Behavior determines times to care home admission and death in semantic dementia: Insights from a cohort of 90 cases Authors: Marc Melkonian, William Lucas, Catherine Merck, Anne Salmon, Francis Eustache, Mickaël Laisney, Serge Belliard Published: 2026 ABSTRACT AbstractINTRODUCTIONIn semantic dementia (SD), no consistent prognostic factors for early care home (CH) admission or mortality have been identified, despite their potential utility in care management.METHODSOver a 31‐year period, we collected sociodemographic, neuroimaging, neuropsychological, and behavioral data within a large cohort of 90 SD patients. Survival and risk factor analyses were conducted to identify variables mostly influencing CH admission and/or death.RESULTSHalf of the patients were admitted to CH 7.5 years after diagnosis, and half died after 8.4 years. Behavior emerged as the sole baseline prognostic factor and even mild behavioral disorders were linked to earlier CH admission (3.0 years) and shorter survival (3.6 years), without influence of the predominant side of anterior temporal atrophy.DISCUSSIONBehavioral disorders are a major risk factor for loss of independence in SD and a usable prognostic marker in clinical practice, regardless of the predominant side of temporal atrophy. BACKGROUND Semantic dementia (SD) is a neurodegenerative disease, belonging to the group of frontotemporal lobar dementias (FTLDs), and characterized by a progressive and prominent loss of semantic knowledge. 1 , 2 , 3 , 4 Neuroimaging typically reveals an asymmetric atrophy of the anterior temporal lobes (ATLs), with the predominant side determining the initial clinical presentation. Patients’ early complaints vary depending on the affected hemisphere: left ATL (L‐ATL) atrophy is typically associated with verbal semantic impairments, while right ATL (R‐ATL) atrophy is linked to non‐verbal and socioemotional semantic deficits. Behavioral disorders may also emerge relatively early and become almost inevitable in later disease stages. 5 , 6 Some authors proposed to separate these two forms into two distinct entities. The former is considered as a primary linguistic disorder, known as the “semantic variant of primary progressive aphasia” (svPPA). 7 Since behavioral disorders are more pronounced in cases of right anterior temporal lobe damage, a novel entity has been proposed for the latter under various names: “semantic behavioral variant of frontotemporal dementia,” 8 “right temporal variant of frontotemporal dementia,” 9 and “frontotemporal dementia with right anterior temporal predominance”. 10 Epidemiology and prognosis SD is the rarest FTLD variant, with an estimated incidence of 0.11 per 100,000 person‐years and a prevalence of 1.16 per 100,000 persons for svPPA in a recent meta‐analysis. 11 Due to this relative rareness, epidemiological and prognostic data on SD remain sparse. Most studies focus on FTLD syndromes as a whole, and often include small samples of SD or svPPA patients. A notable exception is the retrospective longitudinal study by Hodges et al., which followed 100 SD cases over 17 years. 12 These studies consistently highlighted a relative young age at diagnosis in SD, around 65 years, 12 , 13 , 14 as well as a longer symptom duration before diagnosis compared to other FTLD syndromes, ranging from 4 to 6.5 years. 12 , 14 , 15 , 16 , 17 , 18 , 19 Clinically, two‐thirds of SD patients exhibit a L‐ATL predominant atrophy and an initial verbal semantic complaint. 12 Survival in SD has been repeatedly reported as the longest among FTLD syndromes, 20 , 21 , 22 , 23 with median survival times ranging from 9 to 12.8 years. 12 , 13 , 14 , 15 To date, no consistent determinants of loss of independence have been identified in SD. The meta‐analysis conducted by El‐Wahsh et al. 24 revealed that baseline demographic variables (i.e., sex, age at diagnosis, education level), symptom duration, and neuropsychological scores—particularly Mini‐Mental State Examination (MMSE) or Addenbrooke's Cognitive Examination (ACE) scores, used as measures of dementia severity—did not emerge as prognostic factors in FTLD syndromes. These findings have been mirrored by similar recent results. 15 , 25 Additionally, most studies have not identified any prognostic influence of the predominant side of ATL atrophy on survival. 24 Study objectives Like other neurodegenerative diseases, SD impacts patients’ independence and places a heavy burden on their caregivers. 26 Identifying prognostic factors for loss of independence could provide valuable guidance for clinical practice, optimizing patient support, and alleviating caregiver burden. The objectives of this study were two‐fold: (1) to describe the survival course of SD by focusing on two endpoints, that is, care home (CH) admission and death; and (2) to identify baseline prognostic factors for these events. RESEARCH IN CONTEXT Systematic review: The authors reviewed the literature in PubMed and found few studies specifically addressing prognosis in semantic dementia (SD). Most of the available data pertains to small subgroups within larger cohorts of various dementias. One notable exception is the study by Hodges et al. published in 2010, which involved 100 patients with SD. However, it did not identify significant prognostic factors