confidential: for review only - bmj€¦ · confidential: for review only 5 introduction alcohol...

Confidential: For Review O

nly

Moderate alcohol consumption as a risk factor for adverse

brain outcomes and cognitive decline

Journal: BMJ

Manuscript ID BMJ.2016.034088.R2

Article Type: Research

BMJ Journal: BMJ

Date Submitted by the Author: 05-Dec-2016

Complete List of Authors: Topiwala, Anya; University of Oxford, Psychiatry Allan, Charlotte; University of Oxford, Psychiatry Valkanova, Vyara; University of Oxford, Psychiatry Zsoldos, Eniko; University of Oxford, Psychiatry Filippini, Nicola; University of Oxford, Psychiatry

Sexton, Claire; University of Oxford, Psychiatry Mahmood, Abda; London School of Hygiene and Tropical Medicine, Epidemiology and Population Health Fooks, Peggy; University of Oxford, Psychiatry Singh-Manoux, Archana; INSERM, Centre for Research in Epidemiology and Population Health; University College London, Department of Epidemiology and Public Health Mackay, Clare; University of Oxford, Psychiatry Kivimaki, Mika; University College London, Department of Epidemiology and Public Health Ebmeier, Klaus; University of Oxford, Psychiatry

Keywords: Alcohol, MRI, Brain structure, Hippocampal atrophy, Cognition

https://mc.manuscriptcentral.com/bmj

BMJ


nlyWhat this paper adds box

1. What is know on this subject

In December 2015 we searched PubMed, MEDLINE and Google Scholar for relevant

reports with the terms: “alcohol” “moderate drinking” “moderate drinking” “cognitive

impairment” “dementia” “brain” “grey matter” “white matter” “diffusion tensor imaging”

and “magnetic resonance imaging” with no language or date restrictions. Reference lists

of relevant papers were also searched.

One previous systematic review (without meta-analysis) has been published on the

alcohol dependence and neuroimaging. Studies have consistently shown heavy drinking

to be associated with Korsakoff’s syndrome, dementia and widespread brain atrophy.

Whilst smaller amounts of alcohol have been linked to protection against cognitive

impairment, there is a paucity of studies examining the effects of moderate alcohol on

the brain. Those that have been done have methodological limitations especially

regarding the lack of prospective alcohol data, been conflicting and failed to provide a

convincing neural correlate.

2. What this adds

We have shown that moderate alcohol is associated with increased risk of adverse brain

outcomes and steeper cognitive decline in the domain of lexical fluency. The area

particularly vulnerable is the hippocampus, which is the most robust imaging marker of

Alzheimer’s disease, and has not been previously linked negatively with moderate

alcohol use. No protective effect was found for small amounts of alcohol over abstinence.

Previous reports claiming a protective effect of light drinking may have been subject to

confounding by associations between increased alcohol and higher social class or IQ.

These findings support the recent reduction in the UK “safe” drinking cut-offs and

question current US guidelines. Moderate drinking may represent a modifiable risk

factor for later cognitive impairment.

Page 1 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

1

Moderate alcohol consumption as a risk factor for

adverse brain outcomes and cognitive decline:

longitudinal cohort study

Anya Topiwala* – Clinical Lecturer in Old Age Psychiatry, Department of Psychiatry,

University of Oxford, Warneford Hospital, Oxford, UK, OX3 7JX

[email protected], Tel: +44 (0)1865 226469 Fax: + 44 (0)1865 793101

Charlotte L. Allan – Academic Clinical Lecturer in Old Age Psychiatry, Neurobiology of

Aging Group, Department of Psychiatry, University of Oxford, Warneford Hospital, Oxford, UK, OX3 7JX

Vyara Valkanova – Specialist registrar in Old Age Psychiatry, Neurobiology of Aging

Group, Department of Psychiatry, University of Oxford, Warneford Hospital, Oxford, UK,

OX3 7JX

Enikő Zsoldos – DPhil student, Neurobiology of Aging Group, Department of Psychiatry, University of Oxford, Warneford Hospital, Oxford, UK, OX3 7JX

Nicola Filippini – Postdoctoral scientist, Neurobiology of Aging Group, Department of

Psychiatry, University of Oxford, Warneford Hospital, Oxford, UK, OX3 7JX

Claire Sexton – Postdoctoral scientist, FMRIB Centre, Nuffield Department of Clinical Neurosciences, University of Oxford.

