Introduction
Frontotemporal dementia (FTD) represents a set of phenotypically and pathologically heterogeneous syndromes characterized by focal neurodegeneration in frontal, orbital-insular, and anterior temporal brain regions (Bang et al., Reference Bang, Spina and Miller2015). The clinical presentation of FTD encompasses three primary conditions, a behavioral variant, and two language variants – non-fluent (nfvPPA) and semantic (svPPA) primary progressive aphasia. The behavioral variant of frontotemporal dementia (bvFTD) is characterized by early and prominent change in comportment, personality, and social functioning (Rascovsky et al., Reference Rascovsky, Hodges, Knopman, Mendez, Kramer, Neuhaus, Swieten, Seelaar, Dopper, Onyike, Hillis, Josephs, Boeve, Kertesz, Seeley, Rankin, Johnson, Gorno-Tempini, Rosen, Prioleau-Latham, Lee, Kipps, Lillo, Piguet, Rohrer, Rossor, Warren, Fox, Galasko, Salmon, Black, Mesulam, Weintraub, Dickerson, Diehl-Schmid, Pasquier, Deramecourt, Lebert, Pijnenburg, Chow, Manes, Grafman, Cappa, Freedman, Grossman and Miller2011; LaMarre and Kramer, Reference LaMarre, Kramer, Ravdin and Katzen2013), whereas nfvPPA and svPPA are defined by progressive decline in speech production or semantic knowledge (Gorno-Tempini et al., Reference Gorno-Tempini, Hillis, Weintraub, Kertesz, Mendez, Cappa, Ogar, Rohrer, Black, Boeve, Manes, Dronkers, Vandenberghe, Rascovsky, Patterson, Miller, Knopman, Hodges, Mesulam and Grossman2011; Mesulam et al., Reference Mesulam, Rogalski, Wieneke, Hurley, Geula, Bigio, Thompson and Weintraub2014). Early and accurate differentiation has strong implications for the pathological basis of each syndrome and is critical for guiding treatment and for clinical trial recruitment – which will increasingly rely on etiopathologic predictions (Woolley et al., Reference Woolley, Khan, Murthy, Miller and Rankin2011).
Though the anatomical, molecular, genetic, and neurochemical aspects of FTD are increasingly understood in relation to its clinical phenotypes, neuropsychological correlates of pathological substrates have proven to be inconsistent and at times counterintuitive (LaMarre and Kramer, Reference LaMarre, Kramer, Ravdin and Katzen2013; Sitek et al., Reference Sitek, Barczak and Harciarek2015). The variability in FTD neuropsychological profiles has been attributed to multiple factors – low sample size, phenotypically mixed cohorts, differing diagnostic criteria, and the absence of disease staging information (Harciarek and Cosentino, Reference Harciarek and Cosentino2013; LaMarre and Kramer, Reference LaMarre, Kramer, Ravdin and Katzen2013). Despite the difficulties in characterizing cognition in bvFTD, it is generally accepted that its early cognitive manifestations result in gradual changes in executive functioning where visuospatial skills are preserved until later in the course (Ng et al., Reference Ng, Rademakers and Miller2015). Given the difficulties discriminating bvFTD and AD using cognitive tests (Hutchinson and Mathias, Reference Hutchinson and Mathias2007), there has been a considerable focus on qualitative differences in test scores. Kramer et al. (Reference Kramer, Jurik, Sha, Rankin, Rosen, Johnson and Miller2003) and Thompson et al. (Reference Thompson, Stopford, Snowden and Neary2005) found that bvFTD patients made more executive errors across tasks compared with svPPA and AD patients. In the latter study, classification of error types in bvFTD and AD cases had utility over traditional test scores and improved classification accuracy from 71% to 96% (Thompson et al., Reference Thompson, Stopford, Snowden and Neary2005).
Patients with nfvPPA and svPPA demonstrate a cognitive decline in domains beyond speech/language as neurodegeneration progresses. However, findings from autopsy case series and longitudinal studies in FTD contradict the view that cognitive performances converge into a common subtype. Libon et al. (Reference Libon, Xie, Wang, Massimo, Moore, Vesely, Khan, Chatterjee, Coslett, Hurtig, Liang and Grossman2009) found that distinct cognitive profiles observed at disease onset were maintained over an 8-year follow-up period with double dissociations noted between the behavioral and language variants of FTD. Patients with svPPA showed poor naming and relative sparing of working memory over the illness course, whereas the opposite pattern was observed in nfvPPA patients. Similar double dissociations were observed in a pathologically-confirmed FTD sample of tau-negative and tau-positive cases (Grossman et al., Reference Grossman, Xie, Libon, Wang, Massimo, Moore, Vesely, Berkowitz, Chatterjee, Coslett, Hurtig, Forman, Lee and Trojanowski2008).
