Concordance of Clinical Alzheimer Diagnosis and Neuropathological Features at Autopsy

Abstract

It remains unclear what clinical features inform the accuracy of a clinical diagnosis of Alzheimer disease (AD). Data were obtained from the National Alzheimer’s Coordinating Center to compare clinical and neuropathologic features among participants who did or did not have Alzheimer disease neuropathologic changes (ADNC) at autopsy. Participants (1854) had a clinical Alzheimer dementia diagnosis and ADNC at autopsy (Confirmed-AD), 204 participants had an AD diagnosis and no ADNC (AD-Mimics), and 253 participants had no AD diagnosis and ADNC (Unidentified-AD). Compared to Confirmed-AD participants, AD-Mimics had less severe cognitive impairment, while Unidentified-AD participants displayed more parkinsonian signs, depression, and behavioral problems. This study highlights the importance of developing a complete panel of biomarkers as a tool to inform clinical diagnoses, as clinical phenotypes that are typically associated with diseases other than AD may result in inaccurate diagnoses.

INTRODUCTION

The accuracy of the clinical diagnosis of dementia and a clinician’s ability to distinguish between dementia subtypes is important for clinical care, minimizing caregiver burden, clinical trial participation, and developing effective treatments. Thus, it is critical to strive for the most accurate assessment of a patient’s underlying etiology of cognitive decline. However, since postmortem neuropathological examination remains the current gold-standard for identifying the cause of dementia, the accuracy of clinical diagnoses remains in question while a patient is alive (1). The original National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) criteria for Alzheimer disease (AD) diagnosis were published in 1984 (2). These were updated to the National Institute on Aging—Alzheimer Association (NIA-AA) criteria in 2011 and 2018 to elaborate on the preclinical and mild cognitive impairment phases of AD and the importance of incorporating biomarkers into AD diagnosis (3–5). Several studies have examined the accuracy of clinical diagnosis of AD dementia. Beach et al (6) found that sensitivity ranged from 70.9% to 87.3% and specificity from 44.3% to 70.8%. In a clinical imaging/pathology series of 57 individuals clinically diagnosed with AD, 13 (23%) had no or sparse beta amyloid (Aβ) plaques at autopsy. The study concluded that diagnostic accuracy of clinical AD is variable, even among experts (7). Understanding factors associated with an inaccurate AD diagnosis could lead to more accurate diagnoses of dementia etiology during life.

Previous studies have compared clinical and neuropathological characteristics of patients who received accurate and inaccurate AD diagnoses. Using data from the National Alzheimer’s Coordinating Center (NACC), Shim et al (1) compared 533 patients who were diagnosed with AD, referring to the 119 patients who did not meet neuropathological criteria for AD as AD-Mimics. AD-Mimics were more likely to have a history of pacemaker insertion, congestive heart failure, hypertension, and resting tremor than participants who had autopsy-confirmed AD. Compared with AD-Mimics, participants with autopsy-confirmed AD had a higher prevalence of delusions, agitation or aggression, depression or dysphoria, and motor disturbances. Autopsy-confirmed AD participants also showed more impairment on neuropsychological tests and clinically evaluated scales. The authors examined neuropathological diagnoses among the AD-Mimics rather than neuropathological features, which prevented them from addressing the potential effects of mixed pathologies present in the AD-Mimics participants.

Another study using NACC data compared accurately and inaccurately diagnosed probable AD cases where a participant did or did not meet the defined neuropathological threshold for AD at autopsy (i.e. moderate to frequent Consortium to Establish a Registry for Alzheimer’s disease (CERAD) score or Braak stages III–VI). The authors found that misdiagnosed participants were less likely to be depressed. In addition, they found that 18.8% of misdiagnosed participants were treated with potentially inappropriate medications, highlighting the importance of diagnostic accuracy and the need for additional research to establish biomarkers as a diagnostic tool to avoid misdiagnosis and inappropriate treatment in dementia patients (8). These studies addressed the clinical characteristics associated with AD-Mimics, but the literature on underlying neuropathological features that might be misdiagnosed clinically as AD is sparse. Another study data aimed to address this, using NACC data to compare hippocampal sclerosis (HS) pathology with AD pathology, finding less cognitive impairment in persons with HS versus AD pathology. Despite this, over 80% of participants with “pure” HS pathology (HS pathology and no AD pathology) were diagnosed with possible or probable AD prior to death, indicating that HS pathology strongly mimics AD pathology clinically (9).

This study expands on previous studies by exploring clinical and neuropathological characteristics in participants with and without Alzheimer disease neuropathologic change (ADNC). Particularly, we aimed to examine a wide range of neuropathological features rather than focusing on a particular neuropathological diagnosis, which allows for exploring possible mixed pathologies. Our primary goals were to identify antemortem characteristics that lead to clinical AD diagnosis in participants who have low or do not have AD neuropathologic features at autopsy, and to explain why other participants seem to slip under the radar, never receiving an AD diagnosis during life but demonstrating AD neuropathologic features at autopsy. We explored this additional group because AD pathological changes begin in the brain well before symptoms present clinically; however, once cognitive impairment begins it is important to understand what clinical presentations may be influencing clinicians to miss an AD diagnosis as this is uncommon.

