Provisional research criteria for the behavioral variant of Alzheimer’s disease A systematic review and meta-analysis

Search strategy, selection and outcomes

This study was conducted following pre-specified methods (PROSPERO registration number: CRD42021243497) and reported following the PRISMA guidelines (Table-S1). We performed a systematic literature search in PubMed/Medline and Web-of-Science databases. We searched studies including clinically diagnosed 1) AD cases with “frontal” or “behavioral” presentations, or 2) bvFTD cases with neuropathological evidence of AD (see full database queries in Table-S2). We included peer-reviewed articles, written in English and presenting original research with human data only. Screening was first conducted at the title/abstract level in Rayyan (https://rayyan.qcri.org/). Reference lists were additionally cross-checked for eligible studies. Two independent reviewers (RO/EHS) screened titles and abstracts. Ambiguous records were discussed with a third author (YALP) to reach consensus. Studies were eligible when 1) including cases or groups of patients presenting with early and predominant behavioral changes with a clinical diagnosis, biomarker support and/or neuropathological evidence of AD, and 2) behavioral/neuropsychiatric, neuropsychological, neuroimaging and/or neuropathological data were presented. Studies were excluded when 1) describing patients with isolated executive dysfunction in the absence of behavioral symptoms, 2) there was biomarker and/or neuropathological evidence for a non-AD pathology as primary etiology. Studies were only eligible for the meta-analysis if a bvAD group was compared against typical AD (tAD) and/or bvFTD groups. We extracted demographic (age, sex), clinical (behavioral features per bvFTD criteria¹³ or neuropsychiatric symptoms per Neuropsychiatric Inventory [NPI¹⁴]), neuropsychological (Mini-Mental State Examination [MMSE], memory and executive function tests), neuroimaging (structural MRI, [¹⁸F]FDG-PET, perfusion SPECT, amyloid-PET, tau-PET) and neuropathological (amyloid-β and tau) characteristics from all studies. After eligibility assessment for inclusion, meta-analyses were constructed using pooled clinical (behavioral or neuropsychiatric questionnaires), neuropsychological (MMSE, memory and executive functioning tests) and neuropathological (amyloid-β and tau) data. The lack of uniform reporting of effect sizes among neuroimaging methods across studies did not allow a meta-analysis, hence these findings were analyzed using systematic review (Table-1).

Statistical analysis

Meta-analysis was used to examine whether bvAD differed from tAD and bvFTD in terms of behavioral/neuropsychiatric and neuropsychological features, and whether bvAD differed from tAD in the distribution of amyloid-β and tau pathology defined at autopsy. Missing data were requested from the authors of three studies (3/3 responded).^15-17 We calculated the pooled standardized mean differences and 95% confidence intervals using Hedges’ g random-effects models in the “meta” package (R v4.0.2), with significance levels of p<0.05. We used random effects because we assumed that the true effect size would be study dependent, due to high heterogeneity in samples, methodology and outcomes among studies.

Statistical heterogeneity for the meta-analyses was assessed using the I² statistic, with I²>75% indicating substantial heterogeneity. Heterogeneity across studies was substantial for analyses including behavioral/neuropsychiatric symptoms, memory and executive measures (range I²: 70-96%) and moderate for analyses including neuropathological data (range I²: 0-51%). Publication bias was assessed by visual inspection of funnel plots, which indicated substantial publication bias (Figure-S1-S3). Two authors (EHS/CG) independently assessed risk of bias using the ROBINS-I risk of bias tool for non-randomized studies. The overall risk of bias was serious for two studies and moderate for 11 studies in the meta-analysis (Table-S3, Figure-S4).

We additionally calculated the prevalence of each behavioral feature in the core clinical bvFTD criteria¹³ (i.e., presence of disinhibition, apathy, lack of empathy, compulsiveness and hyperorality) and the 12 items of the neuropsychiatric inventory (NPI, prevalence score ≥1) across studies, and compared bvAD, bvFTD and tAD groups using χ² tests. For the NPI analysis only, we used 769 Aβ-positive tAD cases from the Amsterdam Dementia Cohort (mean age[SD]: 65.9[7.7], 52.4% females, mean[SD] MMSE: 20.3[5.1]).¹⁸

Participants

The systematic literature search yielded 1,257 records, of which 116 studies were assessed at full-text level for eligibility and 83 studies met inclusion criteria (see Figure-S5 for flow chart). Thirteen studies were eligible for meta-analysis. Table-S4 provides an overview of the participant characteristics for all 83 included studies. Across these studies, 591 bvAD cases were enrolled, with a mean(SD) age-at-diagnosis of 62.0(7.3), and (38.3%) of participants were female. The mean(SD) MMSE was 20.1(5.9) and 47.5% carried an APOEε4 allele.