Diagnosis of FTD requires an assessment of a patient's social and emotional functioning. Clinicians typically rely on caregiver ratings, though performance-based measures of social cognition are increasingly utilized in research and clinical contexts. Impaired social conversation (Rousseaux et al., Reference Rousseaux, Sève, Vallet, Pasquier and Mackowiak-Cordoliani2010) and deficits in theory of mind and empathy (Rankin et al., Reference Rankin, Kramer and Miller2005; Eslinger et al., Reference Eslinger, Moore, Troiani, Antani, Cross, Kwok and Grossman2007; Bora et al., Reference Bora, Walterfang and Velakoulis2015) are frequent findings in FTD patients (Eslinger et al., Reference Eslinger, Dennis, Moore, Antani, Hauck and Grossman2005). Social cognition has been examined in prior meta-analyses in bvFTD (Hutchinson and Mathias, Reference Hutchinson and Mathias2007; Bora et al., Reference Bora, Walterfang and Velakoulis2015, Reference Bora, Velakoulis and Walterfang2016; Beeldman et al., Reference Beeldman, Raaphorst, Klein Twennaar, Govaarts, Pijnenburg, de Haan, de Visser and Schmand2018); however, the magnitude of these deficits have not been systematically compared between bvFTD, svPPA and nfvPPA. Similarly, the utility of olfactory measures has been well-documented in Alzheimer's and Parkinson's disease (Murphy et al., Reference Murphy, Gilmore, Seery, Salmon and Lasker1990; Mesholam et al., Reference Mesholam, Moberg, Mahr and Doty1998; Devanand et al., Reference Devanand, Michaels-Marston, Liu, Pelton, Padilla, Marder, Bell, Stern and Mayeux2000, Reference Devanand, Lee, Manly, Andrews, Schupf, Doty, Stern, Zahodne, Louis and Mayeux2015) but is relatively unknown in FTD (Luzzi et al., Reference Luzzi, Snowden, Neary, Coccia, Provinciali and Lambon Ralph2007). Olfaction is linked to hedonic capacity (Kamath et al., Reference Kamath, Moberg, Kohler, Gur and Turetsky2011), altered food preferences (Piwnica-Worms et al., Reference Piwnica-Worms, Omar, Hailstone and Warren2010), apathy (Seligman et al., Reference Seligman, Kamath, Giovannetti, Arnold and Moberg2013), and increased preference for sweets (Seeley et al., Reference Seeley, Crawford, Rascovsky, Kramer, Weiner, Miller and Gorno-Tempini2008), all of which are diagnostic features of bvFTD and are observed in the language variants. These linkages suggest the assessment of olfaction may be pertinent to FTD pathophysiology.
Meta-analytic methods can increase our understanding of cognitive profiles in FTD subgroups through the statistical synthesis of independent experiments. To date, no study has comprehensively characterized and compared cognitive, social cognitive, and olfactory profiles in FTD subgroups. In this study, we compared effect sizes of these domains within and between bvFTD, svPPA, and nfvPPA groups and employed meta-regression methods to analyze the influence of demographic and disease characteristics on effect size magnitude.
Methods
Literature search
A computerized literature search was conducted on PubMed using index terms for each patient group and task domain (see online Supplementary Materials Appendix A). The search parameters were limited to English-language publications with human participants available from 1980 through 26 January 2018. The search yielded 3682 publications that were reviewed for inclusion. Additional articles were identified from references and review articles.
Study selection
Two authors (VK/GC) screened abstracts and articles for the following inclusion criteria:
• The presence of an FTD cohort.
• An age-matched comparison group without subjective cognitive/neurologic complaints. Studies using normative data were excluded. Single case studies were excluded, whereas case series were included if control data were provided.
• Sufficient data to generate an effect size.
In cases of sample and outcome redundancy, the publication with the largest sample size and set of outcome variables was used. Baseline data were used in studies that collected outcomes across multiple time points. For the current meta-analysis, 470 articles were retained. See Fig. 1 for a flow chart of article inclusion/exclusion and online Supplementary Materials Appendix B for a list of included studies.
Fig. 1. Flowchart of literature search and study selection. *Two publications reported different outcome measures but used the same patient and control sample and were treated as one publication.
Data extraction
Demographic/clinical characteristics
Data representing 11 782 FTD patients and 19 451 controls were analyzed. Patient groups were classified as bvFTD, bvFTD phenocopy, svPPA, nfvPPA, and mixed-FTD. The mixed-FTD group comprised patients spanning more than one FTD diagnostic group or patients delineated by their cerebral atrophy patterns (e.g. temporal v. frontal variant FTD). The following demographic and clinical information was coded: mean control age, mean patient age, percentage of men in patient group, percentage of men in control group, mean education for controls, mean education for patients, age of illness onset, illness duration, and criteria used for diagnosis (see Table 1 for descriptive information). Though the heterogeneity of the mixed-FTD group and a small number of studies within the bvFTD phenocopy group precluded their inclusion in analyses, their descriptive information is included in Table 1.
Table 1. Sample characteristics
bvFTD, behavioral variant frontotemporal dementia; svPPA, semantic variant primary progressive aphasia; and nfvPPA, non-fluent variant PPA.
a Numbers for mixed-FTD exceed the total number of studies as five subgroups comprised patients meeting criteria for more than one diagnostic criteria.
Outcome measures
Overall, 5720 measures were imputed from 470 studies and cross-checked for accuracy. Outcome measures were coded as one of three primary domains: cognition, social cognition, or olfaction. Cognitive domain tasks were further assigned to one of nine cognitive subdomains: global cognition, intellectual functioning (IQ), processing speed, attention, language, ideational fluency, memory, visuospatial functioning, and executive functioning. Facets of each cognitive subdomain were also coded. Task domains, cognitive subdomains, facets of each subdomain, and example tests are presented in online Supplementary Materials Appendix C. Disagreements regarding categorization were resolved by discussion (see online Supplemental Materials Appendix D). We tabulated an index of executive errors across all measures similar to Kramer et al. (Reference Kramer, Jurik, Sha, Rankin, Rosen, Johnson and Miller2003).
Statistical analysis
We performed analyses with Comprehensive Meta-Analysis, Version 3.0 (CM3). Due to the range of populations and outcomes assessed, we used a random-effects model to account for within- and between-study variation in effect size estimates as this model assumes greater variability between studies than sampling error alone. Random-effects models assume that effect sizes are sampled from a universe of effect sizes. Studies were weighted according to their inverse variance estimates to account for differences in study sample size when mean effect sizes were computed.