for care home admission and survival. Interpretation: Our analyses identified behavioral disorders as a key prognostic factor at baseline, significantly influencing both CH admission rates and mortality in SD. Future directions: Further studies are needed to validate these findings and explore which behavioral components most strongly impact dependence, as well as identifying other potential prognostic factors, not assessed in our study. This approach, combined with addressing limitations like comorbidity assessment, standardized neuropsychological evaluation over time, and refined behavioral phenotyping may improve prognostic accuracy. Systematic review: The authors reviewed the literature in PubMed and found few studies specifically addressing prognosis in semantic dementia (SD). Most of the available data pertains to small subgroups within larger cohorts of various dementias. One notable exception is the study by Hodges et al. published in 2010, which involved 100 patients with SD. However, it did not identify significant prognostic factors for care home admission and survival. Systematic review: The authors reviewed the literature in PubMed and found few studies specifically addressing prognosis in semantic dementia (SD). Most of the available data pertains to small subgroups within larger cohorts of various dementias. One notable exception is the study by Hodges et al. published in 2010, which involved 100 patients with SD. However, it did not identify significant prognostic factors for care home admission and survival. Interpretation: Our analyses identified behavioral disorders as a key prognostic factor at baseline, significantly influencing both CH admission rates and mortality in SD. Interpretation: Our analyses identified behavioral disorders as a key prognostic factor at baseline, significantly influencing both CH admission rates and mortality in SD. Future directions: Further studies are needed to validate these findings and explore which behavioral components most strongly impact dependence, as well as identifying other potential prognostic factors, not assessed in our study. This approach, combined with addressing limitations like comorbidity assessment, standardized neuropsychological evaluation over time, and refined behavioral phenotyping may improve prognostic accuracy. Future directions: Further studies are needed to validate these findings and explore which behavioral components most strongly impact dependence, as well as identifying other potential prognostic factors, not assessed in our study. This approach, combined with addressing limitations like comorbidity assessment, standardized neuropsychological evaluation over time, and refined behavioral phenotyping may improve prognostic accuracy. METHODS This retrospective cohort study with prospective follow‐up adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines. Patients On January 1, 2019, the medical records of 112 patients diagnosed with SD between 1992 and 2018 were extracted from the SD cohort of the Memory Resource and Research Center (CMRR) at the University Hospital of Rennes, France. Patients were enrolled by experienced cognitive neurologists based at the center or its surrounding regions in French Brittany. All diagnoses were confirmed by the center's experts, establishing the date of diagnosis. Patients diagnosed before 1998 met criteria for early and prominent semantic deficits, consistent with the cases described by Snowden et al. 2 and Hodges et al. 3 All patients were retrospectively evaluated according to the diagnostic criteria of Neary et al., 27 also applied to patients diagnosed after 1998. Only patients for whom SD was the primary condition and who presented no other severe medical comorbidities were included. Psychiatric disorders, drug or alcohol misuse, signs of concomitant neurodegenerative diseases, history of traumatic brain injury or stroke, and severe chronic cardiovascular, respiratory, hepatic, or renal diseases were exclusion criteria. To be included, patients had to have undergone a neuropsychological assessment by the center's experienced neuropsychologists within 6 months of diagnosis. They also had to have undergone a brain magnetic resonance imaging (MRI) with T1‐weighted or fluid‐attenuated inversion recovery (FLAIR) coronal images of the ATLs within 6 months before or after diagnosis. Medical records of included patients were then screened to extract demographic and neuropsychological data, as well as behavior‐related information. Twenty‐two patients were excluded due to insufficient data, resulting in the final inclusion of 90 patients. The follow‐up period was extended to January 1, 2024, in order to gain additional insight into the progression of the most recently diagnosed patients. During follow‐up, patients were typically seen annually until loss of independence, and their functional status was regularly updated. When patients were no longer followed at the CMRR, active efforts were made to collect dates of CH admission and/or death, mainly by telephone. Demography Demographic data collected included age, sex, and years of education. Symptom onset was determined as the earliest date of either patient's first complaints or changes observed by relatives consistent with a diagnosis of SD. Neuroimaging