MATERIALS AND METHODS

Data Acquisition and Participants

We used data from NACC. NACC maintains the Uniform Data Set (UDS), which includes standardized clinical data reported by past and present NIA-funded Alzheimer’s Disease Centers (ADCs) throughout the United States. Each ADC recruits participants according to its own protocols; participants may come from clinician referral, self-referral by patients or family members, and active community recruitment. Data were collected by clinicians or trained interviewers at each ADC. Assessments were conducted approximately yearly using a standard protocol (10–12). Informed consent was obtained from all participants at the individual ADCs. Neuropathologic data is available for those participants who consented to autopsy and later died.

This study consisted of a cross-sectional sample of NACC UDS participants with data collected from UDS visits conducted between September 2005 and August 2018. The study was limited to participants who died within 2 years from their last UDS visit.

Neuropathological Features

Neuropathological evaluations were performed at individual ADCs according to their own protocols. NACC maintains the Neuropathology Data Set, which includes standardized data on these evaluations. In this study, we examined 3 groups—Confirmed-AD, AD-Mimics, and Unidentified-AD. Participants in the Confirmed-AD group (n = 1854) were diagnosed with primary or contributing AD-dementia at their last UDS visit before death and met our criteria for ADNC: moderate to frequent neuritic plaques and Braak stages III through VI, which loosely aligns with NIA-AA criteria for intermediate or high ADNC (13). Participants in the AD-Mimics group (n = 204) were also diagnosed with primary or contributing AD-dementia at their last UDS visit before death, but at autopsy did not exhibit neuritic plaque burden (CERAD score 0) and therefore did not exhibit ADNC. Participants in the final study group, Unidentified-AD (n = 253), had a dementia diagnosis at their last UDS visit before death but did not receive a primary or contributing clinical diagnosis of AD. Nonetheless, participants in this group did meet the criteria for ADNC. In addition to ADNC, other neuropathological features were assessed in the Neuropathology Data Set and were explored in all 3 groups. Lewy body disease (LBD) pathology was characterized according to established guidelines (14). Frontotemporal lobar degeneration (FTLD) subtypes were separated into tau-positive subtypes (FTLD-tau) and nontauopathy subtypes (FTLD-transactive response DNA-binding protein (TDP)-43/other). FTLD-tau pathology included Pick disease, corticobasal degeneration, progressive supranuclear palsy, and other tauopathies. FTLD-TDP-43/other pathology included TDP-43 pathology, ubiquitin positive/tau-negative inclusions, and not specified but not a tauopathy. Gross infarcts, microinfarcts, hemorrhages, cortical laminar sclerosis, and HS were also assessed.

Diagnostic Criteria and Covariate Definitions

Clinical diagnoses including cognitive status were made by a clinician or a consensus team. A presumptive etiologic diagnosis of AD was made following either the NINCDS-ADRDA criteria in UDS versions 1 and 2, or the NIA-AA criteria in UDS version 3 (2, 3). In this study, depression was defined as having either self-reported active depression within 2 years from the last UDS visit before death, or a clinical diagnosis of depression at the last UDS visit before death. Parkinsonian signs were defined as having changes in motor function suggestive of Parkinsonism at the last UDS visit before death. Cardiovascular/cerebrovascular comorbidities were defined as having self or coparticipant report of any of the following cardiovascular/cerebrovascular conditions as being “active” (i.e. within the last year or still requires active management) at the last UDS visit before death: heart attack/cardiac arrest, congestive heart failure, diabetes, stroke, or transient ischemic attack. Behavioral problems were defined as having any of the following clinically indicated at the last UDS visit before death: apathy, depressed mood, hallucinations (visual or auditory), delusions, disinhibition, irritability, agitation, personality changes, rapid eye movement (REM) sleep behavior disorder, anxiety, or any other meaningful change in behavior. Language decline was defined as having clinically meaningfully impaired language abilities such as hesitant speech, have trouble finding words, or the use inappropriate words without self-correction at the last UDS visit before death.

Statistical Analyses

We performed statistical analyses of neuropathological characteristics and clinical characteristics associated with the 3 study groups.

Clinical characteristics investigated included length of follow-up, age at death, years of education, sex, CDR Dementia Staging Instrument sum of boxes (CDR-SB) from the last UDS visit before death, presence of the APOEε4 allele, depression, parkinsonian signs, cardiovascular/cerebrovascular comorbidities, behavioral problems, and language decline. Additional comparisons examined the primary and contributing diagnoses from the last UDS visit among the participants in the 3 study groups. Two chi-squared tests were conducted for each clinical characteristic or diagnosis—1 test for the equality of the proportions in the Confirmed-AD group and the AD-Mimics group and another test for the equality of the proportions in the Confirmed-AD group and the Unidentified-AD group. For comparisons of continuous variables, t-tests were used. To correct for multiple comparisons, we additionally adjusted p values using the Bonferroni method. We then performed 2 separate multivariable logistic regressions using all of the clinical characteristics as covariates, as all of these clinical characteristics were significantly different in at least 1 bivariate comparison. The first logistic regression estimated the odds ratio (OR) associated with inclusion in the AD-Mimics group relative to the Confirmed-AD group for each of the above clinical variables. The second model estimated the OR associated with inclusion in the Unidentified-AD group relative to the Confirmed-AD group.