To standardize group differences, we calculated Hedges’ g, the mean difference between the patient and control group divided by the pooled standard deviation. A negative g indicates poorer performance by patients. Effect size direction was inverted for tests in which larger scores indicated poorer performance (e.g. time, errors). Effect sizes were classified as small (g = −0.2), medium (g = −0.5), or large (g ⩾ −0.8). We also conducted sensitivity analyses to identify potential outliers. Using the ‘one study removed’ CM3 module, a random-effects mean and standard error was calculated as each study is removed one at a time from the analysis (Tobias, Reference Tobias1999). To examine publication bias, analyses were conducted using a funnel plot for graphical representation and an adjusted rank-correlation test, using methods by Begg and Mazumdar (Reference Begg and Mazumdar1994) and Egger et al. (Reference Egger, Davey Smith, Schneider and Minder1997).
Homogeneity in effect sizes was assessed using the Cochran's Q-statistic. To evaluate variability among effect sizes, clinical and demographic characteristics were explored. We also calculated an index of heterogeneity or I 2-statistic, which describes the proportion of observed dispersion reflecting differences in true scores rather than sampling error (see online Supplementary Materials Appendix E).
Continuous moderator variables were assessed using meta-regression analyses. Categorical moderator variables were assessed using procedures described by Borenstein et al. (Reference Borenstein, Hedges, Higgins and Rothstein2009) in which between-group variance (Q B) is used to test whether effect sizes differ between categories of a given moderator. Analyses that involved domains with fewer than five studies were considered preliminary and were conducted for descriptive purposes only (Turner et al., Reference Turner, Bird and Higgins2013).
Results
Comparisons of domains, subdomains, and error index between FTD groups
Overall domains
We first compared effect sizes of neuropsychological, social cognitive and olfactory domains between the bvFTD, svPPA, and nfvPPA groups (see Table 2 and Fig. 2). For the cognitive domain, the three groups differed significantly (Q B[2] = 25.61, p < 0.001). SvPPA patients showed the largest cognitive deficits followed by those with nfvPPA and bvFTD, respectively. However, effect sizes were comparable between groups when contributions of the language subdomain were excluded (Q B[2] = 1.66, p = 0.44). Effect sizes for the social cognition (Q B[2] = 1.95, p = 0.38) and olfaction (Q B[2] = 0.52, p = 0.77) domain was not statistically different between groups.
Fig. 2. Effect sizes (± 95% CI) for domains and cognitive subdomains by FTD group. bvFTD , behavioral variant frontotemporal dementia; svPPA, semantic variant primary progressive aphasia; nfvPPA, non-fluent variant PPA; COG, Cognition; SOC, Social Cognition; OLF, Olfaction; GLOB, Global Measures; IQ, Intellectual Functioning; ATT, Attention; PS, Processing Speed; FLU, Ideational Fluency; LANG, Language; MEM, Memory, VISUO, Visuospatial; EF, Executive Functioning.
Table 2. Effect sizes (Hedges’ g ± 95% CI) for task domains and subdomains in bvFTD, svPPA, and nfvPPA groups
bvFTD or B, behavioral variant frontotemporal dementia; svPPA or S, semantic variant primary progressive aphasia; and nfvPPA or N, non-fluent variant PPA; Cochran's Q-statistic and p values are presented for comparisons across all three FTD groups. When overall comparisons were statistically significant, post-hoc domain contrasts were completed; ≈ indicates that study effects sizes were not statistically significant from one another.
a Effect sizes in brackets are presented for descriptive purposes due to the small number of studies included in the analysis (Turner et al., Reference Turner, Bird and Higgins2013).
Cognitive subdomains
On global cognitive screening measures, bvFTD and nfvPPA patients had comparable effect sizes (Q B[1] = 0.26, p = 0.61); both groups had fewer deficits relative to svPPA patients (all p’s ⩽ 0.001). Effect sizes for IQ and math did not differ significantly between groups. Compared with svPPA and bvFTD, the nfvPPA group had the largest attention and processing speed deficits. SvPPA cases had greater attentional deficits compared with bvFTD cases (Q B[1] = 35.69, p < 0.001). In contrast, bvFTD cases tended to show greater processing speed deficits compared with svPPA cases (Q B[1] = 3.51, p = 0.06). Consistent with expectation, the svPPA group had greater language deficits relative to nfvPPA and bvFTD patients (p’s < 0.001), and nfvPPA patients had greater language impairment compared with bvFTD patients (Q B[1] = 11.40, p < 0.001).
BvFTD and nfvPPA patients had comparable effect sizes for executive functioning tasks (Q B[1] = 0.86, p = 0.35); both groups had greater deficits than the svPPA group. The PPA groups had similar effect sizes on ideational fluency measures (Q B[1] = 1.30, p = 0.25), but greater deficits than the bvFTD group (p's ⩽ 0.03). The bvFTD and svPPA groups showed a similar degree of memory impairment (Q B[1] = 0.15, p = 0.70); both groups had greater deficits than the nfvPPA group (p’s < 0.001). On visuospatial tasks, bvFTD patients showed greater impairment compared with the nfvPPA (Q B[1] = 6.83, p < 0.01) and svPPA groups (Q B[1] = 36.20, p < 0.001); the nfvPPA group performed more poorly on visuospatial tasks relative to svPPA patients (Q B[1] = 6.10, p = 0.01).
