18)|29.7% (22)|23.0% (17)|23.0% (17)	18.8% (48)|36.9% (94)|28.6% (73)|15.7% (40)	16.0% (41)|39.1% (100)|24.2% (62)|20.7% (53)	0.2487		0.2863		0.3503
Other Neuropathologic Features
LBD Stage (0|1|2|3)	65.8% (48)|8.2% (6)|21.9% (16)|4.1% (3)	61.0% (152)|1.6% (4)|24.9% (62)|12.4% (31)	55.6% (139)|0.8% (2)|30.0% (75)|13.6% (34)	0.0071		0.0002		0.4544
LATE Stage (0|1|2|3)	68.4% (39)|5.3% (3)|24.6% (14)|1.8% (1)	54.7% (111)|12.3% (25)|26.6% (54)|6.4% (13)	59.5% (88)|17.6% (26)|17.6% (26)|5.4% (8)	0.1467		0.0652		0.1593
Hippocampal Sclerosis	7.6% (6/73)	11.3% (32/251)	12.7% (37/255)	0.2897		0.1603		0.5639
ARTAG	60.0% (12/20)	44.9% (31/69)	34.8% (23/66)	0.2350		0.1449		0.6652
Gross Infarct	16.0% (12/75)	12.6% (32/254)	11.4% (29/255)	0.4470		0.2855		0.6703
Microinfarcts	14.9% (11/74)	24.3% (62/255)	17.3% (44/255)	0.0850		0.6276		0.0495
Gross Hemorrhage	1.4% (1/73)	3.2% (8/253)	3.1% (8/255)	0.4103		0.4150		0.9872
Arteriolosclerosis (0|1|2|3)	15.7% (11)|44.3% (31)|27.1% (19)|12.9% (9)	16.5% (39)|33.3% (79)|33.3% (79)|16.9% (40)	14.8% (35)|31.8% (75)|38.1% (90)|15.3% (36)	0.3880		0.1413		0.2800
WMR (0|1|2|3)	50.8% (31)|19.7% (12)|21.3% (13)|8.2% (5)	38.0% (92)|37.6% (91)|15.7% (38)|8.7% (21)	30.0% (73)|29.2% (71)|27.2% (66)|13.6% (33)	0.0576		0.0228		0.0019
Genetic Features
APOE Status
≥ 1 ε2 Allele	22.3% (15/67)	12.3% (30/243)	7.1% (17/240)	0.0388		0.0003		0.0511
≥ 1 ε4 Allele	37.3% (25/67)	57.2% (139/243)	66.7% (160/240)	0.0039		< 0.0001		0.0322
ε2/2 | ε2/3 | ε2/4 | ε3/3 | ε3/4 | ε4/4	1.5% (1)|13.4% (9)|7.5% (5)|47.8% (32)|25.4% (17)|4.5% (3)	0% (0)|6.6% (16)|5.8% (14)|36.2% (88)|42.0% (102)|9.5% (23)	0.4% (1)|2.5% (6)|4.2% (10)|30.4% (73)|45.0% (108)|17.5% (42)	0.0152		< 0.0001		0.0206

*Given the paucity of non-white participants, statistics for race are calculated as “white” and “non-white”; 

†Cognitive impairment was by definition not reached for minimal progression group, therefore the commensurate cognitive exam was used for this measure

^p-values were adjusted for multiple comparisons with the Benjamini, Krieger, and Yekutieli method, p-values less than 0.0079 are considered significant after false discovery rate (FDR) adjustment

The overall cohort had a mean CDR-SB of 12.2 ± 0.2 and an MMSE of 18.2 ± 0.4 at their final cognitive exam. Between the three cognitively-defined groups of high-level ADNC participants, there were no significant differences in participant race or level of education after adjusting for multiple comparisons; however, the MinP participants were significantly older than either the ModP or MaxP participants, and the MaxP participants were significantly younger than the ModP participants (Table 1). Additionally, there were no significant differences between the MinP, ModP, and MaxP groups and timeframe for year of death or year of cognitive evaluation.