Brain 1.5T MRI scans were available for 85 patients. Two experienced neurologists (S.B., A.S.), blind to the neuropsychological and behavioral data, independently assessed ATL atrophy using T1‐ or T2‐weighted coronal views with a four‐point visual rating scale, inspired by the scales proposed by Galton et al. 28 and Davies et al. 29 . Our scale was defined as follows: 0 indicated no visible atrophy, 1 mild widening of temporal sulci without loss of gyral volume, 2 marked widening of temporal sulci with loss of the gyral volume, 3 severe atrophy. In cases of disagreement between raters, the mean of the two scores was used. Two additional radiological scores were calculated: the total ATL atrophy score, obtained by summing the individual atrophy scores of both ATLs, and an asymmetry index, derived by subtracting the atrophy score of the R‐ATL from that of the L‐ATL. Patients with a positive asymmetry index were classified into the L‐ATL‐predominant atrophy (LAPA) group, while those with a negative index were classified into the R‐ATL‐predominant atrophy (RAPA) group. Patients with symmetric atrophy were not assigned to a group. The five patients without available MRI scans for visual rating were classified based on the predominant side of temporal atrophy reported in radiologists’ notes, without atrophy scores. Neuropsychological assessment At baseline, each patient underwent a comprehensive neuropsychological battery. Global cognitive functioning was assessed using the MMSE 30 (n = 62), as well as the PM47 test 31 for non‐verbal efficiency. Working memory was evaluated using Forward and Backward Digit Span tests. Verbal presemantic abilities were assessed with the single‐word repetition task from the French version of the Boston Diagnostic Aphasia Examination (BDAE) 32 and reading of regular words of the Croisile's battery. 33 Non‐verbal presemantic abilities were evaluated using the Benton Facial Recognition Test, 34 the Rey–Osterrieth Complex Figure Test, 35 and the Embedded Figure Test of the Protocole d’Évaluation des Gnosies Visuelles. 36 General semantic knowledge about objects was assessed through confrontation oral naming tasks (from the BECS‐GRECO battery 37 or the DO80 38 ), and verbal and visual semantic matching subtests of the BECS‐GRECO battery. Prior to the BECS‐GRECO validation in 2011, the semantic sorting test 39 was used; this test involves categorizing pictures and names of items under four categories (animals, food, tools, and kitchenware) at both superordinate and subordinate levels. Person‐specific knowledge (i.e., the ability to recognize and identify famous people) was assessed using the French Rennes Celebrities Questionnaire. 40 Briefly, this questionnaire involves recognizing 10 famous people mixed with ten unknown people. For each correctly recognized celebrity, participants have to answer questions about some of their characteristics (dead or alive, nationality, occupation, and some physical attributes) and their biography. The questionnaire has two versions for the same celebrities: a name‐based and a photograph‐based. Due to the longitudinal nature of the cohort, some tests were replaced over time. To enable inter‐patient and inter‐group comparisons, normalized composite scores (overall percentages of correct responses) were derived when multiple tests assessed the same cognitive function. For each composite score, a composite cutoff was calculated as the mean of the rescaled tests’ cutoffs. These first three composite scores are: (1) the “object semantic knowledge” score, corresponding to the total score obtained on the semantic sorting test or the semantic matching task of the BECS‐GRECO battery (cutoff < 95%); (2) the “verbal semantics” score, derived from word version of the semantic sorting test or the semantic matching task of the BECS‐GRECO battery (cutoff < 95%); and (3) the “non‐verbal semantics” score, based on image versions of the same tests (cutoff < 95%). The three other composite scores, one semantic and two presemantic scores, correspond to situations where one or more tests were used to assess a specific cognitive domain: (4) the “people recognition” score, integrating performance in recognizing names and faces of both famous and unknown people (cutoff < 97%); (5) the “verbal presemantics” score, based on performance in reading regular words and single‐word repeating tasks (cutoff < 93%); (6) the “non‐verbal presemantics” score, incorporating results from the Benton Facial Recognition Test, the Rey–Osterrieth Complex Figure test, and the Embedded Figure Test (cutoff < 77%). Behavioral assessment All available behavior‐related information was gathered from medical and paramedical reports by a neurologist. Behavior was systematically assessed at each visit by asking patients and their caregivers for any change in behavior and/or personality trait. These changes were also scrutinized by clinicians during examination and written down if present. Behavioral changes were independently rated by two other clinicians—one neurologist and one neuropsychologist—blind to the neuropsychological and radiological data, on a five‐point Likert scale (“behavioral score,” BS), as follows: 0, no changes in personality or behavior; 1, very minor behavioral or personality changes (e.g., tendency