Executive error index
BvFTD patients had greater deficits on the executive error index compared with nfvPPA (Q B[1] = 5.82, p = 0.02) and svPPA groups (Q B[1] = 10.28, p < 0.001); nfvPPA and svPPA cases showed highly similar effect sizes on this index (Q B[1]<0.001, p > 0.99).
Meta-analytic results in bvFTD
Across all bvFTD studies (n = 289), the overall effect size was large (g = −1.61, 95% CI −1.70 < δ < −1.51) and significantly heterogeneous (Q B[288] = 1566.16, p < 0.001). The sensitivity analysis revealed that the smallest effect size of −1.56 (95% CI −1.64 < δ < −1.48) and the largest effect size of −1.61 (95% CI −1.71 < δ < −1.52) fell within the confidence interval of the mean effect size reported indicating minimal influence from an individual study. We compared effect sizes between domains within the bvFTD group. Effect sizes for measures of cognition (n = 285; g = −1.60; 95% CI −1.70 < δ < −1.50), social cognition (n = 52; g = −1.76; 95% CI −1.99 < δ < −1.53), and olfaction (n = 4; g = −1.90; 95% CI −2.78 < δ < −1.02) were large and not statistically different (Q B[2] = 1.84, p = 0.40). Assessment of publication bias for each domain revealed significant rank correlation tests and Egger tests (p's < 0.03). The classic fail-safe N indicated that 405 611 studies (cognition), 14 174 (social cognition), and 56 (olfaction) studies reporting a zero effect would be required to reduce the significance of the observed effect to a p value greater than 0.05. As such, publication bias imposed minimal influence on the results for the bvFTD sample.
Moderator analysis in bvFTD
Demographic and clinical characteristics
We next evaluated the potential effect size moderators within the bvFTD sample. Sex composition in bvFTD patients was a significant predictor of study effect size; a lower proportion of men was associated with greater deficits on all tasks (Z = 2.37, p = 0.02). This effect was significant for the cognitive domain only (Z = 2.19, p = 0.03) – and, within the cognitive domain, the association was significant for attentional deficits (Z = 2.62, p < 0.01). Higher patient education was associated with smaller deficits (Z = 2.67, p < 0.01). In contrast, control sex composition s (Z = 0.89, p = 0.37), patient age (Z = 0.59, p = 0.55), control age (Z = 0.27, p = 0.79), and control education (Z = 0.94, p = 0.35) were not significant moderators. In bvFTD patients, illness duration (Z = 1.57, p = 0.12) and age of onset (Z = −0.30, p = 0.76) were not associated with effect size.
Subdomain comparisons in bvFTD
For descriptive purposes, we compared olfactory and social cognitive domains to cognitive subdomains. The greatest deficits in bvFTD were observed for measures of olfaction, global cognition (n = 250; g = −1.84; 95% CI −1.96 < δ < −1.72), ideational fluency (n = 102; g = −1.83; 95% CI −2.02 < δ < −1.64), and social cognition. Relatively smaller deficits were observed for mathematic ability (n = 19; g = −0.97; 95% CI −1.34 < δ < −0.54), attention (n = 85; g = −0.96; 95% CI −1.16 < δ < −0.76), and visuospatial functioning (n = 75; g = −0.84; 95% CI −1.06 < δ < −0.63). Effect sizes for domains and subdomains are shown in Table 2. See online Supplementary Materials Appendix F1 for Cochran's Q-statistic and p values for contrasts between domains and cognitive subdomains.
Comparisons between facets of cognitive subdomains in bvFTD
We compared effect sizes between facets of each cognitive subdomain (see Table 2).
IQ
BvFTD patients had comparable verbal and performance IQ deficits (Q B[1] = 1.24, p = 0.27).
Attention
BvFTD patients showed greater working memory impairments relative to simple attention tasks (Q B[1] = 7.67, p < 0.01).
Ideational fluency
Differences in verbal and novel design fluency were not statistically significant (p's ⩾ 0.10).
Language
BvFTD patients showed the largest deficits for reading and naming and the smallest deficits for repetition and semantic knowledge. Cochran's Q-statistic and p values for contrasts between language facets are shown in online Supplementary Materials Appendix G.
Memory
Effect sizes were comparable for immediate verbal and visuospatial memory (Q B[1] = 0.10, p = 0.75) and between measures of verbal and visuospatial recognition memory (Q B[1] = 0.69, p = 0.41). In contrast, greater deficits were observed for delayed verbal recall tasks compared with delayed visuospatial recall (Q B[1] = 5.77, p = 0.02).
Visuospatial skills
Effect sizes for motor-free visuoperceptual and visuoconstruction measures were nearly identical (Q B[1] = 0.01, p = 0.94).
Executive functioning
Effect sizes for response inhibition, visual set-shifting, and concept formation were not statistically different (p's ⩾ 0.21).
Meta-analytic results in svPPA
The overall effect size for svPPA studies was large (n = 199, g = −1.99, 95% CI −2.12 < δ < −1.86) and significantly heterogeneous (Q B[198] = 848.53, p < 0.001). Sensitivity analysis revealed that the lowest (g = −1.97; 95% CI −2.10 < δ < −1.84) and highest effect sizes (g = −2.00; 95% CI −2.13 < δ < −1.87) fell within mean effect size confidence interval, which indicated minimal influence from a single study. Effect sizes for measures of cognition (n = 199; g = −2.02; 95% CI −2.15 < δ < −1.89), social cognition (n = 20; g = −1.53; 95% CI −1.93 < δ < −1.13), and olfaction (n = 5; g = −2.33; 95% CI −3.22 < δ < −1.44) were heterogeneous (Q B[2] = 5.59, p = 0.05). Deficits for cognitive and olfactory domains were comparable (Q B[1] = 0.44, p = 0.51); both were more impaired than the social cognition domain (p’s ⩽ 0.05). Publication bias indices were statistically significant for each domain (p's < 0.05); calculation of the classic fail-safe N indicated that 178 024 (cognition), 1366 (social cognition), and 89 (olfaction) studies reporting a zero effect would be required to reduce the significance of the observed effect to a pvalue greater than 0.05. As such, publication bias imposed minimal influence on the reported svPPA results.