Neuropathological characteristics investigated included the proportion of participants in each group who displayed neurofibrillary degeneration (defined by Braak stage), neuritic plaques (CERAD score), LBD pathology, FTLD-tau, FTLD-TDP-43/other pathology, arterial infarcts, hemorrhages, cortical laminar sclerosis, and HS. As with the clinical characteristics, we compared neuropathological characteristics using chi-square for the Confirmed-AD group versus the AD-Mimics and for the Confirmed-AD group versus the Unidentified-AD group, with additional Bonferroni correction for multiple comparisons. We then performed 2 additional multivariable logistic regressions using the neuropathological characteristics that were significantly different in at least 1 bivariate comparison. Since all 3 groups in the study were defined by their Braak stage and frequency of neuritic plaques, we did not include these in the model comparing the neuropathological characteristics between the Confirmed-AD group and AD-Mimics.

RESULTS

Comparison of Clinical Characteristics

Compared to the Confirmed-AD group, AD-Mimics had higher mean ages at time of death, were less likely to possess the APOEε4 allele and had lower CDR-SB scores at their last UDS visit. Furthermore, AD-Mimics were more likely to have a cardiovascular/cerebrovascular comorbidity present and less likely to have language decline (Table 1).

TABLE 1.

Demographic and Clinical Characteristics of the 3 Study Groups at Last Visit Before Death

Characteristics	Confirmed-AD (n = 1854)	AD-Mimics (n = 204)	Unidentified-AD (n = 253)	p Value (Confirmed-AD vs AD-Mimics)	p Value (Confirmed-AD vs Unidentified-AD)
Length of follow-up (years), mean (SD)	3.5 (2.8)	3.2 (2.8)	2.2 (2.2)	0.102	<0.001
Age at death, mean (SD)	80.5 (10.7)	83.9 (11.0)	75.3 (9.1)	<0.001	<0.001
Education (years), mean (SD)	15.2 (3.3)	15.2 (3.4)	15.6 (3.1)	0.950	0.085
≥1 APOEε4 allele, n (%)	965 (52.1)	28 (13.7)	109 (43.1)	<0.001	0.007
Male, n (%)	1006 (54.3)	109 (53.4)	163 (64.4)	0.821	0.002
Non-white, n (%)	141 (7.6)	19 (9.3)	15 (5.9)	0.387	0.340
CDR sum of boxes at last visit, mean (SD)	14.0 (4.7)	10.3 (5.1)	12.8 (4.9)	<0.001	<0.001
Depression, n (%)	752 (40.6)	96 (47.1)	146 (57.7)	0.309	0.001
Parkinsonian signs, n (%)	279 (15.1)	29 (14.2)	97 (38.3)	0.934	<0.001
Cardiovascular/cerebrovascular comorbidity* present, n (%)	284 (15.3)	55 (27.0)	51 (20.2)	<0.001	0.581
Behavioral problems† present, n (%)	1549 (83.6)	159 (77.9)	235 (92.9)	0.113	0.002
Language decline reported, n (%)	1562 (84.3)	124 (60.8)	220 (87.0)	<0.001	0.263

^*Cardiovascular/cerebrovascular comorbidities include at least one of the following: heart attack/cardiac arrest, congestive heart failure, diabetes, stroke, or transient ischemic attack.

^†Behavioral problems include at least one of the following: apathy, depressed mood, hallucinations (visual or auditory), delusions, disinhibition, irritability, agitation, personality changes, REM sleep behavior disorder, anxiety, or any other meaningful change in behavior.

Table 1 also compares the Confirmed-AD group and the Unidentified-AD group. Compared to the Confirmed-AD group, the Unidentified-AD group had a higher proportion of males, died at a younger age, had slightly lower CDR-SB scores, and were less likely to possess the APOEε4 allele. Only the comparison for APOEε4 allele became insignificant after Bonferroni correction. Participants in the Unidentified-AD group also had significantly higher proportions of depression, behavioral problems, and parkinsonian signs compared to the Confirmed-AD group.

We compared the clinical diagnoses from the most recent visit before death among the 3 groups (Table 2). Participants in both the AD-Mimics and Confirmed-AD groups were required to have either a primary or contributing AD diagnosis at their last visit prior to death. Comparing those 2 groups, a smaller proportion of AD-Mimics received a primary AD diagnosis and a larger proportion received a contributing diagnosis of AD compared to those with Confirmed-AD. Participants in the AD-Mimics group were more likely to be diagnosed with primary FTLD and primary vascular dementia or stroke than participants in the Confirmed-AD group. Among participants in the Unidentified-AD group, the most common primary diagnoses were frontotemporal lobar degeneration (41.1%), LBD (43.9%), and vascular dementia or stroke (6.7%).

TABLE 2.