to self‐centeredness, less flexibility regarding schedules) reported by relatives but unnoticeable during examination, without causing disturbances in daily life nor distress on family members; 2, mild disturbances (e.g., marked rigidity regarding schedules, pronounced self‐centeredness) observed by clinicians, and with mild impact on daily life reported by caregivers; 3, moderate and recurrent behavioral disorders causing significant interpersonal disturbances, including social misbehavior (e.g., behavioral disinhibition, shoplifting, repeated traffic violations); 4, severe disorders requiring institutionalization or medication (e.g., self‐/hetero‐aggression, self‐endangerment). Intermediate scores were allowed when caregivers reported the most severe disturbances as occasional and were not observed by healthcare professionals. The behavioral score was established by rating the most disabling and/or the most prominent disorder, among all reported disorders. In case of disagreement between raters, the final behavioral score was obtained by averaging the two ratings. Cerebrospinal fluid biomarkers Cerebrospinal fluid (CSF) was collected in a subset of 33 patients, diagnosed in the most recent years of the cohort. The analyses included a screening for Alzheimer's disease (AD) biomarkers: total tau (t‐Tau), phosphorylated tau at threonine 181 (p‐Tau181), amyloid‐beta peptides Aβ1‐40 and Aβ1‐42 levels, as well as the Aβ1‐42/Aβ1‐40 ratio. These markers were considered indicative of AD pathology when positive, according to the pathological thresholds established by the manufacturers and the international consensus criteria in use at the time of the analyses. Data analyses All analyses were performed using R Statistical Software (v4.5.2). 41 Survival analyses were conducted with the survival (v3.8‐3) package, 42 and graphics were generated using the ggplot2 (v3.5.2) 43 and the survminer (v0.5.1.999) 44 packages. Statistical analyses Binary and ordinal variables were compared using Fisher's exact test. Prior to comparing means, the normality of data was assessed with the Shapiro–Wilk test; means were then compared using either Welch's t‐test or the Wilcoxon rank‐sum test with continuity correction, as appropriate. Correlations among variables were calculated using Kendall's tau‐b (τ), with a statistical significance threshold set at p < 0.05. Times to 50% CH admission and mortality, from symptom onset and first visit, were estimated using the Kaplan–Meier method, and group‐level curves were compared using the log‐rank test. Given the absence of consistently identified prognostic factors in previous studies, an exploratory approach was undertaken. Demographic data, neuropsychological composite scores, and neuroimaging atrophy scores were individually examined as potential prognostic factors in univariate Cox proportional hazards (PH) regression models. The proportional hazards assumption was tested using Schoenfeld residuals (Grambsch–Therneau test). 45 Variables that reached statistical significance in univariate analyses were considered to enter multivariable Cox PH regression models. To limit the discovery of spurious significant correlations and to increase statistical power, the number of predictors included in multivariable models was limited to ensure at least ten events per variable. 46 Each significant variable was subsequently evaluated in Cox PH regression models, with sex, age, and symptom duration at first visit as controlling covariates. Collinearity between covariates was assessed using variance inflation factors (VIF), applying a threshold of VIF < 5 to exclude moderate to high multicollinearity. RESULTS Baseline cohort characteristics Our cohort comprised 50 men and 40 women (Table 1). At diagnosis, the mean age was 65.2 years (range: 47.6–83.5), with mean symptom duration of 3.8 ± 2.6 years. More than half of patients were diagnosed within their sixth decade (n = 48), mostly between 65 and 70 years (n = 26) (see Figure S1 for age distribution). Mean MMSE score was 24.0 ± 4.6. Presemantics composite scores were within the normal range, while all semantic composite scores were altered, with less than 85% correct answers (Table 1, see Table S1 for raw neuropsychological scores). TABLE 1Demographic data, MMSE, behavioral, and ATL atrophy scores at baseline.Baseline dataVariablesOverall, n = 90LAPA, n = 68RAPA, n = 21LAPA vs. RAPA (p‐values)DemographyGender M/F, n 50/4038/3010/11/261 a Education, y 11.1 (2.8) [n = 77]11.4 (2.8) [n = 57]10.2 (2.6) [n = 20] 0.032 c Age at diagnosis, y 65.2 (7.2) [n = 90]64.6 (7.7) [n = 68]67.3 (5.2) [n = 21]0.07 b Age at onset, y 61.5 (7.6) [n = 89]61.1 (8.1) [n = 67]63.3 (5.4) [n = 21]0.16 c Symptoms duration at diagnosis, y 3.8 (2.6) [n = 89]3.7 (2.5) [n = 67]4.0 (2.6) [n = 21]0.60 c Neuropsychological testsMMSE24.0 (4.6) [n = 62]23.1 (4.8) [n = 45]26.6 (2.9) [n = 16] 0.010 c PM4727.5 (5.9) [n = 58]28.8 (5.5) [n = 41]24.4 (5.7) [n = 17] 0.009 c People recognition CS79.0 (16.2) [n = 58]81.0 (16.3) [n = 43]73.0 (14.7) [n = 15]0.10 c Object knowledge CS78.6 (13.1) [n = 71]77.9 (13.4) [n = 53]80.5 (12.3) [n = 18]0.45 b Verbal semantics CS74.3 (15.4) [n = 73]72.9 (15.9) [n = 55]78.4 (13.6) [n = 18]0.17 c Non‐verbal semantics CS84.3 (13.5) [n = 72]84.5 (14.0) [n = 54]83.7 (12.1) [n = 