Behavioral/neuropsychiatric symptoms

Meta-analysis indicated that patients with bvAD showed more severe behavioral/neuropsychiatric symptoms than tAD patients (standardized mean difference [SMD, 95% confidence interval]: 1.16[0.74–1.59], p<0.001), and a trend towards less severe behavioral/neuropsychiatric symptoms compared to bvFTD (SMD:-0.22[-0.47–0.04], p=0.10, Figure-1A). Results remained similar when separating bvFTD core criteria and neuropsychiatric features (Figure-S6).

• Download figure
• Open in new tab

Figure 1.

Meta-analyses of behavioral and neuropsychiatric features

Panel A shows the results of meta-analyses on behavioral and neuropsychiatric total scores between bvAD and typical AD (left) and bvAD and bvFTD (right). “n” represents the number of bvAD cases in each study. Panel B shows the mean weighted percentages of participants per diagnostic group fulfilling specific bvFTD core clinical features proposed by Raskovsky et al. (left) or presence of specific neuropsychiatric symptoms measured using the Neuropsychiatric Inventory (NPI, right). Panel C summarizes the differences and similarities between diagnostic groups.

bvAD=behavioral variant of AD, tAD=typical AD, bvFTD=behavioral variant frontotemporal dementia. NPI=Neuropsychiatric Inventory, FAB=Frontal Assessment Battery, PBAC=Philadelphia Brief Assessment of Cognition, DCQ=Dépistage Cognitif de Québec, SMD=standardized mean difference.

Next, we compared proportions of bvFTD features and NPI items as reported in previous studies (Figure-1B, Table-S5^19-25). Compared to bvFTD, bvAD patients less frequently showed compulsive behaviors (45.0%vs68.5%, χ²=22.5, p<0.001) and hyperorality (35.9%vs64.1%, χ²=32.8, p<0.001), but no differences on disinhibition (60.8%vs68.6%, χ²=2.8, p=0.10), apathy (68.8%vs77.4%, χ²=3.7, p=0.05) and lack of empathy (54.6%vs53.6%, χ²=0.1, p=0.83). On the NPI, bvAD patients more frequently showed agitation (67.9vs43.4%, χ²=8.8), hallucinations (28.2vs9.0%, χ²=12.8) and delusions (36.6vs13.4%, χ²=13.4) compared to bvFTD (all p<0.001). Furthermore, bvAD more frequently showed night-time behaviors (39.6vs19.5%, χ²=12.9), euphoria (16.6vs6.8%, χ²=7.9), anxiety (54.2vs31.7%, χ²=10.8), agitation (67.9vs14.8%, χ²=90.3), hallucinations (28.2vs3.1%, χ²=71.2), delusions (36.6vs9.2%, χ²=37.2) and motor behaviors (50.4vs19.8%, χ²=26.2) compared to tAD patients (all p<0.01).

Cognition

Meta-analyses of cognitive data indicated that at initial assessment bvAD patients showed no differences on MMSE compared to tAD (SMD:-0.18[-0.56–0.20], p=0.35) and bvFTD (SMD:-0.22[-0.78–0.35], p=0.46, Figure-2). bvAD showed worse executive performance compared to tAD (SMD:-1.03[-1.74–-0.32], p<0.01), but not compared to bvFTD (SMD:-0.61[-1.75–0.53], p=0.29). Finally, bvAD showed a trend towards worse memory performance compared to bvFTD (SMD:-1.31[-2.75–0.14], p=0.08), but did not differ from tAD (SMD:0.43[-0.46–1.33], p=0.34).

• Download figure
• Open in new tab

Figure 2.

