The protocol has been described in detail elsewhere . A total of 193 participants aged 40–59, of whom 168 had repeat MRI were recruited through multiple sources. Initially they were identified from the dementia register database held at a London National Health Service (NHS) Trust, part of the UK National Health Service. Other participants were recruited via the Join Dementia Research website (https://www.joindementiaresearch.nihr.ac.uk/), through information about the study on the Internet and public presentations. The study aimed to recruit approximately half of the subjects with a parental family history of dementia and half without. All MRI scans obtained on participants were formally reported by an experienced consultant neuroradiologist. Any subject with significant incidental findings (e.g., tumour, stroke) was excluded. Approval for the study was given by the NHS Research Ethics Committee London Camberwell St-Giles (REC reference: 12/LO/1023). All participants provided informed written consent.
Participants underwent multimodal 3 T structural magnetic resonance imaging on a single scanner (Siemens Verio) including volumetric T1-weighted scans (176 slices, 1.0 × 1.0 mm, 1.0 mm slice thickness, repetition time = 2300 ms, echo time = 2.98 ms, flip angle 9°). Scans were repeated after approximately 2 years on the same scanner using the same protocol. Imaging findings from these participants have been previously reported [9,10,11].
Percentage brain and ventricular volume change between the two scans were determined using the FSL SIENA program (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki). Values were then divided by the time between scans to give rates per year. SPM12 (https://www.fil.ion.ucl.ac.uk/spm) was used to segment the brain into grey matter, white matter and CSF, and spatially normalise the images to standard space. To determine hippocampal volume, the hippocampal region of interest (ROI) from the AAL template  was used to determine the mean of the normalised grey matter segmentation image. Left and right were averaged, and the longitudinal change was determined by subtracting the baseline measurement from repeat. Total intracranial volume (ICV) was calculated from the sum of grey matter, white matter and CSF segmentation at baseline. All output was manually checked to ensure correct brain segmentation, and alignment of baseline/repeat scans.
Demographic data and information on drinking and smoking status were collected from case report forms which hold participants’ data collected at interview. We also included data from the participants’ medical history, which included history of alcoholism and diabetes. Participants were classified as current smokers or non-smokers (never and former).
We used the Lifetime of Experiences Questionnaire (LEQ)  to extract data about physical, social and cultural activity (art, music, hobbies) from the age of 30 years to present. The three physical questions used from the LEQ asked participants how often they engaged in activities that were (1) mildly energetic (such as walking), (2) moderately energetic and (3) vigorous (such as running). For each activity, participants were asked to indicate their response on the six-point Likert scale from Never to Daily. We dichotomised the scores as to whether people did the activities more vs. less frequently than once a month.
Alcohol consumption in units per week was determined using the Scottish Collaborative Group Food Frequency Questionnaire version 7.1 (https://www.foodfrequency.org) completed by the participant at baseline and repeat visit. The questionnaire asks for the typical amount and frequency of consumption of 175 food and drink items over the last 2–3 months. We took the average of baseline and repeat alcohol consumption, except for eight participants for whom repeat values were missing, in which case we used the baseline value. Following previous research  and the UK guidelines of 14 units per week as a safe limit (https://www.drinkaware.co.uk/alcohol-facts/alcoholic-drinks-units/latest-uk-alcohol-unit-guidance/), alcohol usage was divided into abstaining (< 1 unit/week), light (> = 1 and < 7 units/week), moderate (> = 7 and < 14 units/week), high (> = 14 and < 21 units/week) and very high (> = 21 units per week). No distinction was applied between men and women in this classification. For a subgroup analysis on male and female separately, we divided alcohol use into three groups: (0–5), (> 5–15), (> 15) units per week to avoid having too few male subjects in each group.
The Framingham risk score was used to calculate 10-year cardiovascular risk from current smoking status, systolic BP, BP medication status, sex, age at baseline, total and HDL cholesterol and diagnosis of diabetes. Social class was determined using the UK National Statistics Socio-economic classification (NS-SEC).
Brain volume data and clinical variables were analysed with R version 3.5.2. Independent t tests were used to compare groups for continuous variables. Linear regression was used to investigate association of alcohol with brain atrophy rates. Total brain and ventricular volume change were expressed as percentage of baseline volume (as calculated by SIENA). For hippocampal analysis, we included ICV as a covariate to adjust for head size. Residuals were checked by eye to verify normality, heteroscedasticity and absence of nonlinear associations.