Moderator analysis in svPPA
Demographic and clinical characteristics
We evaluated demographic and clinical characteristics of the svPPA sample to examine their influence on study effect size. Age, education, and sex composition within svPPA patients (p’s > 0.28) and controls (p’s > 0.61) were not associated with study effect size. In svPPA patients, illness duration (Z = −0.42, p = 0.67) and age of onset (Z = 0.73, p = 0.46) were not statistically significant.
Subdomain comparisons in svPPA
Aside from language (n = 141; g = −3.05; 95% CI −3.22 < δ < −2.88), the largest cognitive deficits were on measures of ideational fluency (n = 99; g = −2.43; 95% CI 2.62 < δ < −2.25), global cognition (n = 153; g = −2.28; 95% CI −2.43 < δ < −2.14), and olfaction. Relatively smaller deficits were observed on measures of math (n = 30; g = −0.79; 95% CI −1.11 < δ < −0.47), attention (n = 92; g = −0.55; 95% CI −0.73 < δ < −0.37), and visuospatial functioning (n = 94; g = −0.29; 95% CI −0.47 < δ < −0.11). Effect sizes of all task domains and subdomains are shown in Table 2. See online Supplementary Materials Appendix F2 for Cochran's Q-statistic and p values for contrasts between cognitive subdomains.
Comparisons between facets of cognitive subdomains in svPPA
We compared effect sizes between facets of each cognitive subdomain (see Table 2).
IQ
SvPPA patients had greater verbal than performance IQ deficits (Q B[1] = 81.59, p < 0.001).
Attention
SvPPA patients had comparable deficits for working memory and simple attention (Q B[1] = 0.23, p = 0.64).
Ideational fluency
Patients with svPPA had significantly greater deficits for category-guided relative to letter-guided verbal fluency and novel design fluency tasks (all p's < 0.001). Greater deficits were observed on letter-guided verbal fluency compared with design fluency (Q B[1] = 24.33, p < 0.001).
Language
Consistent with expectation, the largest deficits in the language subdomain was observed for naming tasks. The smallest deficits were observed for repetition measures. For descriptive purposes, Cochran's Q-statistic and p values for contrasts between language facets are shown in online Supplementary Materials Appendix G.
Memory
As expected, patients with svPPA had greater deficits on verbal memory tasks (immediate recall, delayed recall, and recognition memory) when compared with visuospatial memory tasks (all p’s < 0.003).
Visuospatial skills
Though the overall effect size for the visuospatial domain was small, greater deficits were observed for motor-free visuoperceptual tasks when compared with visuoconstruction measures (Q B[1] = 4.16, p = 0.04).
Executive functioning
Individuals with svPPA showed greater impairment on measures of response inhibition relative to measures of visual set-shifting (Q B[1] = 16.20, p < 0.001) and concept formation (Q B[1] = 6.94, p < 0.01). Furthermore, greater deficits were observed for visual set-shifting tasks relative to measures of concept formation (Q B[1] = 3.84, p = 0.05).
Meta-analytic results in nfvPPA
The overall effect size across nfvPPA studies was large (n = 110, g = −1.73, 95% CI −1.88 < δ < −1.59) and significantly heterogeneous (Q B[109] = 361.34, p < 0.001). Sensitivity analysis revealed that the smallest (g = −1.71; 95% CI −1.85 < δ < −1.56) and largest effect sizes (g = −1.75; 95% CI −1.89 < δ < −1.61) fell within the mean effect size confidence interval, which indicated minimal influence from a single study. Effect sizes for measures of cognition (n = 110; g = −1.75; 95% CI: −1.89 < δ < −1.60), social cognition (n = 9; g = −1.38; 95% CI −1.87 < δ < −0.88), and olfaction (n = 2; g = −1.79; 95% CI −2.93 < δ < −0.65) were large and not statistically different within the nfvPPA group (Q B[2] = 2.03, p = 0.36). The fail-safe N algorithm was significant for the cognition and social cognition domains (p's < 0.001); whereas the Egger test was not statistically significant for the social cognition domain. To reduce the significance of the observed effect (i.e. p > 0.05), 50 020 (cognition) and 211 (social cognition) studies reporting a zero effect would be required. Publication bias indices could not be assessed for the olfactory domain due to the low sample size.
Moderator analysis in nfvPPA
Demographic and clinical characteristics
In nfvPPA, sex composition had a marginal influence on effect size magnitude (Z = −1.77, p = 0.08); a higher percentage of men in the patient group was marginally associated with greater deficits. In contrast, control sex composition was not associated with study effect size (Z = −0.34, p = 0.74). Patient age (Z = 1.46, p = 0.15), control age (Z = −0.30, p = 0.77), patient education (Z = −0.58, p = 0.56), and control education (Z = −0.84, p = 0.40) were not statistically significant moderators. Among nfvPPA patients, illness duration (Z = −1.61, p = 0.11) and age of onset (Z = 1.37, p = 0.17) were not significantly associated with study effect size.