Meta-analyses of cognitive performance

Results of meta-analyses on Mini-Mental State Examination, episodic memory and executive function for the contrast bvAD vs typical AD (left) and bvAD vs bvFTD (right).

bvAD=behavioral variant of AD, tAD=typical AD, bvFTD=behavioral variant frontotemporal dementia. SMD=standardized mean difference.

Neuroimaging

Table-2 provides an overview of neuroimaging studies in bvAD. Structural MRI studies (number of studies [“k”]=6, number of participants [“n”]=92) showed temporoparietal²³, frontotemporal and insular^17,24,25, or frontoparietal²⁶ predominant atrophy patterns across bvAD patients. bvAD did not differ from tAD in three studies^23,26,27, and showed moderately more involvement of frontal regions in bvAD compared to tAD in three other studies^17,24,25. Studies assessing glucose metabolism with [¹⁸F]FDG-PET or perfusion with SPECT (k=7, n=88) also showed heterogeneous results, ranging from a predominantly temporoparietal hypometabolic pattern^27,28 to a mixed frontal and temporoparietal^15,16,29,30 or predominantly frontal pattern³¹. Amyloid-PET studies (k=2, n=28) showed no differences in amyloid-β burden or distribution between bvAD and tAD patients.^26,31 For tau-PET (k=2, n=22), one study showed a temporoparietal pattern with higher uptake in anterior regions in bvAD compared to tAD²⁶, whereas another study showed heterogeneous patterns across bvAD patients³². Findings on functional connectivity (k=3, n=54) and white matter hyperintensities (k=1, n=29) in bvAD are presented in Table-S6.

We distilled two distinct bvAD neuroimaging-phenotypes from the literature, including a posterior-predominant (more “AD-like”) and an anterior-predominant (more “bvFTD-like”) phenotype (Figure-3A). We propose that these phenotypes occur on a continuum (Figure-3B), with the posterior-predominant phenotype being most prevalent (Figure-3C).

• Download figure
• Open in new tab

Figure 3.

Neuroimaging features in bvAD

The top panel (A) shows two cases which serve as examples of two distinct bvAD neuroimaging phenotypes: an “AD-like” posterior-predominant atrophy/tau load pattern and a “bvFTD-like” anterior-predominant atrophy/tau load pattern. In the middle panel (B) we propose that these neuroimaging phenotypes are part of a spectrum that ranges from a typical AD regional distribution to a classical bvFTD regional distribution (from left to right). The brain template images were obtained from https://smart.servier.com/. In the lower panel (C), we present a literature-informed estimated distribution of the regional distribution in bvAD, indicating that “typical AD” and “bvAD-AD-like” patterns are more common than “bvAD-bvFTD-like” and “typical bvFTD”.

Neuropathology

In line with amyloid and tau PET findings, the meta-analyses on neuropathological data showed that bvAD and tAD did not differ in the neuropathological burden of amyloid-β (k=3, n=20) in the frontal cortex (SMD:0.23[-0.36–0.81], p=0.45), medial temporal lobe (SMD:-0.06[-0.65– 0.53], p=0.84) or occipital regions (SMD:-0.16[-1.05–0.73], p=0.73, Figure-S7). Furthermore, there was no difference in tau burden (k=4, n=28) in the frontal cortex (SMD:-0.05[-0.56–0.46], p=0.84), medial temporal lobe (SMD: 0.32[-0.19–0.83], p=0.22), or occipital lobe (SMD:-0.36[-0.95–0.23], p=0.24, Figure-S7).