Cognitive features 

While there was considerable variation between individuals in terms of the timing and trajectory of global cognitive decline (Fig. 1A and B), there was clear separation in the smoothed mean cognitive trajectories of each of the three groups (Fig. 2A and B). As expected based on the experimental design, there were significant differences in the rate of cognitive decline (slope) in mixed-effects modeling of both CDR-SB and MMSE between all three groups (Supplemental Figure S2 and Supplemental Table S1). The MinP participants were more likely to be determined by their clinician to have normal cognition/behavior at their final cognitive exam compared to ModP and MaxP participants and less likely to have clinical diagnoses of dementia and multidomain amnestic dementia. No significant differences were found in frequency of posterior cortical atrophy (PCA) or primary progressive aphasia (PPA) diagnoses (Table 2).

Fig. 2.

Smoothed cognitive and neuropsychological trajectories for participants with minimal progression (MinP), moderate progression (ModP), and rapid/maximal progression (MaxP), including A Clinical Dementia Rating Sum of Boxes (CDR-SB), B Mini-Mental State Examination (MMSE), C logical memory immediate recall (LMI), D logical memory delayed recall (LMD), E digit span forward (DSF), F digit span backward (DSB), G trail making test part A (TMT-A), H trail making test part B (TMT-B), I WAIS digit substitution test (WAIS DS), J animal naming test (Animals), K vegetable naming test (Vegetables), and L Boston naming test (BNT). Mixed effects models for each test are displayed in Supplemental Figure S2 and statistical comparisons of rate of decline for each test are noted in Supplemental Table S1

Table 2.

Clinical features of participants with minimal, moderate, and maximal cognitive progression

	Minimal Progression	Moderate Progression	Maximal Progression	p-value* (MinP vs. ModP)	p-value* (MinP vs. MaxP)	p-value* (ModP vs. MaxP)
Medical & Systemic Factors
Lifetime tobacco use	35.0% (7/20)	50.0% (30/60)	53.8% (28/52)	0.2439	0.1518	0.5926
Cancer history	29.1% (16/55)	19.5% (37/190)	19.8% (40/202)	0.1271	0.1390	0.9348
Diabetes	12.7% (7/55)	11.1% (21/189)	8.4% (17/203)	0.6486	0.3242	0.4554
Myocardial infarction	3.6% (2/55)	7.3% (14/191)	3.9% (8/203)	0.3277	0.9173	0.1431
Congestive heart failure	7.3% (4/55)	7.4% (14/190)	3.0% (6/202)	0.9809	0.1435	0.0493
Atrial fibrillation	29.6% (16/54)	18.0% (34/189)	11.4% (23/201)	< 0.0001	0.0010	0.0006
Hypertension	51.8% (28/54)	47.6% (91/191)	46.3% (93/201)	0.5849	0.4657	0.7851
Hypercholesterolemia	41.5% (22/53)	45.7% (86/188)	45.2% (90/199)	0.5840	0.6285	0.9184