18]0.39 c Verbal presemantics CS96.8 (5.2) [n = 72]96.1 (5.8) [n = 55]98.9 (1.6) [n = 16]0.16 c Non‐verbal presemantics CS89.3 (5.8) [n = 84]89.9 (5.8) [n = 63]87.6 (5.7) [n = 20]0.11 c BehaviorBehavioral score1.5 (1.0) [n = 88]1.4 (1.1) [n = 66]1.8 (0.8) [n = 21]0.35 c Neuroimaging atrophy scoresLeft ATL2.0 (0.7) [n = 85]2.2 (0.6) [n = 64]1.5 (0.8) [n = 20] < 0.001 c Right ATL1.4 (1.0) [n = 85]1.0 (0.7) [n = 64]2.7 (0.5) [n = 20] < 0.001 ‡ Asymmetry Index d 1.2 (0.6) [n = 85]1.2 (0.6) [n = 64]1.2 (0.5) [n = 20]0.65 c Global3.4 (1.3) [n = 85]3.2 (1.2) [n = 64]4.2 (1.2) [n = 20] 0.002 c Note: Results are shown as means with their standard deviation and the number of patients with available data.Abbreviations: ATL, anterior temporal lobe; CS, composite score; LAPA, left anterior temporal lobe‐predominant atrophy; MMSE, Mini‐Mental State Examination; RAPA, right anterior temporal lobe‐predominant atrophy. a Fischer's exact test. b Welch's t‐test. c Wilcoxon's rank‐sum test with continuity correction. d Absolute values.John Wiley & Sons, Ltd. Demographic data, MMSE, behavioral, and ATL atrophy scores at baseline. Note: Results are shown as means with their standard deviation and the number of patients with available data. Abbreviations: ATL, anterior temporal lobe; CS, composite score; LAPA, left anterior temporal lobe‐predominant atrophy; MMSE, Mini‐Mental State Examination; RAPA, right anterior temporal lobe‐predominant atrophy. Fischer's exact test. Welch's t‐test. Wilcoxon's rank‐sum test with continuity correction. Absolute values. Behavioral assessment showed strong agreement between raters (weighted Cohen's kappa = 0.85; range: 0.78–0.91). Seventy‐eight percent of patients displayed behavioral changes and 30% had noticeable disorders during examination, mildly impacting daily life (BS = 2) (Table 2). The cohort's mean BS was 1.5 ± 1.0. Only one patient, with symmetric ATL atrophy, had a BS exceeding 3 (BS = 4). TABLE 2,Distribution of severity of behavioral disorders at baseline.Behavioral scoreOverall, n = 87RAPA, n = 21LAPA, n = 66 p‐value a 019 (22%)2 (9.5%)17 (26%)0.110.56 (6.9%)0 (0%)6 (9.1%)14 (4.6%)1 (4.8%)3 (4.5%)1.513 (15%)7 (33%)6 (9.1%)226 (30%)6 (29%)20 (30%)2.59 (10%)3 (14%)6 (9.1%)310 (11%)2 (9.5%)8 (12%) Note: Results are presented as n (%).Abbreviations: LAPA, left anterior temporal lobe‐predominant atrophy; RAPA, right anterior temporal lobe‐predominant atrophy. a Fisher's exact test.John Wiley & Sons, Ltd. Distribution of severity of behavioral disorders at baseline. Note: Results are presented as n (%). Abbreviations: LAPA, left anterior temporal lobe‐predominant atrophy; RAPA, right anterior temporal lobe‐predominant atrophy. Fisher's exact test. Neuroimaging revealed LAPA in 76% of cases (n = 68), RAPA in 23% (n = 21), and symmetric atrophy in approximately 1% (n = 1). Comparison of LAPA and RAPA groups at baseline At first visit, groups with LAPA and RAPA predominant atrophy did not significantly differ in terms of age at diagnosis, gender, or symptom duration (Table 1). Patients in the LAPA group were slightly more educated (11.4 vs. 10.2 years; p = 0.032). This group had a lower mean MMSE score (23.1 vs. 26.6; p = .010), while the RAPA group performed worse on the PM47 task (28.8 vs. 24.4; p = .009). The LAPA group showed lower composite scores in verbal semantics and object knowledge, whereas the RAPA group had lower scores in people recognition and non‐verbal semantics. None of these differences, however, reached statistical significance (Table 1). Famous people face recognition was significantly impaired in the RAPA group compared to the LAPA group (93.3% vs. 55.8% impaired; p < 0.001), driving its lower performance in people recognition but with no difference in name recognition or specific knowledge regardless of input modality about celebrities. Both groups performed within the normal range on the Embedded Figure Test, with similar impairment rates (∼5%). Interestingly, behavioral disturbance severity did not significantly differ between groups (Table 2). Neuroimaging revealed bilateral ATL atrophy in the two groups, with similar absolute values of asymmetry indices, but the RAPA group exhibited greater total ATL atrophy (4.2 vs. 3.2; p = .002). Correlations between neuropsychological and behavioral data with ATL atrophy Most neuropsychological deficits did not significantly differ by predominant atrophy laterality, but several scores moderately correlated with the degree of ATL atrophy (Table S2). The object semantic knowledge composite score correlated both with left (τ = −0.28; p = 0.002) and right (τ = −0.20; p = 0.023) ATL atrophy. Left ATL atrophy correlated with MMSE (τ = −0.33; p = 0.001) and verbal semantics (τ = −0.27; p = 0.002) scores, while right ATL atrophy correlated with PM47 (τ = −0.38; p < 0.001), people recognition (τ = −0.30; p = 0.002), non‐verbal semantics (τ = −0.25; p = 0.005), and non‐verbal presemantics (τ = −0.16; p = 0.047) scores. Behavioral scores correlated more strongly with right ATL (τ = 0.24; p = 0.005) than left (τ = 0.18; p = 0.040) ATL atrophy. Survival analysis Patients were followed for a mean of 7.4 years (range, 0.5–20.4). During follow‐up, 49 patients (54.4%) were admitted to a CH, with similar rates between LAPA (n = 35, 51.5%) and RAPA (n = 