Materials and methods

Participants

Secondary data analysis was conducted on participant data from the research clinic at FRONTIER, the frontotemporal dementia clinical research group at the Brain and Mind Centre of the University of Sydney. Data from all eligible patients presenting between August 2008 and June 2021 were included in the study, resulting in 106 probable bvFTD, 43 possible bvFTD, and 42 SD patients. In addition, 112 healthy controls were included, sourced from FRONTIER’s research volunteer pool. Clinical diagnosis was established by consensus following a comprehensive examination involving clinical interview by an experienced neurologist, a cognitive assessment, and review of neuroimaging, in line with diagnostic criteria for bvFTD [1] and SD [2]. Specifically, to distinguish between possible and probable bvFTD, neuroimaging evidence from 3 T MRI structural brain imaging was closely examined. For all participants classified as possible bvFTD, there was no frontal and/or anterior temporal atrophy sufficient to provide neuroimaging support for a probable diagnosis. In addition, for the 24 possible bvFTD participants who subsequently returned for at least one follow up visit, additional neuroimaging continued to support a diagnosis of possible bvFTD (i.e. no participants converted from possible to probable bvFTD within this sample). Information from previous neuropsychological assessments or clinical imaging provided by the referring clinician was also considered within the diagnostic evaluation. While not used as a primary indicator of SD, we acknowledge that where no previous cognitive testing was available, data from the Sydney Language Battery (see measures below) was part of the information available to the neurologist who confirmed the diagnosis. Healthy controls underwent the same neuropsychological assessments and neuroimaging procedures.

Exclusion criteria included the presence of any additional suspected psychiatric, neurodegenerative or neurological disorder (e.g., bipolar disorder, motor neuron disease), very severe disease stage or limited general cognitive ability: <40/100 on the Addenbrooke’s Cognitive Examination-III [ACE-III][28]; n = 7), no formal years of education in English or an inability to provide answers in English (n = 9), substantial missing data (> 50% of the language or neuropsychological variables; n = 13), and for healthy controls, a score below 88/100 on the ACE-III [29]. For SD patients, an additional exclusion criterion was applied for those participants whose neuroimaging indicated right-sided asymmetric anterior temporal lobe atrophy. This was due to both insufficient numbers to analyse this subgroup separately, as well as lack of diagnostic clarity around this clinical phenotype (also known as right temporal variant of FTD [30]).

All participants or their person responsible (e.g., spouse, family member etc.) provided written informed consent in accordance with the Declaration of Helsinki. Original data collection was approved by the South Sydney Area Health Service, the University of New South Wales and the University of Sydney Human Research Ethics Committees (HREC 2020/224), with this retrospective study approved by University of Newcastle Human Research Ethics Committee (H-2021-0399).

Assessment measures

Disease severity was evaluated using the informant-based Frontotemporal Dementia Rating Scale (FRS [31]; a measure of functional ability and behaviour). General cognitive ability was assessed using the ACE-III [29] or, if the assessment pre-dated this version, the ACE-Revised [32], with scores converted to ACE-III using a validated algorithm [28]).

To assess language, the Sydney Language Battery (SYDBAT [19]) was used to test picture naming, word comprehension (word-picture matching), semantic association (picture-picture matching), and word repetition. Each subtest is based on the same 30 polysyllabic target words, generating a subtest total score of 30, where higher scores denote better performance. For the semantic association task, this involves matching a picture of each target word with a picture of an associated item. Across the four tasks, each item was scored as either correct or incorrect. The SYDBAT has been validated in dementia populations, with high convergent validity and reliability [18, 19]. In addition, naming and comprehension items of the ACE-III were combined to form a “ACE-III Semantic” measure out of 16.

Participants also completed neuropsychological tests of executive function. The available measures from the existing dataset included: Digit Span subtest of the Wechsler Adult Intelligence Scale – Third edition (WAIS-III [33]); Trail Making Test – Part A and B (TMT [34]); plus two verbally mediated tasks—Hayling Sentence Completion Test [35], and letter fluency using the letters ‘F’, ‘A’ and ‘S’ [36]. From these tests, four key executive functioning measures were extracted. While we acknowledge that no test provides a pure measure of executive function, the maximum span achieved from the Digit Span backwards subtest (DSB) was calculated as an indicator of working memory. Total time taken to complete TMT-B (capped at 180 s) was used as a measure of speeded mental flexibility, in the context of visual, attention and motor demands. Word generativity was measured using the total correct words generated, and response initiation and suppression was measured by the overall scaled score of the Hayling. The overall scaled score was selected because it integrates performance across response initiation, response suppression, and error components thereby capturing executive control more broadly.

Given the task demands of the Hayling, which requires semantic knowledge for meaningful completion, SD patients were not administered this task.

Statistical analyses

Analyses were conducted using IBM SPSS Statistics, version 29.0 (SPSS Inc., Chicago, Ill., USA). Normality of variables was assessed using Shapiro-Wilks tests and visual inspection of histograms. Demographic variables (age, education) were analysed via one-way Analysis of Variance (ANOVA), with Games-Howell post-hoc testing to control for multiple comparisons, given unequal group sample sizes. Categorical measures (e.g., sex) were analysed using chi-squared tests. Group differences on disease severity, general cognition, language and executive function, were all compared using Kruskal–Wallis tests. Dunn’s procedure was applied to correct for multiple comparisons.

Given the non-normal distribution of SYDBAT scores, the frequency of impairment on each SYDBAT subtest was determined by identifying the cut off score for each subtest that signalled performance below the fifth percentile of the control-group distribution. The number and proportion of participants in each patient group meeting this criterion were then calculated.

To permit comparison of relative performance across SYDBAT subtests with different control distributions, raw scores were standardised using the control-group mean and standard deviation for each subtest. The resulting z scores indicated each participant’s performance relative to controls and were compared across subtests using paired Wilcoxon signed-rank tests. Given the non-normality of some score distributions, these analyses were interpreted as comparisons of relative standardised performance rather than as precise estimates of normally distributed impairment severity. Comparisons were made between naming and comprehension, naming and semantic association, naming and repetition, comprehension and semantic association, comprehension and repetition, and semantic association and repetition.

The relative and independent statistical contribution of language and executive functioning to SYDBAT performance in bvFTD was then examined using standard multiple regression. This method was chosen given its robustness in the presence of non-normal data [37]. Assumptions relating to independence, homoscedasticity, and multicollinearity were all met. For each model, ACE-III Semantic was entered as the measure of semantic functioning, while DSB, FAS, TMT-B, and Hayling Overall were entered as executive-function measures. All variables were entered simultaneously using the Enter method. Analyses were conducted separately for the possible and probable bvFTD groups. R-squared and beta coefficients were reported as indices of effect size.