Abda Mahmood – Research assistant, Neurobiology of Aging Group, Department of

Psychiatry, University of Oxford, Warneford Hospital, Oxford, UK, OX3 7JX

Peggy Fooks – Medical student, Neurobiology of Aging Group, University of Oxford,

Warneford Hospital, Oxford, UK, OX3 7JX

Archana Singh-Manoux – Professor of Epidemiology and Public Health, Department of Epidemiology and Public Health, University College London

Clare E. Mackay – Associate Professor, Translational Neuroimaging Group, Department

of Psychiatry, University of Oxford, Warneford Hospital, Oxford, UK, OX3 7JX

Mika Kivimäki – Professor, Department of Epidemiology and Public Health, University

College London

Klaus P. Ebmeier – Professor of Old Age Psychiatry, Neurobiology of Aging Group,

University of Oxford, Warneford Hospital, Oxford, UK, OX3 7JX

* Corresponding author

Article word count: 4932

Abstract word count: 281 Number of references: 78

Number of tables: 5 Number of Figures: 5

Supplementary materials: methods and results

Page 2 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

2

Abstract

Objectives

To investigate whether moderate alcohol consumption has favourable, adverse or no

association with brain structure and function.

Design

Observational cohort study with weekly alcohol intake and cognitive performance

measured repeatedly during a 30-year period (1985 to 2015). Multimodal Magnetic

Resonance Imaging (MRI) was performed at study endpoint (2012-5).

Setting

Community-dwelling adults enrolled in the Whitehall II cohort based in the United

Kingdom (the Whitehall II imaging sub-study).

Participants

We included 550 men and women aged 43.0 years (±5.4) at study baseline, none scoring

“alcohol dependent” in the CAGE screening questionnaire, and safe for brain MRI at

follow-up. Exclusions (n=23) were due to incomplete or poor quality imaging data or

gross structural abnormality (e.g. a brain cyst), incomplete alcohol use,

sociodemographic, health or cognitive data.

Page 3 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

3

Main outcome measures

Structural brain measures included hippocampal atrophy, grey matter density and

white matter microstructure. Functional measures included cognitive decline over the

study.

Results

More alcohol consumption over the 30-year follow-up was associated with increased

odds of hippocampal atrophy in a dose dependent fashion. Whilst those consuming over

30 units weekly were at the highest risk (odds ratio 5.8 95% CI: 1.8-18.6 p=<0.001),

even those drinking moderately (14-21 units per week) had three times the odds of

right-sided hippocampal atrophy compared with abstainers (odds ratio 3.4 95% CI: 1.4-

8.1 p=0.007). We found no protective effect of light drinking (1- <7 units weekly) over

abstainers. Higher alcohol use was also associated with differences in corpus callosum

microstructure and faster decline in lexical fluency.

Conclusions

Alcohol consumption, even at moderate levels, is associated with adverse brain

outcomes including hippocampal atrophy. These results support the recent reduction in

UK alcohol guidance and question the current USA safe limits.

Funding

Study funding: “Lifelong Health and Wellbeing” Program Grant: “Predicting MRI abnormalities with longitudinal data of the Whitehall II Substudy” (UK Medical Research

Council: G1001354; PI: KPE), the Gordon Edward Small's Charitable Trust (SC008962;

Page 4 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

4

PI: KPE), and the HDH Wills 1965 Charitable Trust (Charity Number: 1117747; PI: KPE). MK was supported by the Medical Research Council (K013351) and NordForsk.

Page 5 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

5

Introduction

Alcohol use is widespread and increasing across the developed world.1-3 It has

historically been viewed as harmless in moderation,4 defined variably from 9-18 units

(72-144g) per week.5,6 Recent evidence of associations with cancer risk 7 has prompted

revision of UK government alcohol guidance, although US Federal Dietary guidelines

(2015-2020) allow up to 24.5 units weekly for men.8 Even light drinking (midpoint

<12.5g daily/8 units per week) has been associated with increased risk of

oropharnygeal, oesophageal and breast cancer.7,9 Whilst chronic dependent drinking is

associated with Korsakoff syndrome and alcoholic dementia,10 the long-term effects of

non-dependent alcohol consumption on the brain are poorly understood. Robust

evidence of adverse associations would have vital public health implications.

Some authors have suggested an inverted U-shaped relationship between alcohol use

and brain outcomes, similar to cardiovascular disease. Light-to-moderate drinking has

been associated with lower dementia risk,11,12 a reduced incidence of myocardial

infarction13 and stroke.14 However, brain-imaging studies have thus far failed to provide

a convincing neural correlate that could underpin any protective effect. Research into

the effects of moderate alcohol on the brain are inconsistent.15 Moderate alcohol

consumption in older subjects has been associated with reduced total brain volume,16

increased ventricle size,17 grey matter atrophy11 and reduced frontal and parietal grey

matter density,18,19 but others have not found such relationships,20 or only at higher

consumption levels.21 Associations between moderate alcohol consumption and white

matter findings are also inconsistent. De Bruin22 reported increased white matter

volume in moderate drinkers compared with abstainers, whereas Anstey23 found the

inverse relationship. Similarly, whereas increased white matter hyperintensities have

Page 6 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

6

been described in moderate drinkers compared with abstainers24 others found no

association.17,23,25

Unresolved questions persist because of design limits to existing studies of non-

dependent drinking and brain imaging. Alcohol consumption cannot be randomised.