13, 62.0%) groups (Fisher's exact test, p = 0.43). Median time to CH admission was 5.8 years (interquartile range [IQR], 3.2–7.6) from diagnosis and 9.8 years (IQR, 6.4–12.0) from symptom onset. Sixty‐seven patients died after a median of 8.0 years (IQR, 5.8–11.0) from diagnosis, and 11.0 years (IQR, 8.8–16.0) from symptom onset. Kaplan–Meier method estimated a 50% CH admission rate at 7.5 years (IQR, 6.6–9.1) and a 50% survival rate at 8.4 years (IQR, 7.2–10.7) after diagnosis (Figure 1). These rates were respectively reached at 11.6 years and 11.9 years from symptom onset (Figure S2). Kaplan–Meier curves illustrating time‐to‐event outcomes for care home admission (top row) and survival (bottom row) at the overall cohort level and within the two behavioral subgroups. Risk factors for earlier CH admission In univariate Cox models, baseline BS emerged as the sole risk factor for earlier CH admission (hazard ratio [HR] = 1.80; 95% confidence interval [CI], 1.32–2.45; p < 0.001; Table S3). However, in the corresponding model, the PH assumption appeared slightly violated (p = 0.020). Subsequently, a modified Cox model including a time‐varying coefficient was fitted, showing a good overall fit (likelihood ratio test = 19.96; p < 0.001), with a good fit of the estimated HR function of time (p = 0.026). This model estimated a baseline HR of 3.82, exponentially decreasing over time to null risk (HR = 1) at 9.6 years after diagnosis. BS remained significant after adjusting for sex, age, and symptom duration (adjusted HR [AHR] = 1.92; 95% CI, 1.38–2.68; corrected p < 0.001). In this model, the estimated multicollinearity was very low (all VIFs = 1.1). However, a slight deviation of the PH assumption was still present when estimating the HR of the BS (p = 0.032). As a consequence, the regression coefficient of the BS was also modeled as a linear function of time, resulting in a well‐fitted model (likelihood ratio test = 20.96; p < 0.001). The modified model indicated an AHR of 4.13 at baseline, exponentially slowly decreasing over time (p = 0.038), reaching null risk at 10.4 years after diagnosis. The plots of the coefficients estimated from these models over Schoenfeld residuals are shown is Figure S3. BS stayed a significant risk factor for earlier CH admission when adjusted with controlling variables. In addition, none of the multivariable models incorporating only controlling variables, with and without the MMSE score, yielded a significant HR. We next investigated which BS stratification provided the most robust prognostic discrimination (Table S4). Patients with a score greater than 1 exhibited significantly earlier CH admission, with higher scores corresponding to progressively earlier institutionalization. While reduced survival was also observed in these patients, further stratification using higher score thresholds did not reveal additional significant differences. To maximize prognostic clarity, we categorized patients into two subgroups: those with no or minimal behavioral disturbances (BS ≤ 1) and those with mild‐to‐severe behavioral changes (BS > 1). Kaplan–Meier analysis revealed that patients in the BS > 1 subgroup experienced a significantly shorter median time to CH admission—6.3 years compared to 9.3 years in the BS ≤ 1 subgroup (log‐rank test, p = 0.007) (Figure 1). This stratification underscores the clinical relevance of even subtle behavioral disturbances at baseline in predicting accelerated loss of independence. Risk factors for shorter survival Older age at diagnosis (HR = 1.04; 95% CI, 1.00–1.08; p = 0.042) and higher BS (HR = 1.40; 95% CI, 1.11–1.75; p = 0.004) were associated with shorter survival. No variable regression coefficient showed deviation from the PH assumption in the corresponding models. In multivariable models, only BS remained significant (AHR = 1.42; 95% CI, 1.12–1.79; p = 0.014). Kaplan–Meier curves showed that patients with BS > 1 had a 50% survival rate 3.6 years earlier than those with BS ≤1 (7.7 vs. 11.3 years; log‐rank test, p < 0.001) (Figure 1). ATL atrophy scores were not significant in univariate models. The RAPA group showed a non‐significant trend toward earlier loss of independence (CH admission: 6.3 vs. 8.2 years, p = 0.06; survival: 7.7 vs. 9.5 years, p = 0.37; Figure S4). AD biomarkers Patients with positive (n = 5) and negative (n = 28) AD CSF biomarkers showed comparable times to CH admission (p = 0.70) and survival (p = 0.68). The HR for biomarker positivity was not significant (HR = 0.96; 95% CI, 0.33–2.84), although the small sample size limits interpretation. DISCUSSION Behavioral disorders as a pivotal prognostic indicator in SD This study, drawing from one of the most extensive and longest longitudinal cohorts of patients with SD, establishes that baseline behavioral disorders represent the foremost predictor of accelerated CH admission and shortened survival. This is the first demonstration of such a robust prognostic link in SD, highlighting the imperative for routine behavioral evaluation in clinical settings. Behavior disturbances have already been highlighted as a common predictive factor of CH admission across FTLD syndromes 23 , 47 and are associated with reduced dependence‐free survival in the behavioral variant of FTD (bvFTD). 