B12 deficiency	11.3% (6/53)	9.3% (17/182)	8.0% (16/199)	0.6694	0.4521	0.6521
Thyroid disease	21.2% (11/52)	22.6% (43/190)	22.3% (45/202)	0.8206	0.8617	0.9330
Other Clinical Diagnoses, Psychiatric & Behavioral Features
Posterior cortical atrophy	1.8% (1/55)	3.1% (6/193)	2.5% (5/203)	0.6102	0.7783	0.6959
Primary progressive aphasia	4.0% (2/50)	8.7% (22/252)	10.9% (28/256)	0.2586	0.1313	0.4037
Depression/dysphoria	34.8% (23/66)	36.0% (86/239)	31.2% (72/231)	0.8648	0.5719	0.2693
Delusions	6.1% (4/66)	28.7% (68/237)	24.9% (55/221)	0.0001	0.0009	0.3586
Hallucinations	4.5% (3/66)	24.1% (57/237)	21.7% (48/221)	0.0004	0.0014	0.5531
Apathy	19.7% (13/66)	50.8% (121/238)	62.1% (144/232)	< 0.0001	< 0.0001	0.0141
Disinhibition	4.5% (3/66)	26.7% (64/240)	26.5% (62/234)	0.0001	0.0001	0.9664
Nighttime behavior	12.5% (8/64)	33.8% (80/237)	40.9% (96/235)	0.0009	< 0.0001	0.1109
Anxiety	28.8% (19/66)	46.7% (112/240)	38.0% (90/237)	0.0093	0.1690	0.0547
Agitation/aggression	24.2% (16/66)	39.5% (94/238)	45.2% (108/239)	0.0225	0.0022	0.2084
Irritability/lability	34.8% (23/66)	46.3% (111/240)	37.4% (88/235)	0.0982	0.6691	0.0519
Appetite disturbance	21.5% (14/65)	30.4% (73/240)	43.5% (104/239)	0.1597	0.0013	0.0030
Motor disturbance	7.6% (5/66)	30.5% (73/239)	41.7% (98/235)	0.0002	< 0.0001	0.0114
Medication Use
Total number medications	7.5 ± 0.6	8.0 ± 0.3	5.8 ± 0.3	0.4391	0.0087	< 0.0001
Anticoagulant use	52.1% (37/71)	51.4% (127/247)	29.0% (71/245)	0.9177	0.0003	< 0.0001
Anti-AD medication use	18.3% (13/71)	70.0% (173/247)	46.5% (114/245)	< 0.0001	< 0.0001	< 0.0001
Anti-Parkinson medication use	8.5% (6/71)	6.9% (17/247)	4.9% (12/245)	0.6530	0.2554	0.3500
Any anti-HTN medication use	66.2% (47/71)	58.3% (144/247)	41.2% (101/245)	0.2311	0.0002	0.0003
ACE-I use	15.5% (11/71)	17.0% (42/247)	4.9% (12/245)	0.7633	0.0025	< 0.0001
Angiotensin II inhibitor use	14.1% (10/71)	10.1% (25/247)	6.1% (15/245)	0.3412	0.0286	0.1075
Beta blocker use	43.7% (31/71)	20.6% (51/247)	15.9% (39/245)	< 0.0001	< 0.0001	0.1749
Calcium channel blocker use	15.5% (11/71)	19.8% (49/247)	11.0% (27/245)	0.4095	0.3076	0.0068
Diuretic use	31.0% (22/71)	10.9% (27/247)	9.8% (24/245)	< 0.0001	< 0.0001	0.6796
Lipid-lowering medication use	42.3% (30/71)	40.9% (101/247)	18.8% (46/245)	0.8371	< 0.0001	< 0.0001
NSAID use	43.7% (31/71)	46.6% (115/247)	28.2% (69/245)	0.6660	0.0134	< 0.0001
Diabetes medication use	16.9% (12/71)	12.6% (31/247)	4.9% (12/245)	0.3579	0.0008	0.0025
Anti-depressant use	28.2% (20/71)	54.3% (134/247)	53.1% (130/245)	0.0017	0.0002	0.3444
Antipsychotic use	1.4% (1/71)	15.8% (39/247)	28.6% (70/245)	0.0013	< 0.0001	0.0006
Estrogen therapy	1.4% (1/71)	0.8% (2/247)	0.4% (1/245)	0.6456	0.3494	0.5673