Subdomain comparisons in nfvPPA
For descriptive purposes, we compared study effect sizes between cognitive subdomains and the social cognitive and olfactory domains. Cochran's Q-statistic and p values for domain contrasts between cognitive subdomains are shown in online Supplementary Materials Appendix F3. The largest deficits in nfvPPA were observed for ideational fluency (n = 47; g = −2.23; 95% CI −2.47 < δ < −1.99), global cognition (n = 91; g = −1.88; 95% CI −2.05 < δ < −1.71), language (n = 67; g = −1.74; 95% CI −1.94 < δ < −1.54), and processing speed (n = 24; g = −1.92; 95% CI −2.25 < δ < −1.59). Relatively smaller deficits were observed for memory (n = 48; g = −0.95; 95% CI −1.17 < δ < −0.72) and visuospatial functioning (n = 45; g = −0.54; 95% CI −0.76 < δ < −0.31).
Comparisons between facets of cognitive subdomains in nfvPPA
We compared effect sizes between facets of each cognitive subdomain in nfvPPA (see Table 2).
IQ
Consistent with expectation, individuals with nfvPPA had greater verbal IQ deficits compared with performance IQ deficits (QB[1] = 22.87, p < 0.001).
Attention
NfvPPA patients showed comparable deficits on simple attention and working memory measures (Q B[1] = 0.004, p = 0.95).
Ideational fluency
Deficits in category-guided verbal fluency tasks were not significantly different from letter-guided or design fluency tasks (p's > 0.22). Letter-guided and novel design fluency tasks were comparable (Q B[1] = 3.18, p = 0.08).
Language
The greatest deficits were observed for speech output and reading. The smallest were observed for semantic knowledge and comprehension. For descriptive purposes, domain contrasts between facets of the language subdomain are shown in online Supplementary Materials Appendix G.
Memory
Contrary to expectation, deficits for immediate verbal and visuospatial memory were not statistically different (Q B[1] = 0.39, p = 0.53), nor were measures of verbal and visuospatial recognition memory (Q B[1] = 0.26, p = 0.61). In contrast and consistent with expectation, nfvPPA patients showed greater deficits for delayed verbal recall tasks than for delayed visuospatial recall (Q B[1] = 4.23, p = 0.04).
Visuospatial skills
Deficits for motor-free visuoperceptual tasks were significantly smaller than for visuoconstruction measures (Q B[1] = 4.43, p = 0.04).
Executive functioning
Individuals with nfvPPA had greater deficits on measures of response inhibition v. measures of visual set-shifting (Q B[1] = 9.49, p = 0.002) and concept formation (Q B[1] = 5.12, p = 0.02). Effect sizes were comparable for visual set-shifting and concept formation (Q B[1] = 1.26, p = 0.26).
Discussion
In this meta-analysis, we found that bvFTD, svPPA and nfvPPA patients displayed varying profiles of impairment on measures of cognition, social cognition, and olfaction. A limited number of demographic factors moderated effect size magnitude within each group. Higher education in bvFTD patients was associated with smaller deficits, a finding consistent with the positive influence of cognitive reserve on symptom progression either by delaying detection or slowing progression (Perneczky et al., Reference Perneczky, Diehl-Schmid, Drzezga and Kurz2007a; Premi et al., Reference Premi, Gazzina, Bozzali, Archetti, Alberici, Cercignani, Bianchetti, Gasparotti, Turla, Caltagirone, Padovani and Borroni2013; Lenehan et al., Reference Lenehan, Summers, Saunders, Summers and Vickers2015). In bvFTD, samples with higher percentages of women had greater attentional deficits. Though research regarding sex differences between FTD pathology subtypes and clinical syndromes is scarce, one study found that men with bvFTD showed greater cognitive reserve (Perneczky et al., Reference Perneczky, Diehl-Schmid, Förstl, Drzezga and Kurz2007b). Furthermore, sex differences in the prevalence of FTLD mutations have been observed (Curtis et al., Reference Curtis, Masellis, Hsiung, Moineddin, Zhang, Au, Millett, Mackenzie, Rogaeva and Tierney2017) as have mutation-specific patterns of cognitive decline (Jiskoot et al., Reference Jiskoot, Panman, van Asseldonk, Franzen, Meeter, Donker Kaat, van der Ende, Dopper, Timman, van Minkelen, van Swieten, van den Berg and Papma2018). Future work examining the influence of sex on cognitive profiles in bvFTD could improve the prognostic utility of assessment measures. Contrary to expectation, illness duration was not significantly associated with study effect size in any cohort. These findings are difficult to reconcile with existing longitudinal studies of cognition in FTD. Libon et al. (Reference Libon, Xie, Wang, Massimo, Moore, Vesely, Khan, Chatterjee, Coslett, Hurtig, Liang and Grossman2009) examined cognitive trajectories in the behavioral and language variants of FTD over an 8-year follow-up period and found declines in all cognitive domains assessed. Grossman et al. (Reference Grossman, Xie, Libon, Wang, Massimo, Moore, Vesely, Berkowitz, Chatterjee, Coslett, Hurtig, Forman, Lee and Trojanowski2008) similarly found a longitudinal cognitive decline in a pathologically-confirmed sample of tau-negative and tau-positive cases. It is possible that non-linear cognitive trajectories, differing rates of progression, pathological heterogeneity within cohorts, and analysis across all outcomes obscured the ability to detect associations between study effect sizes and illness duration.