47 However, their prognostic significance in SD had not been previously elucidated. In dementia syndromes, the severity of behavioral disorders correlates with increased caregiver burden, 26 , 48 , 49 often hastening institutionalization. Caregivers of patients with svPPA and bvFTD experience levels of strain comparable to those caring for individuals with bvFTD, with a more rapid worsening observed in svPPA caregivers over time. 26 Our findings corroborate this pattern: The BS employed in this study indirectly measured caregiver burden by evaluating the impact of behavioral changes on daily life and family member distress. Importantly, no neuropsychological measure significantly influenced the timing of CH admission or mortality, reinforcing the notion that global cognitive scores—frequently used as proxies for dementia severity—may not reliably forecast clinical outcomes. Cohort characteristics and external validity The demographic and clinical profile of our cohort aligns closely with prior epidemiological research. The majority of patients received their diagnosis in their sixth decade and nearly a quarter were diagnosed after the age of 70, consistent with earlier reports. 12 , 13 , 14 , 15 The mean symptom duration at the first visit was 3.8 years, comparable to durations documented in previous studies. 11 , 12 , 13 , 15 , 16 , 17 As observed in other cohorts, two‐thirds of our patients exhibited left anterior temporal lobe‐predominant atrophy. Common demographic variables, including sex, age at diagnosis, and educational attainment, did not emerge as significant prognostic factors. Notably, symptom duration did not predict survival, a finding that contrasts with prior work suggesting that shorter symptom duration at baseline is associated with poorer survival. 13 LAPA versus RAPA: Minimal differences with broad implications A central finding of this study is the absence of substantial differences between patients with LAPA and RAPA in terms of demographic characteristics, neuropsychological performance, behavioral profiles, or clinical outcomes. Baseline neuropsychological assessments revealed significant differences solely in general cognitive functioning, non‐verbal reasoning, and famous people face recognition. The lower scores on the MMSE in the left atrophy group is likely to pertain to the test's emphasis on verbal abilities. Conversely, the greater executive dysfunction observed in the right atrophy group—particularly in non‐verbal reasoning tasks—may be attributable to neurodegeneration extending beyond the right ATL into the frontal cortex. Previous studies have indeed reported more extensive frontotemporal atrophy in right‐ compared to left‐predominant atrophy. 50 , 51 , 52 Impairment in famous people face recognition was nearly universal among patients with right‐predominant atrophy, aligning with the non‐verbal semantic deficits typically associated with right ATL neurodegeneration. However, more than half of patients with left‐predominant atrophy also performed poorly on this task, with no significant differences between groups in composite scores. This overlap undermines the notion of a clear cognitive–behavioral divide based on atrophy laterality. Behavioral disturbances were equally prevalent in both groups, though their correlation with right ATL atrophy was stronger. These results suggest that the side of predominant atrophy may not independently determine cognitive or behavioral profiles at diagnosis. SD: A clinically coherent entity, beyond predominant side of atrophy Our findings contribute to the ongoing debate regarding the classification of SD into two distinct entities: A verbal semantic variant (svPPA) and a behavioral semantic variant (sbvFTD). 53 svPPA is characterized by impairments in verbal semantics and left anterior temporal lobe atrophy, while the behavioral semantic variant—proposed for cases with right anterior temporal lobe‐predominant atrophy and serving as the counterpart of svPPA—is defined by deficits in person identification, behavioral changes, and loss of empathy. 8 However, our data reveal minimal differences between left‐ and right‐predominant atrophy groups, suggesting that the cognitive‐behavioral profiles may not reliably predict the predominant side of atrophy at diagnosis. The clinical usability of a clear‐cut distinction of two entities may thus be of limited relevance. High rates of baseline behavioral disorders have been consistently reported in SD and svPPA, ranging from 50% to 80%. 12 , 14 , 16 , 18 , 54 Given the higher prevalence of behavioral disturbances in patients with RAPA, substantial evidence from multiple studies underscores the critical involvement of right hemisphere structures in the pathogenesis of these behavioral changes. 8 , 55 To account for the elevated prevalence of behavioral disturbances in svPPA, Ulugut et al. proposed that a subset of these patients might actually correspond to right temporal lobe variants of frontotemporal dementia. 54 However, consistent with findings from multiple independent studies, our cohort demonstrated no significant differences in the baseline frequency of behavioral disorders between patients with LAPA versus RAPA. 12 , 25 , 56 Supporting this observation, Hodges et al. reported equivalent Neuropsychiatric Inventory‐Evaluation Scale (NPI‐ES) scores across both subgroups of SD, 12 while Kumfor et al. documented comparable baseline rates of social dysfunction, encompassing motivational deficits and stereotyped behaviors. 