Clinical Diagnoses of the 3 Study Groups at Last Visit Before Death

Clinical Diagnoses	Confirmed-AD (n = 1854)	AD-Mimics (n = 204)	Unidentified-AD (n = 253)	p Value (Confirmed-AD vs AD-Mimics)	p Value (Confirmed-AD vs Unidentified-AD)
Primary clinical diagnosis at last visit, n (%)
Alzheimer disease dementia	1737 (93.7)	161 (78.9)	0 (0.0)	<0.001	NA
Lewy body dementia	44 (2.4)	8 (3.9)	104 (41.1)	0.181	<0.001
Frontotemporal dementia	42 (2.3)	15 (7.4)	111 (43.9)	<0.001	<0.001
Vascular dementia/stroke	20 (1.1)	15 (7.4)	17 (6.7)	<0.001	<0.001
Contributing diagnoses (not mutually exclusive) at last visit, n (%)
Alzheimer disease dementia	117 (6.3)	43 (21.1)	0 (0.0)	<0.001	NA
Lewy body dementia	114 (6.2)	10 (4.9)	2 (0.8)	0.478	<0.001
Frontotemporal dementia	42 (2.3)	7 (3.4)	23 (9.1)	0.300	<0.001
Vascular dementia/stroke	166 (9.0)	31 (15.2)	12 (4.7)	0.004	0.024

Table 3 displays the results of 2 logistic regression models examining the association of clinical features with either AD-Mimics (Model 1) or Unidentified-AD (Model 2) compared to Confirmed-AD. In Model 1, participants with at least 1 APOEε4 allele had an 84% reduction (OR = 0.16) in the odds of inclusion in the AD-Mimics group. That is, presence of the ε4 allele was associated with increased likelihood that a given participant with an AD diagnosis exhibited ADNC at autopsy. Increasing CDR-SB score was associated with decreased odds of inclusion in the AD-Mimics group. In Model 2, possession of at least 1 APOEε4 allele was associated with decreased odds of being in the Unidentified-AD group. Increasing age at death and increasing CDR-SB were also associated with decreased odds of being in the Unidentified-AD group. Alternatively, many factors were found to be associated with increased odds of an AD case going unidentified—parkinsonian signs, behavioral problems, and declining language abilities.

TABLE 3.

Adjusted Odds Ratios (ORs) for Clinical Features Associated With AD Mimicry and Unidentified-AD Relative to the Confirmed-AD Reference Group

Variable	Model 1: AD-Mimics (Ref: Confirmed-AD)			Model 2: Unidentified-AD (Ref: Confirmed-AD)
	OR	95% CL	p Value	OR	95% CL	p Value
Age at death	1.01	(0.99–1.03)	0.397	0.95	(0.93–0.97)	<0.001
Male	0.90	(0.64–1.26)	0.528	1.01	(0.69–1.48)	0.949
Education (years)	0.98	(0.92–1.06)	0.653	1.01	(0.96–1.06)	0.775
Length of follow-up (years)	1.00	(0.92–1.08)	0.931	0.94	(0.85–1.04)	0.248
Non-white race	1.36	(0.68–2.69)	0.384	0.71	(0.33–1.55)	0.392
CDR sum of boxes at last visit	0.89	(0.84–0.93)	<0.001	0.95	(0.91–0.98)	0.005
≥1 APOEε4 allele	0.16	(0.10–0.26)	<0.001	0.60	(0.43–0.82)	0.002
Depression	1.28	(0.92–1.80)	0.148	1.15	(0.72–1.83)	0.548
Parkinsonian signs	0.83	(0.49–1.40)	0.481	5.38	(3.24–8.94)	<0.001
Cardiovascular/cerebrovascular comorbidity* present	1.34	(0.82–2.19)	0.243	1.48	(0.99–2.22)	0.055
Behavioral problems† present	1.12	(0.57–2.19)	0.742	2.40	(1.16–4.99)	0.019
Language decline reported	0.66	(0.34–1.30)	0.233	1.95	(1.04–3.64)	0.037

^*Cardiovascular/cerebrovascular comorbidities include at least one of the following: heart attack/cardiac arrest, congestive heart failure, diabetes, stroke, or transient ischemic attack.

^†Behavioral problems include at least one of the following: apathy, depressed mood, hallucinations (visual or auditory), delusions, disinhibition, irritability, agitation, personality changes, REM sleep behavior disorder, anxiety, or any other meaningful change in behavior.

Comparison of Neuropathological Characteristics

Table 4 displays the differences in neuropathologic features among the 3 groups. Compared to the Confirmed-AD group, participants in the AD-Mimics group were more likely to have Braak stages III and IV. However, participants in the Confirmed-AD group were more likely than AD-Mimics to have Braak stages V and VI. Participants in the AD-Mimics group were more likely to have brainstem-predominant Lewy bodies. Conversely, the Confirmed-AD group was more likely than the AD-Mimics group to have LBD pathology; specifically, they had higher proportions of limbic or amygdala predominant, neocortical, and other or unspecified Lewy bodies. The AD-Mimics group was more likely than the Confirmed-AD group to exhibit FTLD-tau pathology, as well as FTLD with TDP-43 pathology. The AD-Mimics group had higher proportion of gross infarcts, microinfarcts, and HS than the Confirmed-AD group. After Bonferroni correction, the comparisons between groups for brainstem-predominant Lewy bodies, limbic or amygdala predominant Lewy bodies, and other or unspecified Lewy bodies became insignificant.