Though svPPA patients showed the greatest impairment across all tasks, these findings were likely driven by their language impairment. The degree of material-specific impairment was expectedly more striking in the svPPA group when compared with the nfvPPA and bvFTD groups. Indeed, svPPA patients showed greater deficits across verbally-mediated measures of IQ, ideational fluency, and memory when compared against the non-verbal or visuospatial counterparts of these measures. When the contributions of language were excluded from the analysis, the FTD groups showed similar effect sizes for cognition. These findings underscore the importance of including measures with reduced semantic burden in the neuropsychological assessment of svPPA as loss of word meaning can result in poor performance in ways that do not reflect deficits in the construct a task is intended to measure (Harciarek and Cosentino, Reference Harciarek and Cosentino2013). Comparison of verbal and non-verbal subscales within assessment batteries can prevent misinterpretation of poor scores.
Though executive deficits are present in bvFTD, they are not consistently so and many patients can produce entirely normal scores on traditional tests of executive functioning years into the illness (LaMarre and Kramer, Reference LaMarre, Kramer, Ravdin and Katzen2013). These inconsistent findings have been attributed to multiple factors, including the heterogeneity of the executive functioning construct, variable patterns of frontal degeneration, individual differences in cognitive reserve, and missing data due to a behavioral disturbance in bvFTD patients (LaMarre and Kramer, Reference LaMarre, Kramer, Ravdin and Katzen2013; Smeding and de Koning, Reference Smeding and de Koning2000). In the current study, the degree of executive deficits in bvFTD was similar in magnitude to that of processing speed, memory, olfaction, IQ, social cognition, and ideational fluency. Furthermore, the magnitude of executive impairment in bvFTD was comparable with nfvPPA. These findings prompted us to examine differences in the executive error index, a metric of all perseverative errors, intrusions, sequencing errors and rule violations. Though this error index was not the largest effect size observed in bvFTD, significantly greater deficits were seen on this metric in bvFTD relative to svPPA and nfvPPA. Kramer et al. (Reference Kramer, Jurik, Sha, Rankin, Rosen, Johnson and Miller2003) and Carey et al. (Reference Carey, Woods, Damon, Halabi, Dean, Delis, Miller and Kramer2008) similarly found greater executive errors and decreased rule monitoring in bvFTD relative to svPPA and AD patients. Hornberger et al. (Reference Hornberger, Savage, Hsieh, Mioshi, Piguet and Hodges2010) found that, across all cognitive and neuroimaging indices, higher errors on a test of inhibitory control and degree of orbitofrontal atrophy separated AD and bvFTD groups with 91% accuracy. Collectively, our findings support the use of error analysis in the differential diagnosis of bvFTD.
Extensive findings indicate that measures of olfaction and social cognition are mediated by orbitofrontal-limbic neurocircuitry (Zald and Andreotti, Reference Zald and Andreotti2010), regions affected early in bvFTD (Perry et al., Reference Perry, Graham, Williams, Rosen, Erzinçlioglu, Weiner, Miller and Hodges2006; Seeley et al., Reference Seeley, Crawford, Rascovsky, Kramer, Weiner, Miller and Gorno-Tempini2008). Our results showed large olfactory deficits that were comparable in magnitude across FTD syndromes and to effect sizes for cognitive and social cognitive domains. The small number of olfactory studies precluded our ability to compare effect sizes by task type (odor identification v. discrimination). This analysis could be interesting as odor identification tasks have greater semantic burden compared with odor discrimination and odor detection threshold tasks. In svPPA, impaired odor identification and odor-picture matching have been observed alongside intact odor discrimination (Luzzi et al., Reference Luzzi, Snowden, Neary, Coccia, Provinciali and Lambon Ralph2007). Though a similar profile emerged in bvFTD, the reduction in odor identification was milder in comparison with svPPA and was associated with executive dysfunction. Thus, examining the relative contributions of semantic and perceptual impairments to olfactory dysfunction could help differentiate FTD syndromes.
Given that social comportment changes are more prominent in bvFTD (Bora et al., Reference Bora, Walterfang and Velakoulis2015, Reference Bora, Velakoulis and Walterfang2016), we did not anticipate comparable effect sizes for social cognition across groups. However, Miller et al. (Reference Miller, Hsieh, Lah, Savage, Hodges and Piguet2012) found emotional processing deficits in svPPA even after accounting for the influence of language and perceptual impairments. Kumfor et al. (Reference Kumfor, Miller, Lah, Hsieh, Savage, Hodges and Piguet2011) found that reducing the attentional-demands of a facial emotion identification task improved performance in bvFTD and nfvPPA patients but not in svPPA. It is worth noting that bvFTD and svPPA show overlap in their clinical features – i.e. word and object agnosia are not uncommon in bvFTD and behavioral impairments are present in svPPA. Both syndromes show overlapping patterns of atrophy in the OFC, insula, and anterior cingulate (Rosen et al., Reference Rosen, Gorno-Tempini, Goldman, Perry, Schuff, Weiner, Feiwell, Kramer and Miller2002). In addition, social cognitive performance can vary by the hemispheric predominance of atrophy, with right-predominant FTD cases showing greater socio-emotional difficulties (Binney et al., Reference Binney, Henry, Babiak, Pressman, Santos-Santos, Narvid, Mandelli, Strain, Miller, Rankin, Rosen and Gorno-Tempini2016). Therefore, atrophy in shared regions could explain the shared clinical features and measurable social cognitive, behavioral and language deficits in both conditions.