Systematic review and meta-analyses

bvAD typically presents at a young age (62.0[7.3] years at time-of-diagnosis), is more prominent in males than in females (61.7%vs38.3%, in line with bvFTD but in contrast with tAD³³) and has a lower frequency of APOEε4 carriership compared to tAD (47.5%vs66.1%³⁴). Clinically, bvAD shows a milder behavioral profile compared to bvFTD with less compulsivity and hyperorality, but greater prevalence of neuropsychiatric symptoms such as agitation, delusions and hallucinations. By definition, bvAD shows greater impairment on a range of behavioral and neuropsychiatric measures compared to tAD. The directionality of findings in the meta-analyses of cognitive data suggest that bvAD might show greater memory and executive function deficits compared to bvFTD, and relatively better memory function and worse executive functioning compared to tAD, but further research in larger cohorts is needed to confirm the significance of these findings. The neuroimaging methodology was too heterogenous across studies to conduct a formal meta-analysis, but a systematic review revealed two distinct phenotypes of brain atrophy, hypometabolism and tau pathology in bvAD, with many cases likely occurring on a continuum. The most prevalent bvAD neuroimaging phenotype is an “AD-like” posterior-predominant pattern comprising bilateral temporoparietal regions with limited involvement of the frontal cortex. This observation is congruent with our meta-analysis on neuropathological data showing that bvAD patients were indistinguishable from tAD patients in both amyloid-β and tau load and spatial distribution. The other bvAD phenotype is characterized by a “bvFTD-like” anterior-predominant neuroimaging pattern, including regions (e.g., anterior cingulate cortex, frontal insula, temporal poles) located in brain networks (e.g., the salience network) that are engaged during socio-emotional processing of information.³⁵ Altogether, our systematic review and meta-analyses further refine the bvAD phenotype, but also highlight the need for larger studies with more uniform methodologies and inclusion/exclusion criteria.

Provisional research criteria for bvAD

Our main objective was to propose provisional research criteria for bvAD guided by the results of the systematic review and meta-analyses. The criteria are based on consensus between all authors, consisting of neurologists, neuropsychologists, neuropathologists and neuroscientists. To facilitate widespread use but also take into account the complexity of this phenotype, we offer four levels of evidence (Table-2). The first level (“clinical bvAD”) can be established solely based on clinical information, while level II and III (“possible bvAD” and “probable bvAD”) add biomarker confirmation of amyloid-β and tau pathology. Level IV (“definite bvAD”) is assigned through histopathological or genetic (i.e., presence of pathogenic APP, PSEN1 or PSEN2 mutations) confirmation of AD in conjunction with a bvAD clinical syndrome.

Several issues warrant further explanation. First, both the literature and our clinical experience align with the notion that bvAD is a combined cognitive and behavioral clinical syndrome. We previously showed that cognitive impairment was among the first symptoms reported by patients and caregivers in ∼75% of bvAD cases.²³ In addition, our meta-analysis suggests that episodic memory performance in bvAD is intermediate between tAD and bvFTD, while bvAD shows greater executive dysfunction compared to bvFTD (Figure-1). To enhance the discriminative accuracy between bvAD and bvFTD, objectively confirmed impairment in either memory or executive domains is therefore mandatory. In addition, 2/5 behavioral features of the diagnostic criteria for bvFTD¹³ (i.e., disinhibition, apathy, lack of empathy, compulsiveness and hyperorality) must be present. This 2:5 ratio was selected to sufficiently distinguish bvAD from tAD, but also acknowledge the generally milder behavioral profile in bvAD compared to bvFTD (where 3/6 bvFTD criteria must be present). Second, despite significant differences between bvAD and both bvFTD and tAD (Figure-1), we deemed it premature to include hallucinations and delusions in the core research criteria, as these observations were derived from only two studies.^21,22 Instead, they were added as supportive features and future prospective studies are needed to assess whether they should be incorporated in the core criteria for bvAD. Third, most AD variants have a clear neurodegenerative signature on MRI and/or [¹⁸F]FDG-PET that corresponds to their clinical phenotype, such as left-hemispheric predominance in lvPPA or occipito-temporal/occipito-parietal damage in PCA.^10,11 The neuroimaging literature in bvAD, however is highly inconsistent. Some studies (mainly case studies or case series) showed anterior neurodegenerative patterns that resemble bvFTD, but most group studies showed either a mix of anterior and posterior involvement or a posterior-predominant pattern.^{17,23-28,36,37} Contrary to PCA and lvPPA, we therefore did not incorporate MRI, CT, SPECT or [¹⁸F]FDG-PET read-outs into the core bvAD research criteria, but only added them as supportive features. Fourth, evidence of amyloid-β pathology provided by PET, CSF or plasma biomarkers can upgrade the diagnosis from “clinical bvAD” to “possible bvAD”. Positive amyloid-β biomarkers substantially increase the likelihood that AD is the primary etiology but, given their limited specificity, the possibility of amyloid-β as comorbid pathology cannot be ruled out, especially in older individuals and in APOEε4 carriers.^38,39 The addition of biomarker evidence for tau pathology further increases the certainty for a bvAD diagnosis (i.e., “probable bvAD”). Here we make the distinction between biofluid and neuroimaging markers of tau pathology. For CSF and plasma biomarkers of tau pathology the differential diagnostic value for distinguishing AD from bvFTD is less well established, and like amyloid-β markers they become abnormal relatively early in the disease course, which lowers their specificity.^40,41 Hence, a full AD-like fluid biomarker profile with abnormalities in both amyloid-β and phosphorylated-tau supports a level II diagnosis of “possible bvAD”. Instead, the currently most widely used tau PET ligands (i.e., [¹⁸F]flortaucipir, [¹⁸F]MK6240 and [¹⁸F]RO948) have consistently shown to bind selectively and with high affinity to the tau aggregates formed in AD (i.e., combinations of 3R/4R tau in paired helical filaments), while neocortical tau PET uptake in sporadic bvFTD is negligible, resulting in excellent discriminative accuracy between AD and bvFTD.^42,43 Furthermore, since tau PET-positivity in the neocortex almost exclusively occurs in amyloid-β positive individuals^42,44 we consider a level I diagnosis (“clinical bvAD”) plus tau PET-positivity in an AD-like pattern⁴⁵ supportive of a level III diagnosis of “probable bvAD”. Fifth, although the question whether bvAD and dysexecutive AD (deAD) reside on a single continuum or represent distinct clinical entities is yet unresolved, we deliberately developed criteria specific to bvAD. This was motivated by our previous study showing that only ∼25% of bvAD cases additionally met deAD criteria²³ (hence bvAD occurs in isolation in the majority of cases), as well as a recent publication proposing specific deAD criteria that explicitly exclude behavioral features.¹²