TABLE 4.

Neuropathological Characteristics of the 3 Study Groups

Characteristics	Confirmed-AD (n = 1854)	AD-Mimics (n = 204)	Unidentified-AD (n = 253)	p Value (Confirmed-AD vs AD-Mimics)	p Value (Confirmed-AD vs Unidentified-AD)
Braak stage, n (%)
AD-type neurofibrillary degeneration not present	0 (0.0)	38 (19.3)	0 (0.0)	<0.001	1.00
Stage I	0 (0.0)	33 (16.8)	0 (0.0)	<0.001	1.00
Stage II	0 (0.0)	50 (25.4)	0 (0.0)	<0.001	1.00
Stage III	78 (4.2)	35 (17.8)	34 (13.4)	<0.001	<0.001
Stage IV	209 (11.3)	31 (15.7)	51 (20.2)	0.002	<0.001
Stage V	539 (29.1)	6 (3.1)	52 (20.6)	<0.001	0.005
Stage VI	1028 (55.5)	4 (2.0)	116 (45.9)	<0.001	0.004
Neuritic plaques, n (%)
No neuritic plaques	0 (0.0)	204 (100.0)	0 (0.0)	<0.001	1.00
Sparse neuritic plaques	0 (0.0)	0 (0.0)	0 (0.0)	1.00	1.00
Moderate neuritic plaques	438 (23.6)	0 (0.0)	87 (34.4)	<0.001	<0.001
Frequent neuritic plaques	1416 (76.4)	0 (0.0)	166 (65.6)	<0.001	<0.001
Lewy body pathology, n (%)
None	1145 (61.8)	160 (78.4)	123 (48.6)	<0.001	<0.001
Brainstem predominant	49 (2.6)	13 (6.4)	9 (3.6)	0.003	0.416
Limbic (transitional) or amygdala predominant	281 (15.2)	18 (8.8)	32 (12.7)	0.014	0.274
Neocortical (diffuse)	253 (13.7)	9 (4.4)	82 (32.4)	0.002	<0.001
Other region or unspecified	113 (6.1)	4 (2.0)	7 (2.8)	0.015	0.031
FTLD-tau, n (%)	18 (1.6)	27 (18.6)	16 (7.6)	<0.001	<0.001
FTLD-TDP-43/other, n (%)	67 (5.0)	36 (21.2)	18 (9.1)	<0.001	0.020
Gross infarcts, n (%)	356 (26.3)	58 (39.2)	44 (20.0)	0.001	0.046
Hemorrhages, n (%)	103 (7.9)	13 (9.1)	10 (4.7)	0.626	0.094
Microinfarcts, n (%)	370 (26.5)	57 (35.0)	44 (19.9)	0.023	0.036
Hippocampal sclerosis, n (%)	221 (12.3)	44 (22.9)	14 (5.7)	<0.001	0.002
Cortical laminar necrosis, n (%)	14 (1.1)	3 (2.3)	2 (1.0)	0.254	0.846

FTLD, frontotemporal lobar degeneration; TDP, transactive response DNA-binding protein.

Compared to the Confirmed-AD group, participants in the Unidentified-AD group were more likely to have Braak stages III and IV and less likely to have Braak stages V and VI. Since ADNC was defined as having moderate to frequent neuritic plaques, both Confirmed-AD and Unidentified-AD participants displayed neuritic plaques; however, compared to the Confirmed-AD group, participants in the Unidentified-AD group were more likely to have moderate neuritic plaques, and less likely to have frequent neuritic plaques. The Confirmed-AD group was more likely to have no LBD pathology and other or unspecified LBD pathology than the Unidentified-AD group. However, the Unidentified-AD group was more likely to have neocortical Lewy bodies. A larger proportion of participants in the Unidentified-AD group had FTLD-tau pathology and FTLD-TDP-43 pathology. The Unidentified-AD group was also less likely to exhibit gross infarcts, microinfarcts, and HS than participants in the Confirmed-AD group. After Bonferroni correction, the comparisons between groups for Braak stages V through VI, other or unspecified Lewy bodies, FTLD-TDP-43 pathology, gross infarcts, microinfarcts, and HS became insignificant.

In Table 5, we display the results of 2 additional, separate logistic regression models. The results from the Model 3 show the adjusted ORs for neuropathological predictors of the AD-Mimics group. The results from Model 4 show the adjusted ORs for neuropathological features associated with inclusion in the Unidentified-AD group. In Model 3, neuropathological factors found to be associated with increased odds of inclusion in the AD-Mimics group included increasing age at death and the presence of both FTLD-tau and FTLD-non-tau pathologies. LBD pathology was associated with decreased odds of inclusion in the AD-Mimics group. In Model 4, LBD pathology, FTLD-tau, and FTLD-TDP-43/other pathology were both associated with increased odds of inclusion in the Unidentified-AD group.

TABLE 5.