56 Notably, Kumfor et al. further provided neuroanatomical evidence linking behavioral regulation to the right hemisphere, demonstrating significant correlations between cortical thickness in right‐lateralized brain regions and the severity of behavioral disturbances. However, these associations did not withstand correction for multiple comparisons, highlighting the need for further investigation into the lateralization of behavioral symptoms in SD. Our results are consistent with these findings. While we did identify significant correlations with the degree of atrophy in each ATL, we did not observe clear dichotomies in neuropsychological and behavioral data based on the predominant side of atrophy. This suggests that the value of atrophy predominance may be limited, as it does not independently predict cognitive and behavioral profiles in clinical practice. Study limitations and future directions While this study provides critical insights into the prognostic significance of behavioral disorders in SD, some limitations must be acknowledged. Some of them are inherent to long‐term longitudinal investigations. First, some variables were not assessed using standardized scales or tests. Severe comorbidities were excluded but not systematically evaluated. A meta‐analysis indicated that a higher comorbidity burden, as measured by the Charlson Comorbidity Index, predicts greater mortality, though data specific to FTLD syndromes are lacking. 57 Additionally, precise measures of functional status at baseline were unavailable. Previous studies have linked worse functional status, as reflected by lower scores on the Disability Assessment for Dementia, 58 to faster functional decline 59 and poorer survival. 13 Furthermore, behavioral assessments were not conducted using scales commonly employed in research, such as the Frontal Behavioral Inventory 60 or the Neuropsychiatric Inventory. 61 Many of these tools were either unavailable or not translated into French during the early years of the cohort. While these scales offer detailed quantification of behavioral aspects, such as motivation, impulsivity, and social norms knowledge, their absence in our study limited fine‐grained phenotyping. Nevertheless, our behavioral evaluations were conducted by clinicians familiar with the distinctive behavioral changes in SD, including self‐centeredness, rigid thinking, parsimony, and food fads. Although our behavioral scale lacks detailed phenotyping, it offers practical utility in clinical settings. Neuroimaging analyses relied on visual ratings of ATL atrophy rather than automated quantitative methods, such as volumetry. Quantitative approaches would have enabled more precise measurement of gray matter atrophy in the ATLs and other brain regions involved in behavior, such as the frontal lobes. Numerous studies have linked specific behavioral disorders to degeneration in various brain areas and networks within FTLD syndromes, extending beyond the ATLs. 62 Our decision to use a visual rating system was driven by the goal of identifying practical clinical markers, following methods developed by other researchers 28 and commonly applied in AD. 63 Finally, our study did not include neuropathological confirmation of diagnoses. For SD diagnosis, we required a documented history of early and predominant semantic deficits at the first visit. Such clinical profiles are highly specific to the distinctive transactive response DNA binding protein of 43 kDa of type C (TDP43‐C) proteinopathy that underlies the majority of SD cases. 64 Further studies are needed to confirm our clinical findings while addressing the limitations inherent to long‐term investigations. Future research could explore the prognostic value of additional neuroimaging or biological markers, such as neurofilament levels, which are associated with axonal neurodegeneration and poorer outcomes in FTLD syndromes. 21 More refined phenotyping of behavioral disturbances may also help identify which behavioral components most strongly influence functional independence, precipitate CH admission, and contribute to reduced survival in SD. CONFLICT OF INTEREST STATEMENT The authors declare no financial, personal, or professional conflicts of interest related to this work. Author disclosures are available in the Supporting Information. CONSENT STATEMENT Written informed consent was obtained from all participants or their legally authorized representative prior to enrollment. All procedures were conducted in accordance with the Declaration of Helsinki and approved by the local ethics committee. Supporting information Supporting information: alz71859‐sup‐0001‐FigureS1.PDF Supporting information: alz71859‐sup‐0002‐FigureS2.PDF Supporting information: alz71859‐sup‐0003‐FigureS3.PDF Supporting information: alz71859‐sup‐0004‐FigureS4.PDF Supporting information: alz71859‐sup‐0005‐TableS1.docx Supporting information: alz71859‐sup‐0006‐TableS2.docx Supporting information: alz71859‐sup‐0007‐TableS3.docx Supporting information: alz71859‐sup‐0008‐TableS4.docx Supporting information: alz71859‐sup‐0009‐ICMJEDisclosureForms.pdf