*p-values were adjusted for multiple comparisons with the Benjamini, Krieger, and Yekutieli method, p-values less than 0.0044 are considered significant after false discovery rate (FDR) adjustment. All data in this table represents the final cognitive exam, ≤ 24 months prior to participant death

The MinP group had significantly slower rates of decline for all assessed neuropsychologic tests as compared to the ModP and MaxP groups. There were differences in the rate of decline for DSF, TMT-A, WAIS DS, vegetable naming, and Boston naming tests between the ModP and MaxP participants, but no significant differences in the rate of decline of logical memory (immediate or delayed), DSB, TMT-B, or animal naming tests between these groups by mixed-effects modeling (Fig. 2, Supplemental Figure S2, and Supplemental Table S1). These differences suggest that the MinP (or “resilient”) participants not only have better global cognition, but are also intact across all cognitive domains, while there are less well defined differences between those with the most severe cognitive decline compared to the ModP participants.

Neuropathologic and genetic features

 Although high-level ADNC represents a fairly narrow range of pathology, comprising Braak stage 5–6 (B3), Thal phase 4–5 (A3), and CERAD NP score moderate-frequent (C2-3) [14, 15], there were a number of subtle differences identified in our cohort. The MinP participants had lower Braak stages compared to ModP (p = 0.0002) or MaxP (p < 0.0001), and the MaxP participants had higher Braak stages compared to the ModP participants (p < 0.0001) (Fig. 3A; Table 1). Similarly, the MinP participants had lower Thal phases compared to ModP (p = 0.0022) or MaxP (p < 0.0001), and the MaxP participants had higher Thal phases compared to the ModP participants (p = 0.0008) (Fig. 3B; Table 1). The MinP participants had lower CERAD NP scores compared to ModP (p = 0.0028) or MaxP (p = 0.0002), but no significant difference was found between the ModP and MaxP participants (Fig. 3C; Table 1). No significant difference was found in frequency of CAA (Fig. 3D). While the difference in p-tau pathology (as extrapolated from Braak stage) could indicate that progression into the striate areas (Braak stage VI) is specifically associated with worse outcome and more rapid progression, there is a wide range of frontal, temporal, and parietal lobe p-tau severity which can be found between individuals with Braak stage V (Fig. 4A-C and E-G), and the p-tau burden in individuals with Braak stage VI generally is at the high end of this spectrum in these neocortical regions (Fig. 4D and H). It is therefore possible that the higher Braak stage associated with the MaxP participants (and lower Braak stage associated with the MinP participants) may also represent differences in neocortical and hippocampal p-tau density rather than simply topographical distribution of p-tau.

Fig. 3.

Alzheimer disease pathology in participants with minimal progression (MinP), moderate progression (ModP), and rapid/maximal progression (MaxP), including A Braak stage, B Thal phase, C CERAD NP plaque score, D cerebral amyloid angiopathy (CAA) severity, E frequency of APOE ε2 alleles, and (F) frequency of APOE ε4 alleles. P-values: ** <0.01, *** <0.001, **** <0.0001

Fig. 4.

Representative paired low- and high-power p-tau (AT8)-stained sections of frontal neocortex demonstrating a spectrum of A, E low p-tau burden, B, F moderate p-tau burden, and C, G high p-tau burden in individuals with Braak stage V, as well as D, H high p-tau burden in an individual with Braak stage VI. Scale bars = 5 mm for panels A-D and 500 μm for panels E-H

There was significantly lower LBD stage in the MinP participants compared to the ModP or MaxP participants, but no difference in LBD stage frequency between the ModP and MaxP groups (Table 1). There were no significant differences in LATE stage, frequency of hippocampal sclerosis, or frequency of ARTAG. There was no significant difference in the prevalence of grossly identifiable infarcts/lacunar infarcts, microinfarcts, hemorrhage, or severity of arteriolosclerosis. However, there was more severe white matter rarefaction in the MaxP participants. No differences were noted between the three groups in terms of the frequency of other concomitant neurodegenerative pathologies, including progressive supranuclear palsy (PSP), corticobasal degeneration (CBD), Pick disease (PiD), frontotemporal lobar dementia with TDP-43 (FTLD-TDP), amyotrophic lateral sclerosis (ALS)/motor neuron disease (MND), chronic traumatic encephalopathy (CTE), or history of traumatic brain injury.

APOE genotyping was available for 550 participants (93.9%) in the overall cohort; 11.3% of which (62/550) had ≥ 1 APOE ε2 allele and 58.9% (324/550) had ≥ 1 APOE ε4 allele. Only 2 individuals had an ε2/ε2 genotype, while 12.4% (68/550) had an ε4/ε4 genotype. The MinP group was enriched for APOE ε2 alleles (Fig. 3E; Table 1). The MinP group also had significantly fewer individuals with at least one APOE ε4 allele compared to the ModP group (p = 0.0039) and the MaxP group (p < 0.0001) (Fig. 3F; Table 1).

Clinical features and medication use

 There were minimal differences between these groups in terms of clinical conditions and general medical conditions, as measured at the time of each participant’s final cognitive exam within the final 24 months of life (Table 2). In terms of psychiatric and behavioral features, participants with minimal cognitive progression were less likely to have psychosis (delusions and/or hallucinations), apathy, disinhibition, nighttime behavior disturbances, and disturbances in motor function compared to those with moderate or maximal cognitive progression. Those with maximal cognitive progression were more likely to have disturbances in appetite compared to the MinP or ModP participants.

In terms of medication, MaxP participants were taking significantly fewer medications overall compared to ModP participants in their final 24 months of life (Table 2), although no significant differences were noted between any groups in terms of the total number of medications at first cognitive exam or at the time of conversion to cognitive impairment. Significantly fewer MaxP participants were taking anti-hypertensive medications, anticoagulants, ACE inhibitors, diabetes medications, NSAIDs, and lipid-lowering medications compared with ModP or MinP participants in the final 24 months of life. More participants with MinP were taking beta-blockers and diuretics and fewer were taking anti-depressants compared to the ModP and MaxP participants. These findings suggest that some of these medications may be protective, or possibly that systemic diseases in the MaxP participants were not being managed as well as those in the MinP and ModP groups, as cognitive impairment may impair adequate treatment in these patients. Significantly fewer participants in the MinP group were taking anti-psychotics compared to the ModP and MaxP participants.