Adjusted Odds Ratios (ORs) for Neuropathological Predictors Associated With AD Mimicry and Unidentified-AD Relative to the Confirmed-AD Reference Group

Variable	Model 3: AD-Mimics (Ref: Confirmed-AD)			Model 4: Unidentified-AD (Ref: Confirmed-AD)
	OR	95% CL	p Value	OR	95% CL	p Value
Age at death	1.05	(1.02–1.08)	<0.001	0.96	(0.94–0.97)	<0.001
Lewy body pathology	0.45	(0.26–0.78)	0.004	1.94	(1.36–2.78)	<0.001
FTLD-tau	17.92	(7.55–42.54)	<0.001	4.28	(1.66–11.01)	0.003
FTLD-TDP-43/other	6.61	(3.19–13.72)	<0.001	2.51	(1.21–5.22)	0.014
Gross infarcts	1.62	(0.99–2.66)	0.054	1.05	(0.65–1.68)	0.855
Microinfarcts	0.92	(0.53–1.60)	0.772	1.04	(0.63–1.70)	0.884
Hippocampal sclerosis	1.59	(0.87–2.89)	0.132	0.48	(0.22–1.05)	0.066

FTLD, frontotemporal lobar degeneration; TDP, transactive response DNA-binding protein.

AD-Mimics were more likely than Confirmed-AD participants to have FTLD pathology despite having fewer behavioral problems and less language decline. We ran a supplemental analysis to independently assess each behavioral problem that was included in the grouped covariate (Supplementary DataTable S1). After Bonferroni correction, we found that AD-Mimics had significantly fewer visual hallucinations and delusions, and less agitation than Confirmed-AD participants, while compared to Confirmed-AD, participants with Unidentified-AD were more likely to have depressed mood, personality change, visual hallucinations, and REM sleep behavior disorder and less likely to have anxiety.

To explore whether there were different levels of mixed pathology present among the 3 groups, we ran a supplemental analysis comparing the frequencies of several pathological combinations (ADNC + one other co-occurring pathology) as well as the frequency of ADNC alone (i.e. LBD, FTLD, HS, and vascular pathologies were not present in addition to ADNC; Supplementary DataTable S2). The results show that in our sample, AD-Mimics had significantly less cumulative mixed pathology, meaning they had less of any 2, 3, or 4 pathologies present when compared to the Confirmed-AD group. However, the Unidentified-AD group had significantly more ADNC + LBD, ADNC + FTLD, and more of any 2 pathologies present compared to the Confirmed-AD group. This supplemental analysis supports the notion that the Unidentified-AD group has higher frequencies of underlying mixed pathology, which may explain their undiagnosed AD and the symptoms they displayed at the last visit before death.

DISCUSSION

We examined the clinical and neuropathological characteristics among 3 groups—Confirmed-AD, AD-Mimics, and Unidentified-AD—to better understand differences that result in clinical misdiagnoses. We found that participants in the AD-Mimics group lived longer than those with Confirmed-AD and have lower CDR-SB scores. These results corroborate previous studies (1, 15) that AD-Mimics are less cognitively impaired than people with Confirmed-AD. A greater degree of cognitive impairment was associated with decreased odds of inclusion in both the AD-Mimics group and the Unidentified-AD group, indicating that more significant cognitive impairment increases the likelihood that an AD diagnosis is correct. In addition, participants in the Confirmed-AD group displayed more severe ADNC (i.e. Braak stages V and VI and frequent neuritic plaques) at autopsy than participants in the AD-Mimics and Unidentified-AD groups.

Previous studies (1, 15) have shown that LBD and cerebrovascular injury are the most common AD-dementia mimics. In this study, we found that among the AD-Mimics, participants tended to display FTLD-tau, FTLD-TDP-43/other, HS, and cerebrovascular pathologies at higher proportions than participants in the Confirmed-AD group. It is possible that this result differs from previous studies because we included participants with either a primary or contributing clinical diagnosis of AD, while previous studies likely focused on primary clinical diagnoses. To determine whether our results were due to the inclusion of participants with contributing AD, we ran a sensitivity analysis restricted to participants with a primary diagnosis of AD in the AD-Mimics group (n = 161; 79% of the original AD-Mimics group; Supplementary DataTables S3–S7). The results of this sensitivity analysis showed no difference in any association or frequency compared to the main analysis, aside from that of HS in Supplementary DataTable S7 (Model 7), which became significant (OR: 2.04, 95% CI: 1.09–3.81; p value: 0.026).

The Unidentified-AD group was more likely to have LBD pathology and displayed higher a proportion of parkinsonian signs than the Confirmed-AD group. Shim et al (1) found that AD-Mimics had more impairment on the Unified Parkinson’s Disease Rating Scale (UPDRS) and exhibited more parkinsonian signs than Confirmed-AD participants, concluding that the presence of these signs should warn clinicians of the possibility of an alternate diagnosis. However, in our study we found that these signs were associated with increased odds of being included in the Unidentified-AD group, indicating that it is difficult to recognize AD in the face of LBD symptoms. In our sample clinicians that noticed these parkinsonian signs opted for alternative diagnoses in participants who subsequently had ADNC present at autopsy.