The nature and extent of episodic memory difficulties in FTD have been debated for some time (for a review, see: Hornberger and Piguet, Reference Hornberger and Piguet2012). Assessment of material-specific memory differences has proved useful in distinguishing bvFTD from AD (Hodges et al., Reference Hodges, Patterson, Ward, Garrard, Bak, Perry and Gregory1999; Kramer et al., Reference Kramer, Jurik, Sha, Rankin, Rosen, Johnson and Miller2003) and nfvPPA from the logopenic variant of PPA (lvPPA; Ramanan et al., Reference Ramanan, Flanagan, Leyton, Villemagne, Rowe, Hodges and Hornberger2016). Poor visuospatial memory has emerged as a potential cognitive index of the amyloid pathology of AD and lvPPA cases. Aspects of our results validate these findings as each FTD subgroup had greater deficits for delayed recall of verbal v. visuospatial information. More recent work has focused on the use of novel spatial memory and topographical orientation tests that probe posterior cingulate regions typically affected in AD and generally spared in FTD cases – including those with significant memory loss (Bird et al., Reference Bird, Chan, Hartley, Pijnenburg, Rossor and Burgess2009; Tu et al., Reference Tu, Wong, Hodges, Irish, Piguet and Hornberger2015). Initial results in differentiating FTD patients are encouraging and warrant further exploration in larger samples.
Category-guided and letter-guided verbal fluency have been studied extensively in bvFTD and AD (Laisney et al., Reference Laisney, Matuszewski, Mézenge, Belliard, de la Sayette, Eustache and Desgranges2009; Kertesz et al., Reference Kertesz, Jesso, Harciarek, Blair and McMonagle2010), as dissociable profiles are attributed to the differential demands of frontally-mediated processes. Though both tasks impose comparable demands on executive control, category-guided tasks rely on the integrity of semantic memory and temporal structures. Our results followed the expected pattern as svPPA patients showed greater category-guided fluency impairment. Conversely, nfvPPA patients had comparable effect sizes across verbal fluency tasks. Similarly, bvFTD patients showed equivalent effect sizes across all ideational fluency tasks, a finding consistent with earlier work in a sample of pathology-confirmed cases of FTD (Rascovsky et al., Reference Rascovsky, Salmon, Hansen, Thal and Galasko2007).
Visuospatial functioning in FTD is relatively spared in comparison with AD and Lewy Body disease (Park et al., Reference Park, Harvey, Johnson and Farias2015; Pal et al., Reference Pal, Biswas, Pandit, Roy, Guin, Gangopadhyay and Senapati2016) but few studies have analyzed facets of visuospatial functioning in FTD. We found visuospatial deficits to be significantly greater in bvFTD than in the FTD-PPA groups. SvPPA and nfvPPA showed a double dissociation in their visuospatial profiles. NfvPPA patients showed greater impairment in visuoconstructional tasks than visuoperceptual tasks; the opposite pattern was observed in svPPA. These data converge with the observation of greater visuoconstruction impairment in tau-positive relative to tau-negative cases (Grossman et al., Reference Grossman, Xie, Libon, Wang, Massimo, Moore, Vesely, Berkowitz, Chatterjee, Coslett, Hurtig, Forman, Lee and Trojanowski2008).
There were several limitations of the current meta-analysis. Studies were included based on clinical diagnosis rather than pathological confirmation. Furthermore, it is likely that a subset of nfvPPA patients diagnosed prior to the implementation to contemporary diagnosis criteria (Gorno-Tempini et al., Reference Gorno-Tempini, Hillis, Weintraub, Kertesz, Mendez, Cappa, Ogar, Rohrer, Black, Boeve, Manes, Dronkers, Vandenberghe, Rascovsky, Patterson, Miller, Knopman, Hodges, Mesulam and Grossman2011) may have had lvPPA, underscoring the need for comparison of our results to findings in lvPPA samples. We could not assess the influence of race/ethnicity on effect size magnitude as this variable was rarely reported. The inclusion of racial characteristics in future studies would be helpful in assessing its influence on task performance. We strived to include the broadest sample of neuropsychological, social cognitive and olfactory outcomes available in FTD patients. This broader inclusion led to greater variation in study quality and between-study variance. To account for within- and between-study variation in effect size estimates, we employed a random-effects model and employed sensitivity analyses to identify sources of potential outliers. We found stable effects that were not significantly influenced by outliers. As such, we believe the benefits of this approach, particularly the ecological relevance, allow for a broader application of these study findings.
Overall, the results of this meta-analysis provide important information regarding the nature and severity of neuropsychological, social cognitive, and olfactory impairment in the behavioral and language variants of FTD. Given the difficulties in recruiting large samples of FTD patients, meta-analytic tools for the systematic synthesis of independent studies are useful for the quantitative characterization of phenotypes. It is hoped that the comprehensive results from this study will guide the clinical assessment of FTD patients and future studies aimed at improving FTD measurement for clinical and research applications.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291718003604.
Author ORCIDs
Vidyulata Kamath 0000-0001-7711-3597
Acknowledgements
We extend our gratitude to Martin D. Smith, M.A., Stephen Aita, Janae Cephas, Madalyn Myers, Maeve Dennigan, Erin Schnappauf, and Alana Spears for assistance with article procurement and data entry. We thank David J. Schretlen, Ph.D., and Paul J. Moberg, Ph.D., for guidance regarding task assignment and data extraction. VK is supported by the Johns Hopkins Clinical Research Scholars Program (grant number KL2TR001077). This work is supported by funding from the Johns Hopkins Department of Psychiatry and Behavioral Sciences (VK), the Jane Tanger Black Fund for Young-Onset Dementias (CUO); and the Nancy H. Hall Fund for Geriatric Psychiatry (CUO). The views and opinions expressed in this article are those of the authors and should not be construed to represent the views of the US Government. The funders had no role in study design, data analysis, manuscript preparation, or decision to publish.
Conflict of interest
None.