Multivariable logistic regression analysis

 We next performed multivariable logistic regression analysis to determine which factors were independently associated with resilience/minimal progression (compared to combined ModP and MaxP groups) and which were independently associated with more rapid/maximal progression (compared to combined MinP and ModP) (Fig. 5). The presence of at least one APOE ε4 allele, higher Braak stage, higher Thal phase, more severe WMR, and unmedicated hyperlipidemia were all independently associated with rapid progression, while unmedicated depression was inversely correlated. Conversely, Thal phase, CERAD NP score, the presence of at least one APOE ε4 allele, and unmedicated psychosis were inversely correlated with resilience/minimal progression, while unmedicated depression was positively correlated.

Fig. 5.

Forest plots of multivariable logistic regression analysis for A predictive factors (demographic, genetic, neuropathological, systemic medical management) associated with the outcome measure minimal progression (longitudinal resilience) compared to combined moderate and maximal progression participants and B predictive factors associated with the outcome measure rapid/maximal progression compared to combined minimal and moderate progression participants. Time between death and final cognitive evaluation was used as a covariate. P-values: * <0.05, ** <0.01

Finally, to determine if undermedication for certain conditions preceded cognitive impairment or perhaps were the result of the difficulty in managing patients with more severe dementia, we evaluated the first cognitive exam in which an individual was cognitively impaired (3.1 ± 0.2 years prior to death for ModP participants and 5.2 ± 0.2 years prior to death for MaxP participants). Since cognitive impairment was by definition not reached in the MinP cohort, a commensurate cognitive exam was used for these participants (4.5 ± 0.1 years prior to patient death, defined as a MinP individual’s cognitive exam closest to the mean between ModP and MaxP groups). While there were no differences in the frequency of unmedicated hypertension, diabetes, depression, or atrial fibrillation, MinP participants had significantly lower rates of unmedicated psychosis. MaxP participants had higher rates of unmedicated hyperlipidemia, although this did not survive FDR correction (Supplemental Table S2).

Discussion

ADNC is the most common pathology underlying dementia and cognitive impairment in the aging population. While there are well-established symptoms associated with this disorder, there remains considerable variation between individuals in terms of severity and progression, and there may be a disconnect between the severity of the pathology and the severity of impairment. Much of this variation may be explained by the presence/absence and severity of comorbid neuropathologies, including LATE, LBD, and CVD, among others [25–30]. These concurrent neurodegenerative processes do not account for all of the variation between individuals, and a number of other factors, including genetics, systemic disease states, and more granular neuropathologic features such as synapse loss, quantitative regional and subregional pathology, asymmetry, and alternative progression of disease, may influence the degree and rate of cognitive decline, and may in part explain some aspects of cognitive reserve and resilience [20–22, 32, 36, 50–54].