Other symptoms associated with inclusion in the Unidentified-AD group included behavioral problems and language decline. Like parkinsonian signs, behavioral problems may indicate to clinicians that an LBD diagnosis is more appropriate than an AD diagnosis, despite the presence of ADNC. This is further supported by the results of our supplementary analysis which examined each behavioral problem included in the grouped covariate. Compared to Confirmed-AD, participants with Unidentified-AD were more likely to have visual hallucinations and REM sleep behavior disorder, which are core clinical features used while diagnosing dementia with Lewy bodies (16).

Klatka et al (15) indicated that if language decline is absent, then this suggests a diagnosis other than AD. Participants in the AD-Mimics group did display less language decline, which further supports that if AD pathologic features are absent, there is a lower likelihood that language decline will be present. Participants in the Unidentified-AD group were younger at death and displayed more neocortical Lewy bodies, FTLD-tau, and FTLD-TDP-43/other pathology than the Confirmed-AD group. Since it is possible that underlying mixed pathologies are driving the clinical symptoms and younger age at death in the Unidentified-AD group, we ran a supplemental analysis comparing the levels of mixed pathology present among the 3 groups, as well as the frequency of ADNC existing alone (i.e. no LBD, FTLD, HS, or vascular pathologies; Supplementary DataTable S2). The results show that in our sample, AD-Mimics had significantly less mixed pathology when compared to the Confirmed-AD group, while the Unidentified-AD group had significantly more mixed pathology than the Confirmed-AD group. The higher levels of mixed pathology in the Unidenitfied-AD group may explain why their clinical symptoms led to diagnoses other than Alzheimer disease. This study adds to the literature showing the importance mixed pathologies have on clinical presentation and prognosis. One review found the prevalence of mixed pathologies in community-based studies to vary between 10% and 74%, with the wide variation in part due to different methods and criteria used in different studies (17). Another study found that over 50% of participants with ADNC had co-occurring LBD pathology or vascular brain injury (18). Other studies found that cognitive trajectories were similar between participants with mixed pathologies and participants who only had ADNC; though, the impact of co-occurring pathologies may have been due to the severity of AD more than the mixed pathologies (18, 19). These findings may explain why this study found that a higher proportion of participants in the Unidentified-AD group had less severe AD (Braak stages III and IV). Unidentified-AD participants may have more mixed pathology but the effect on cognition may have been partially offset by the less severe ADNC.

Misclassification of diagnoses impacts research that informs clinical trials. While we explored clinical characteristics that were associated with incorrect diagnosis, accurate clinical diagnosis of patients with multiple pathologies may be challenging. A more complete panel of biomarkers that capture the extent of neuropathologic features is needed. Our study additionally demonstrates the need to define antemortem criteria for diagnosis of recently distinguished neuropathologic entities. We found that AD-Mimics are older and exhibit more FTLD-TDP pathology than the Confirmed-AD group. In 2018, a consensus group addressed a newly proposed entity, limbic-predominant age-related TDP-43 encephalopathy (LATE). The predominant neuropathological change in LATE is TDP-43 proteinopathy, which is associated with a clinical presentation of amnestic dementia that can mimic Alzheimer’s dementia. LATE is differentiated from FTLD-TDP-43 pathology by distribution of TDP-43 in the brain and by LATE generally affecting older populations (20). It is possible that the AD-Mimics in our sample who had FTLD-TDP-43 and HS pathology may meet the criteria for LATE and future studies should explore the overlap of these groups.

In addition to LATE, it is possible that participants in the AD-Mimics group might exhibit primary age-related tauopathy (PART), another recently defined neuropathological condition that is frequently diagnosed clinically as AD (21). While participants with PART and ADNC both have neuropathological tau deposits, amyloid deposits are absent in definite PART. AD-Mimics and definite PART are both defined as having no neuritic plaques, and like the AD-Mimics in this study, PART participants are generally older and experience less severe cognitive dysfunction (22, 23). Table 4 displays the absence of neuritic plaques in the AD-Mimics, as well as their distribution of Braak stages. Again, it is possible that there is overlap between the AD-Mimics in our sample and PART; future studies comparing AD-Mimics to PART would help elucidate the clinical and neuropathological differences in these 2 groups. LATE and PART are common neuropathologies, indicating the importance of addressing potential relationships between AD-Mimics, and LATE and PART pathology in future research (20, 22).

Our study has some important limitations. First, the NACC database is a case series (or convenience sample) and participants have higher education and are more likely to be white than the general population. This is especially the case for participants who agree to autopsy. Second, we selected participants who consented to autopsy and these participants may differ in characteristics from the overall NACC cohort. Third, the ADCs that contribute data to NACC see an increased proportion of rare neurodegenerative phenotypes such as FTLD, which further limits its generalizability to other populations. Conversely, the prevalence of clinical vascular dementia is lower than that seen in community-based cohorts due to enrollment practices at the ADCs that bias toward participants without vascular dementia. However, despite the selection toward nonvascular dementia at enrollment, there remains substantial vascular disease pathology at autopsy in NACC with up to 70% of participants displaying arteriolosclerosis and 20% displaying microinfarcts at autopsy as of the September 2019 data freeze. A study comparing neuropathological features at autopsy by year of death in the NACC database showed that the presence of microinfarcts has been consistent over time, with 18% of participants having microinfarcts from 2002 to 2011, 19.2% from 2012 to 2013 and 23.9% from 2014 to 2018 (24). Furthermore, a study comparing the frequency of microinfarcts in NACC and the Adult Changes in Thought study (ACT), a longitudinal community-based prospective cohort, found that microinfarcts were common in both populations (19.7% in NACC and 16% in ACT), while between 19.9% and 35% of participants in our 3 study groups displayed microinfarcts (25). Thus, this selection against vascular dementia at enrollment may not have a large effect on the presence of vascular pathology in our sample.

