Posters

INTRODUCTION: Child maltreatment (CM) and adult partner violence correlate with late-onset dementia and mid-life risk factors. Yet, their associations with age at diagnosis and risk of young-onset dementia (YOD) remain unclear.

METHODS: Among 147,938 UK Biobank participants, we identified Alzheimer's-related, vascular, and frontotemporal dementia cases. We examined associations between CM, adult partner violence, age-at-diagnosis, and YOD (diagnosed age 68 years or younger). Weibull accelerated failure time models estimated time from age 18 to diagnosis. Single-mediator and sequential causal mediation models tested CM-YOD direct effects, indirect effects of partner violence and 22 risk factors, and proportions mediated.

RESULTS: Of 915 cases, 128 were YOD. Individuals exposed to CM and partner violence received a dementia diagnosis younger than unexposed participants. CM was associated with YOD (Adjusted OR: 1.51; 95% CI: 1.01, 2.29), with partner violence, socioeconomic deprivation, alcohol excessive consumption, and BMI >30kg/m2, cumulatively mediating 78% of the association. DISCUSSION: Preventing violence and modifying mid-life risk factors may delay dementia.

Introduction: DNA methylation may mediate interactions between genetic and environmental risk factors in Alzheimer’s disease (AD). Previous AD epigenome-wide association studies (EWASs) have largely relied on 450K arrays and have not systematically examined APOE genotype. We conducted EWAS meta-analyses using the EPIC array (≈800K CpGs) to examine associations between DNA methylation, APOE genotype (ε4 risk and ε2 protective variants), and neurofibrillary tangle Braak stage in post-mortem human prefrontal cortex.

Methods: Four post-mortem brain cohorts with matched EPIC array, APOE genotype and Braak stage data were analysed. After quality control, normalisation, and confounder adjustment, EWASs were performed for APOE ε4 carriers vs non-carriers (n=1,308; excluding ε2: n=1,198) and ε2 carriers vs non-carriers (n=1,308; excluding ε4: n=731), followed by fixed- and random-effects meta-analyses. Braak stage analyses compared “controls” (Braak 0–II) with “AD cases” (V–VI) across three cohorts (n=846). Inflation was controlled using Bacon correction and surrogate variables. Differentially methylated positions (DMPs) were prioritised based on Bonferroni significance and differentially methylated regions (DMRs) were identified using comb-p.

Results: Meta-analyses identified 8 Bonferroni-significant DMPs associated with ε4 and 2 with ε2. Signals mapped primarily to the extended APOE locus (including APOC1 and TOMM40). Danalysis identified two ε4-associated regions at the APOE/APOC1 locus and one ε2-associated Dat the CAT promoter. The Braak stage meta-analysis identified 167 Bonferroni-significant DMPs, implicating established AD genes (e.g. SPI1, CSF1R, RIN3) and 121 DMRs. Conclusions: This study provides a comprehensive epigenetic meta-analysis of APOE genotype and Braak stage, revealing both APOE-proximal and broader AD-associated methylation changes.

Background: People living with dementia (PLWD) experience frequent emergency admissions, occupying up to 25% of NHS beds at any time. Improved understanding of admission patterns is needed to predict care needs and prevent avoidable admissions, improving care planning and patient outcomes.

Methods: We analysed anonymised electronic health records for n=220,010 individuals (n=31,573 PLWD) in the Discover-NOW/WSIC dataset, comprising data from 3 million individuals across eight boroughs in North West London. Emergency admissions were propensity-matched 1:1 on age, sex, deprivation, and ethnicity between PLWD and controls (CON). We quantified demographic determinants affecting admission risk, the proportion of bed days attributed to different causes of admission, and outcomes of readmission and mortality following discharge.

Results: Across 108,694 admissions in 2017-2024, PLWD experienced 965,691 bed-days (4,024 per 1000 person-years). This was 30% greater than in matched controls over the same number of admissions (745,759 bed-days; 3,108 per 1000 person-years). Infections were the primary cause for the greatest proportion of bed-days in PLWD (37.1%;27.1% CON), followed by injury (15.4% PLWD;15.0% CON). Upon discharge, 6-month mortality was greater in PLWD than matched controls (Overall HR=1.44, 95%CI=[1.39-1.49], p<0.001), and particularly high in deprived individuals (Deprivation quintile (continuous): 0.99[0.98-0.99], p<0.0001; higher quintile=less deprived).

Conclusion/Impact: PLWD experience longer emergency admissions and worse outcomes, effects compounded by socioeconomic inequalities. More than half (52%) of emergency bed-days in PLWD are due to potentially preventable infections and injury. Proactive prevention, e.g. digital monitoring and community support, may substantially reduce NHS burden.