Limitations

There are several limitations. First, bvAD is a rare AD phenotype that, for the most part, has been described in single case studies and case series. The bvAD literature therefore consists of relatively few cohort studies that are generally characterized by modest sample sizes, which resulted in reduced statistical power to detect differences between bvAD versus bvFTD and tAD groups. This was further complicated by substantial heterogeneity in patient samples and outcome measures and subsequent substantial risk of bias across studies. Second, the variability across neuroimaging studies did not allow a meta-analytical approach, hence we interpreted this literature using a systematic review. Third, in the behavioral, cognitive and neuropathological meta-analyses, we combined comparable yet distinct study outcome measures such as different neuropsychological tests for memory and executive functions, questionnaires for neuropsychiatric/behavioral features, or staining methods and selection of brain regions for histopathological assessment of amyloid-β and tau. Fourth, we did not account for possible co-pathologies (e.g., Lewy bodies) that may contribute to the clinical phenotype. Fifth, there were only limited data on behavioral presentations of AD in diverse populations.

Future directions

Akin to the development of diagnostic criteria for PCA, we consider the currently proposed provisional research criteria as a steppingstone towards internationally established consensus criteria for bvAD. For PCA, provisional research criteria were first proposed by two research groups and were subsequently applied by other groups to establish a PCA diagnosis for several years^46,47, followed by widely supported formal diagnostic criteria based on consensus by an international working group.¹⁰ Similarly, our bvAD criteria should improve the consistency and reliability of future research and possibly aid in the clinical assessment of bvAD, which in turn would enhance the diagnostic accuracy of future bvAD criteria to be established by a working group of worldwide experts. There are several promising novel biomarkers and behavioral features that could be included in future bvAD criteria, for example more objective measurements of behavior like social cognition in conjunction with biometric information (e.g., eye-tracking, face reading, galvanic skin response)⁴⁸ or blood-based biomarkers of AD pathology (e.g., p-tau, amyloid-β) and neurodegeneration (e.g., neurofilament light chain [NfL]).⁴⁹ Furthermore, the diagnostic utility of potential bvAD-specific features (e.g., relatively preserved disease insight, presence of hallucinations and delusions) or measures of disease severity (e.g., the FTLD-modified Clinical Dementia Rating scale⁵⁰) should be further investigated.