The prevalence of mixed pathologies may also be different between this clinical case series and a community-based cohort, which may limit the generalizability of our results in regards to the effect of mixed pathology on an inaccurate diagnosis. A study examined the difference between NACC and ACT and found that, while there was more vascular pathology in ACT participants, they were also older. Mixed pathologies were common in both populations, and once the age distribution was accounted for the distribution of mixed pathologies among NACC participants and ACT participants was similar, with AD+ LBD and AD+Vascular Brain Injury being the most common mixed pathologies in those populations (18). A final limitation of this study is that we did not explore any differences in diagnostic accuracy by whether Centers use biomarkers to help make their diagnosis as this data is very limited. Future studies should consider the temporal trends in diagnostic accuracy as clinic practices have changed over time to include imaging and other biomarkers. Despite these limitations, the study had the strengths of using data on a large group of individuals from multiple institutions across the United States, all of which use standardized techniques and reporting methods for both clinical and autopsy data (26). This allows us to draw reasonable conclusions regarding the differences in clinical and neuropathological characteristics among 3 distinct groups—Confirmed-AD, AD-Mimics, and Unidentified-AD.

In conclusion, this study found that AD-Mimics experienced less severe cognitive impairment and display more FTLD-tau, FTLD-TDP-43/other, HS, and cerebrovascular pathology than Confirmed-AD participants. Unidentified-AD participants were more likely to display clinical characteristics that warn clinicians of diagnoses other than AD and were more likely to have LBD, FTLD-tau, FTLD-TDP-43/other, and HS present at autopsy. However, the prevalence of rare neurodegenerative diseases such as FTLD in this highly selected sample may not reflect the frequencies seen in community-based cohorts and future studies should explore these neuropathological features in other cohorts. This study supports previous literature suggesting that AD-Mimics have less severe cognitive impairment than Confirmed-AD participants and indicates that the accuracy of a clinical diagnosis of AD suffers in the presence of clinical phenotypes associated with other diseases. Underlying mixed pathologies may also contribute to inaccurate diagnoses, which highlights the importance of developing a complete panel of in vivo biomarkers unique to each disease pathology. This study also demonstrates the need to define antemortem criteria for the diagnosis of recently distinguished neuropathologic entities of LATE and PART.

This project was funded in whole or in part with federal funds from the National Cancer Institute, National Institutes of Health (Contract No. HHSN261200800001E). Additional support was provided by the Nancy and Buster Alvord endowment (to C.D.K.).

The NACC database is funded by NIA/NIH Grant U01 AG016976. NACC data are contributed by the NIA-funded ADCs: P30 AG019610 (PI Eric Reiman, MD), P30 AG013846 (PI Neil Kowall, MD), P50 AG008702 (PI Scott Small, MD), P50 AG025688 (PI Allan Levey, MD, PhD), P50 AG047266 (PI Todd Golde, MD, PhD), P30 AG010133 (PI Andrew Saykin, PsyD), P50 AG005146 (PI Marilyn Albert, PhD), P50 AG005134 (PI Bradley Hyman, MD, PhD), P50 AG016574 (PI Ronald Petersen, MD, PhD), P50 AG005138 (PI Mary Sano, PhD), P30 AG008051 (PI Thomas Wisniewski, MD), P30 AG013854 (PI M. Marsel Mesulam, MD), P30 AG008017 (PI Jeffrey Kaye, MD), P30 AG010161 (PI David Bennett, MD), P50 AG047366 (PI Victor Henderson, MD, MS), P30 AG010129 (PI Charles DeCarli, MD), P50 AG016573 (PI Frank LaFerla, PhD), P50 AG005131 (PI James Brewer, MD, PhD), P50 AG023501 (PI Bruce Miller, MD), P30 AG035982 (PI Russell Swerdlow, MD), P30 AG028383 (PI Linda Van Eldik, PhD), P30 AG053760 (PI Henry Paulson, MD, PhD), P30 AG010124 (PI John Trojanowski, MD, PhD), P50 AG005133 (PI Oscar Lopez, MD), P50 AG005142 (PI Helena Chui, MD), P30 AG012300 (PI Roger Rosenberg, MD), P30 AG049638 (PI Suzanne Craft, PhD), P50 AG005136 (PI Thomas Grabowski, MD), P50 AG033514 (PI Sanjay Asthana, MD, FRCP), P50 AG005681 (PI John Morris, MD), P50 AG047270 (PI Stephen Strittmatter, MD, PhD).

The authors have no duality or conflicts of interest to declare.