Background: Cerebral microbleeds (CMBs) are common in Parkinson’s disease (PD) and Lewy body dementia (LBD). Ultra-high field 7T MRI improves CMB detection compared with 1.5T, but whether it provides meaningful additional sensitivity over 3T MRI, and whether this has clinical relevance, remains unclear. We compared 7T and 3T MRI for CMB detection in PD and LBD and examined associations with clinical measures.

Methods: 72 participants were included: PD without dementia (PD-NC, n=45), LBD (n=8), controls (HC, n=19). All underwent both 3T and 7T MRI. SWI was used for visual detection of CMBs, quantified using the Microbleed Anatomical Rating Scale (MARS). Sensitivity of 3T versus 7T MRI for CMB was compared, and relationship between CMB and cognitive and motor performance assessed.

Results: At 3T, 84 CMBs were detected in 17 participants (4 LBD, 9 PD-NC, 4 HC). At 7T, 109 CMBs were detected in 21 participants (3 LBD, 9 PD-NC, 8 HC), representing a 30% increase. Seven participants (2 PD-NC, 4 HC, 1 LBD) showed CMBs only at 7T. These lesions were either too small to be resolved at 3T or could not be reliably distinguished from background signal. No associations were observed between CMB burden and cognitive or motor performance at either field strength.

Conclusion: 7T MRI showed a slightly higher sensitivity in detecting CMBs compared to 3T in PD and DLB. At neither field-strength, an association was found between CMB quantity and cognitive or motor performance.

Background: Alzheimer’s Disease (AD) is the most common form of dementia and a leading contributor to global mortality. Brain volume changes are used to aid diagnosis as they are both non-invasive and widely used clinically. The rate of brain atrophy on MRI scans can be accurately quantified using Boundary Shift Integrals (BSI). When applied in longitudinal scans, they are preferable to cross-sectional scans due to their ability to provide a more individualised reference, however the need to wait for repeat scans delays prognostication. To address this problem, we explored the possibility of using supervised machine learning models to accurately predict the rate of brain atrophy in patients with AD, using BSI, from baseline T1-weighted MRI scans during first presentation.

Methods: T1-weighted baseline MRI scans from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) were used to train the following machine learning models, Random Forest, Ridge Regression and Support Vector Regression, to predict 12-month BSI outcomes. 10-fold cross validation, R2 (Coefficient of Determination) score, was used to compare the accuracy of predicting Ventricular BSI (1555 scans), Whole-brain BSI (1563 scans).

Results: Average cross-validated prediction performance for Ventricular BSI (VBSI) was significantly higher (R2 score of 0.38, 95% CI 0.36-0.41) than that for Whole-brain BSI (0.23, 95% CI 0.21-0.25).

Conclusion: We found that predictions with Ventricular BSIs were more accurate than Whole-brain BSIs, however overall model performance remains insufficient following cross-validation. Future work should consider incorporation of additional clinical, biomarker and genomic data to better capture the complexity of neurodegeneration in AD.

Background: Our previous work demonstrated that electroencephalogram (EEG) slowing predicts accelerated cognitive aging, increased neurodegenerative disease risk, and earlier all-cause mortality in the general population. Building on this, we hypothesised EEG slowing may serve as a neural marker of overall biological aging. Here we investigate whether EEG slowing predicts accelerated biological aging, measured by two second-generation epigenetic clocks (DNAm PhenoAge and GrimAge2), and more rapid development of aging-related health outcomes, including physical frailty and multimorbidity.

Methods: 10,624 adults (ages 36-96) from five population-based cohorts underwent EEG at baseline and were followed over 15-years. Theta-to-alpha ratio (TAR) was calculated using spectral analysis from 3-minutes of resting EEG. Multivariable regression models examined whether higher TAR at baseline predicted accelerated epigenetic aging, incident frailty, and faster accumulation of multiple chronic diseases (multimorbidity).

Results: At baseline, participants with higher TARs were epigenetically older, frailer, more multimorbid, and had poorer self-rated health (all p’s <0.05). Among healthy participants, higher TAR was associated with substantially increased risk of frailty (Incidence rate ratio (IRR) for high TAR vs low TAR = 2.11, p = 0.004) and accelerated accumulation of multimorbidity (IRR for high TAR vs low TAR = 1.51; p = 0.004) over follow-up.

Discussion: EEG slowing predicts accelerated whole-body aging trajectories and may serve as a scalable, non-invasive marker of biological aging in the general population.