In this study, we utilized the NACC dataset to isolate 586 individuals with high-level ADNC, in an effort to maintain a reasonable level of consistency in ADNC severity within the confines of the current National Institute on Aging-Alzheimer’s Association (NIA-AA) criteria [14, 15], to investigate factors which may influence cognition in this pathologically defined group. Segregating these participants by their longitudinal trajectories into MinP, ModP, and MaxP groups, we demonstrated significant differences in both the overall magnitude and rate of global cognitive decline between all groups. There were consistent differences between the MinP (resilient) participants and the other groups in more specific neuropsychologic tests, while the MaxP (rapid/severe progression) had fewer differences with regard to specific neurocognitive domains (Fig. 2, Supplemental Fig. 2, and Supplemental Table 1). Despite the narrow range of ADNC corresponding to high-level ADNC, there were still numerous differences in variables related to AD between these groups. Higher Braak stage, higher Thal phase, and more frequent APOE ε4 alleles were associated with worse cognitive progression while lower Braak stage, lower Thal phase, lower CERAD NP score, and less frequent APOE ε4 alleles were associated with better cognitive outcome in univariate models (Fig. 3; Table 1). Notably, there were relatively frequent comorbid neurodegenerative pathologies in all three groups with trends toward more severe comorbid neuropathology with more severe cognitive outcomes (Table 1), but only WMR had an independent association with the most severe cases in multivariable models (Fig. 5). It is notable that recent work has identified WMR as both a feature of CVD and of neurodegenerative disease, including ADNC, where it may reflect disease severity [55]. These findings indicate that even though these cases are grouped together as “high-level ADNC”, there is still a great deal of nuance within this pathologic category which may significantly influence cognition. It is important to note that while we directly assessed Braak stage and Thal phase, these staging systems only take into account the topographical distribution of these pathologies and not the overall or regional density of pathology. It is clear that while Braak stage V indicates neocortical p-tau-positive neurofibrillary tangles that have not yet progressed into the striate cortices [5, 6], there can be a wide spectrum of pathologic severity in hippocampal structures and neocortical areas which may better correlate to cognitive outcomes (Fig. 4). Furthermore, p-tau pathology in the frontal, temporal, and parietal cortices is generally on the more severe end of the spectrum in Braak stage VI (in addition to striate cortex involvement), so the observed effect of Braak stage may be mediated in part by increasing density of p-tau pathology throughout the brain. Given the available NACC data, it is not possible to evaluate this further in the current cohort, however this is an active area of further research.

Another notable finding in these results is the variation in rates of medication use for underlying medical conditions in the final 24 months of life, despite statistically similar rates of those underlying medical conditions (Table 2). As a group, the MaxP participants are taking significantly fewer medications overall than either the MinP or ModP participants and fewer MaxP individuals are taking anticoagulants, ACE inhibitors, diabetes medications, NSAIDs, and lipid-lowering medications in particular. In addition, the MinP participants have lower rates of being unmedicated or undermedicated for various underlying medical conditions. These findings suggest that in addition to the genetics and underlying ADNC, general clinical management of underlying medical conditions may have a significant effect on the progression of cognitive impairment in these individuals. In particular, while there were no significant differences between groups in terms of frequency of hyperlipidemia, participants in the MaxP cohort were significantly more likely to be undertreated with lipid-lowering medications in the final 24 months of life (Fig. 5; Table 2), consistent with previous meta-analyses indicating that taking adequate lipid-lowering medications might lower risk of AD and other forms of dementia [56, 57], although this finding was only present at the trend level at the time of conversion to cognitive impairment (Supplemental Table S2). Depression did not fit this trend, however. The MinP cohort had significantly lower anti-depressant use despite similar rates of depression and unmedicated depression was positively correlated with minimal progression and negatively correlated with rapid progression (Fig. 5), although this was not found earlier in life (Supplemental Table S2). This apparent contradiction may be explained in part by the recent association between anti-depressant use with faster cognitive decline and higher risk for severe dementia [58]. Use of anti-psychotic and anti-AD medications also do not fit this pattern, as significantly fewer participants in the MinP group were taking these compared to the ModP and MaxP participants. This is likely because the MinP groups had fewer individuals with delusions/hallucinations, and less severe clinical AD symptoms, including amnestic dementia, and so were less likely to be placed on anti-AD therapies. Interestingly, ADNC participants with a history of psychosis were previously shown to have worse longitudinal cognitive deficits, more advanced AD neuropathology, and more severe right temporal lobe regional atrophy [49], and other studies have shown a correlation between psychosis/anti-psychotic medication use and dementia outcomes, including disease severity and mortality in dementia patients [59, 60]. The MaxP group had fewer individuals on anti-AD therapies compared to the ModP group, which may reflect their more severe symptomology and progression.

There are a number of strengths and limitations in this study. The overall cohort is relatively large with relatively tight control on the main pathologic measure (ADNC), allowing for investigation into the subtleties of these and other neuropathologic measures between participants. In addition, we were able to construct longitudinal models for cognitive and neuropsychologic variables from the NACC dataset, which are more able to detect subtle cognitive changes and differences between groups [61, 62]. However, participants in the NACC dataset may not be representative of the general population [63, 64]. There may be population-specific recruitment biases, and the NACC dataset is notably enriched for white individuals and individuals with at least one APOE ε4 allele, more severe dementia, more significant neuropathologic findings (including more frequent rare diseases), and high levels of education compared to the general population. Individuals with the most severe cognitive decline may also be somewhat underrepresented in these data, since these individuals are more likely to be lost to follow-up, however we chose to exclude patients without cognitive data in the final 24 months of life to ensure that the neuropathologic findings were strongly representative of the final cognitive states. Other inclusion criteria may also introduce a source of bias. For example, the requirement that participants be ≥ 65 years of age at death excludes a significant population that may have different genetic and environmental risk factors, but conversely serves to maintain a more robust MinP cohort. Future studies should include an exploration into younger patients and those with specific genetic findings (including APOE ε4 allele carriers).

