Objectives: The association between alcohol and cognitive function is complex. Several pathways are involved. In this work we focus on the potential associations between drinking frequency and cerebrospinal fluid (CSF) biomarkers of neurodegenerationin a cohort of dementia-free middle- aged-adults.

Methods: This analysis is part of the ALBION study. Each volunteer went throughan extensive neuropsychological assessment. Sociodemographic characteristics were alsorecorded.Dietary intake, including energy and alcohol intakes, was assessed through four 24-hour recalls. Diet quality was evaluated through the adherence to the Mediterranean diet (MedDietScore). Other lifestyle parameters were recorded, such as sleep duration, which using a wrist actigraph for 7 days, and physical activity, assessed (through the International Physical Activity Questionnaire). CSF sample was collected. Multiple logistic regression analyses were conducted using drinking frequency subgroups (abstainers, less than 2 standard drinks/week, above 2 standard drinks/week) as independent variables and CSF Alzheimer Disease (AD) biomarkers [amyloid-β (Aβ) accumulation, Tau/Aβ and Phospho-tau/Aβ ratios] as dependent variables.

Results: Of the 195 individuals without dementia, the majority were female (66%), with an average age of 65±9.4 years and 13.8±3.6 years of education. The average sleep duration, as assessed by wrist actigraph, was 5.91±4.48 hours. A weekly alcohol intake exceeding 2 standard drinks was associated with higher Aβ positivity compared to alcohol abstinence [OR:2.98 (1.29- 6.90)], even after adjusting for confounders (age, sex, education years, energy intake and diet quality).

Conclusions: The present study showed that light-to-moderate alcohol consumption was associated with increased Aβ accumulation in cognitively intact adults. These findings potentially increase our understanding of how alcohol consumption might influence vulnerability to AD.

Objectives: The pathophysiological mechanism linking sleep apnea and cognitive impairment remains poorly understood. We aimed to explore the association between the Apnea- Hypopnea Index (AHI), cognitive state, and cerebrospinal fluid biomarker pTau181/aβ42 (Amyloid status) in non-demented individuals.

Methods: In this cross-sectional study, 115 non-demented participants from the ALBION cohort underwent a one week actigraphy assessment and a one-night WatchPAT evaluation to determine sleep microarchitecture and AHI, respectively.Total sleep time (TST) was calculated as the median of one week assessments of daily TST through actigraphy.Participants were categorized by both cognitive status (patients with Mild Cognitive Impairment (MCI) or cognitively normal individuals (CN)) and Amyloid status ((A+) or (A-).Binary logistic regression analysis, adjusted for age, sex, years of education, BMI and TST was used to assess the presence of an association with AHI, respectively. Participants were further divided by AHI levels (Low < 15 or High ≥ 15) and regression analysis evaluated its interaction with amyloid status, using cognitive status as the outcome.

Results: For each unit increase in AHI, the odds of being classified as MCI were 7% higher (OR=1.07 ,p=0.003) and the odds of being classified as A+ were 4% higher (OR=1.04, p=0.043).Compared to the reference group [A(-)/AHI(low)], the odds ratio for MCI was 3.84 (p=0.100) in A(+)/AHI(low) group, 4.48, p=0.022 in the A(-)/AHI(high) and 15.30(p=0.0017) in the A(+)/AHI(high) group.

Conclusions: We find significant associations between AHI, MCI and amyloid-beta brain pathology. Although more longitudinal studies are needed, higher AHI levels could potentially contribute to cognitive impairment either by interacting with amyloid-beta or independently.

Wake After Sleep Onset (WASO) — the amount of time spent awake after initially falling asleep — is a key marker of sleep fragmentation. Elevated WASO has been previously associated with elevated risk of Alzheimer's disease (AD). Less is known about its association with cognitive decline in non-demented population. In the current study, we opted to investigate the association between WASO, measured via actigraphy, and cognitive performance in dementia-free adults (>40 years).

Data were gathered form the Aiginition Longitudinal Biomarker Investigation Of Neurodegeneration (ALBION). A total of 183 people (65% women, mean age 64.9, SD 9.6 years) without dementia were included in the current study. Participants completed actigraphy monitoring (Philips Actiwatch) for 7 consecutive nights to quantify sleep parameters, including WASO. Cognitive function was assessed using a detailed standardized neuropsychological battery. CSF samples were analyzed for amyloid-β and tau to classify participants based on the ATN criteria. Regression models evaluated the relationship between a) mean WASO and Mild Cognitive Impairment diagnosis (MCI) b) mean WASO and cognitive decline, adjusting for age, sex, and education.

WASO was not associated with MCI diagnosis or participants’ cognitive performance. However, we used ATN criteria and divided our sample in a) people without AD pathology, b) with amyloid pathology and c) with AD biomarkers (amyloid and tau pathology). We found that in the group of people without AD pathology, higher WASO was significantly associated with poorer performance in memory (β = -0.016, p=0.003) and attention/speed (β = -0.009, p=0.06) after adjusting for covariates.

According to the study results, greater sleep fragmentation was associated with worse memory and attention performance in older individuals without evidence of AD pathology. These findings highlight sleep disruption as a potential independent risk factor for cognitive decline and suggest that better sleep continuity may help preserve cognitive health.

Aims:This study investigates whether baseline actigraphy features can predict subsequent cognitive decline using wearable-derived measures as early risk markers.

Methods: We analyzed actigraphy data from 101 participants in the ALBION dataset. Cognitive performance was assessed using neuropsychological tests across five domains and summarized as z-scores, standardized against a cognitively normal reference group and adjusted for age and sex. Cognitive trajectories were defined using the average composite z- scores between baseline (T0) and follow-up (T1), where T1-T0 ≤ 2 years. Cognitive status was classified as "stable" if the composite z-score remained unchanged or improved, and "declined" if decreased (Δzscore < 0), resulting in 53 stable and 48 declined samples. We extracted seventeen actigraphy features, including multi-cosinor and nonparametric parameters, from one-day averaged actigraphy time series at baseline. To address the limited sample size, we employed repeated nested cross-validation (rnCV) with 10 rounds × 5 outer × 5 inner folds for unbiased performance estimation. Feature selection employed Minimum Redundancy Maximum Relevance (mRMR) to identify informative, non-redundant predictors.

Results: We used rnCV to compare baseline models, including Logistic Regression, Linear Discriminant Analysis, Gaussian Naive Bayes, and ensemble tree estimators such as Random Forest, Light radient Boosting Machine, and eXtreme Gradient Boosting, confirming that both model types achieve good performance. LDA emerged as the best estimator without feature selection, reaching median recall 0.70 ± 0.03, specificity 0.70 ± 0.04 and MCC 0.39 ± 0.05. Notably, LightGBM achieved the same median recall and specificity with MCC 0.40 ± 0.05, using only three mRMR-selected features.

Conclusions: Actigraphy-based ML shows promise for predicting cognitive decline, with performance achievable using minimal features, supporting further investigation.