Associative binding, which supports the integration of object and spatial information in working memory, is disproportionately vulnerable in Alzheimer’s disease (AD). Although EEG abnormalities are well documented in AD, conventional analyses provide limited insight into the rapid, large‑scale brain activity patterns that support cognition. EEG microstate analysis offers a complementary approach by segmenting the signal into brief, recurring whole‑brain activity patterns. Here, we asked whether these patterns are altered in AD at rest and during working memory binding, and whether task‑related microstates are specifically linked to misbinding. Forty people with AD and thirty age‑matched healthy controls (HC) underwent high‑density, eyes‑closed resting‑state EEG and completed an object-location working memory task with concurrent EEG. Microstate analysis was applied separately to resting and task data. For each microstate, we quantified occurence, duration, coverage, and its global explained variance (GEV). Behaviourally, AD participants showed clear working memory impairment, with a marked increase in misbinding errors relative to HC. Resting‑state analysis revealed four microstates. Compared with HC, AD showed more frequent but shorter episodes and reduced time in one microstate, together with longer duration of another, indicating a reorganisation of cortical activity. During the working memory task, four task‑specific microstates were identified; AD participants showed reduced time in a task‑related microstate. Within the AD group, lower GEV of this microstate was associated with greater misbinding, independent of identification accuracy. These findings link altered large‑scale brain dynamics to a core cognitive deficit in AD and support EEG microstates as a cognition‑anchored biomarker candidate.

Background: Dementia with Lewy Bodies (DLB) is the second most common neurodegenerative disorder in adults aged over 65 years. The diagnostic accuracy for DLB is low due to disease heterogeneity and overlap with other dementias. This requires diagnostic tools that are non-invasive, accessible, and cost-effective. We develop ML-based classifiers to differentiate DLB from healthy controls (HC) and Alzheimer’s disease (AD) using features extracted from structural MRI in a large, multi-centre dataset.

Methods: We included T1-weighted MRI scans of 201 DLB, 120 AD and 244 age- and sex-matched HC from the European DLB consortium (E-DLB), National Alzheimer’s Coordinating Center (NACC), Neuroimaging of Inflammation in Memory and Related Other Disorders (NIMROD), Multimodal Imaging in Lewy Body Disorders (MILOS) and AMyloid imaging for Phenotyping LEwy body dementia (AMPLE) cohorts. Structural MRI features, extracted using Freesurfer and fractalbrain toolkit, were corrected for age and sex, rescaled, and split into 80:20 training and test datasets. To differentiate DLB from HC and AD, we employed Logistic regression (LR), Random Forest (RF) and Support Vector Machine (SVM) with Lasso regularization for feature selection. Results: For DLB vs HC classifiers, we achieved the highest accuracy of 83% using RF, followed by 82% with LR, and 79% with SVM. For DLB vs AD classifiers, we achieved the highest accuracy of 77% using LR, followed by 75% using RF, and 74% with SVM.

Conclusion: ML-based models using T1 MRI features show reasonable performance in diagnosing DLB. Incorporating multimodal MRI features and deep learning may improve these classifiers for clinical implementation.

Background: Low resting heart-rate (HR) is associated with increased cerebral pulsatility (PI), which is associated with cerebral small vessel disease (cSVD). The OXVASC study showed that cSVD severity, vascular cognitive impairment (VCI), and recurrent stroke were strongly associated with PI reflecting excess energy transmitted through stiffened great vessels. We investigated whether population-level HR->vascular dementia (VaD) associations support this haemodynamic hypothesis.

Methods: UK-Biobank is a large, community-based, longitudinal cohort recruiting >500,000 UK participants since 2007, with face-to-face assessment and healthcare data-linkage. Associations between baseline HR and VaD were determined by incidence rate and Cox proportional hazards regression with adjustment for age, sex, metabolic and cardiovascular risk factors. Secondary analyses examined HR×diastolic blood pressure (DBP) interactions.

Results: In 467,444 participants (mean age 56.5 years, 54% female) without baseline dementia during 6.2 million person-years(Pyrs) (median 13.6 years), 2,080 VaD events occurred. VaD was increased at both low (<60bpm: HR 1.16, 95%CI 1.03-1.31) and high HR (>90bpm: HR 1.93, 95%CI 1.63-2.28). After adjustment, high HR retained significant VaD risk (HR 1.37, 95%CI 1.09-1.74, p=0.008). However, VaD risk was attenuated in low HR (HR 1.12, 95%CI 0.96-1.31, p=0.149). VaD incidence was highest when high HR was combined with low DBP (113.8 per 100,000Pyrs), consistent with PI mediated injury.

Conclusions: Resting heart rate shows a J-shaped association with vascular dementia, independent of cardiovascular risk factors. The HR×DBP interaction supports a haemodynamic mechanism involving cerebral pulsatility. These findings suggest HR may be a modifiable target for vascular dementia prevention.