Likewise, our chosen definitions for MinP, ModP, and MaxP groups may influence the results, and certain outcome measures may vary in a study where global CDR of 2–3 was used as a threshold for MaxP, for example. In addition, we did not have access to histologic and immunohistochemical sections for these cases and thus had to rely on less granular measures of pathologic assessment (e.g., Braak stage was available but relative regional p-tau density was not). The NACC medication data also lacks granularity. The dose and specific medication included in each medication category could not be fully assessed and included in statistical models (e.g., numerous lipid-lowering medications were grouped into a single category), and because of the way the medications and systemic diseases are recorded over time, it is difficult to definitively separate cause from effect with these data. Individuals with more severe symptoms may be more likely to receive some types of pharmacologic treatment, but conversely, patients in the end stages of dementia may be less likely to have their systemic illnesses comprehensively managed, which limit our interpretation of these data. A clearer understanding of these mechanisms will require additional prospective studies with longitudinal biomarker imaging and more precise and comprehensive collection of data related to systemic disease states and medication exposure.

Conclusions

Despite these limitations, our findings address a gap in the understanding of the relationship between ADNC and cognitive symptoms, particularly what additional factors are involved in conversion to cognitive impairment/dementia and what factors contribute to disease progression. This study also investigates the concept of resilience in terms of longitudinal rate of progression of cognitive symptoms, which to our knowledge has not been fully investigated in autopsy-based cohorts [65, 66]. These data demonstrate that while cognitive impairment generally correlates with Braak stage and CERAD NP score [16], genetic factors and more subtle differences in the severity of p-tau and Aβ pathology may significantly impact the rate and magnitude of cognitive impairment, even within the narrow range of “high-level” ADNC. This suggests that more granular assessment of the underlying pathology is needed to fully understand what is predictive of cognitive decline, a crucial step in the rational design and evaluation of biomarkers and therapeutics, which may provide insight into stratification for clinical trials. In addition, the finding that certain medications and undertreatment of certain chronic systemic illnesses may be associated with cognitive decline suggests a modifiable risk factor which could potentially alleviate or forestall cognitive impairment in a subset of patients, although more research is needed to further refine this point. These data present a significant step forward in our understanding of the clinical phenotype of individuals with the most severe AD pathology, and offers insight into potential therapeutic avenues.

Authors’ contributions

Study concept or design: TER, JMW; Major role in acquisition of data: TER, LC, SH, CMD, JS, KC, CCS, LSKM, LYCC; Analysis or interpretation of data, including statistical analysis: TER, SK, FCA, SKR, GAM, KF, JFC, EVD, SEE, MMG, TGO, JMW; Drafting/revision of the manuscript for content, including medical writing for content: TER, EVD, CLW, SEE, MMG, JMW; All authors have reviewed and approved of the final draft.

Funding 

T.E.R. and J.M.W. are supported in part by National Institute on Aging (NIA) R21 AG078505 and Texas Alzheimer’s Research and Care Consortium (TARCC) grants 957581 and 957607. E.V.D. is supported in part by TARCC grant 957607. The funders had no role in study design, data collection, analysis, decision to publish, or preparation of the manuscript.

Data availability

The data presented in this manuscript were derived from the National Alzheimer’s Coordinating Center (NACC) dataset, and are available upon request from [https://naccdata.org/].

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki. Written informed consent for each participant was obtained at the ARDC of origin, and study protocols were approved by the institutional review boards (IRBs) at each institution.

Consent for publication

Not applicable.

Competing interests

T.E.R. has been a consultant for Servier Pharmaceuticals. T.G.O. has been a consultant for Sonae and Guidepoint, has received fees as a speaker from Eisai and conference fees covered from Roche and Lilly. The authors declare that these disclosures are unrelated to the present work, and that they have no additional competing interests, conflicts of interest, or other relevant disclosures.