Most studies to date have been cross-sectional or with very limited prospectively

gathered alcohol data. People typically underestimate their alcohol intake,26 a problem

likely to be worse in a retrospective study. Studies have also included elderly

participants in whom sub-threshold pre-symptomatic cognitive impairment may

already have an impact on drinking patterns.

In this study we use data on alcohol consumption gathered prospectively over 30 years

to investigate associations with brain structural and functional outcomes in 550 non-

alcohol dependent subjects. Our hypotheses were two-fold:

1. Light drinking (<7 units weekly) is protective against adverse brain outcomes

and cognitive decline.

2. Heavier drinking (above recommended guidelines) is associated with adverse

brain and cognitive outcomes.

Images ©Alcohol Concern 2016

Box 1

Methods

Study design and participants

Five hundred and fifty subjects were randomly selected for the current Whitehall II

imaging sub-study (2012-2015) from the Whitehall II cohort study.27 The Whitehall II

Page 7 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

7

study was established in 1985 at University College London, with the aim of

investigating the relationship between socioeconomic status, stress and cardiovascular

health. It recruited 10,308 non-industrial civil servants across a range of employment

grades.27 Sociodemographic, health and lifestyle variables (including alcohol use) were

measured over a follow-up period of approximately thirty years, at approximately five-

year intervals (Phase 1: 1985-8, Phase 3: 1991-3, Phase 5: 1997-9, Phase 7: 2003-4,

Phase 9: 2007-9, Phase 11: 2011-12). In order to make the sample as representative as

possible of the cohort at baseline, a random list of 1380 participants was drawn from

those who participated in the Whitehall II Phase 11 clinical examination or Phase 10

pilot examination and had consented. Subjects were sampled from high, intermediate

and low socioeconomic groups.

Alcohol variables collected in each phase included: units drunk per week, frequency of

drinking per week over the previous year, as well as the CAGE screening

questionnaire.28 Weekly consumption was used in this analysis, as there is less

likelihood of a ceiling effect in comparison to drinking frequency. Average alcohol use

across the study was calculated as mean consumption per week averaged across all

study phases. Subjects were deemed ‘abstinent’ if they consumed less than 1 unit of

alcohol per week. ‘Light’ drinking was defined as between 1 and <7 units per week and

‘moderate’ drinking as 7 to <14 units per week for women and 7- <21 units for men,

based on use in the existing literature and government guidelines (see box 1). “Unsafe

drinking” was defined according to pre-2016 (21 units (168 g) per week for men and 14

units (112 g) for women) and newly revised UK Department of Health guidelines (>14

units (112 g) for men and women), and further categorised (14-20, 21-30, >30 units

weekly) for the purposes of the logistic regression analysis (see statistical analysis).29

Non-dependent drinkers were defined as those scoring less than 2 on the CAGE

questionnaire.

Page 8 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

8

Age, sex, education, smoking, social activity e.g. attendance at clubs and visits with

family/friends, physical activity, voluntary work, component measures of the

Framingham Stroke Risk Score e.g. blood pressure, smoking, history of cardiovascular

events, cardiovascular medications, were assessed by self-report questionnaire. Social

class was determined according to occupation at Phase 3 (highest class=1, lowest=4).

Medications (number of psychotropics reported to be taking), and lifetime history of

Major Depressive Disorder (assessed by Structured Clinical Interview for DSM IV) were

assessed at the time of the scan. Information about personality traits was determined by

questionnaire at Phase 1 and included trait impulsivity (question: “Are you hot-

headed?”).

Cognitive function was assessed at Phases 3, 5, 7, 9, and again at the time of the MRI

using short-term recall (20 words), lexical (how many words beginning with a letter can

be generated in one minute) and semantic (how many words in a category can be

named in one minute) fluency tests. Full-scale intelligence quotient (IQ) was estimated

using the Test of Premorbid Functioning - UK Version (TOPF UK, assessed at the time of

the scan) with sex and education adjustment.

Participants were included in the imaging sub-study if they were safe to have MRI and

able to give informed consent. Exclusions were due to incomplete or poor quality

imaging data or gross structural abnormality (e.g. a brain cyst), incomplete alcohol use

data, sociodemographic, health or cognitive data (figure 1).

Page 9 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

9

MRI analysis

All MRI scans were acquired at the University of Oxford Functional Magnetic Resonance

Imaging of the Brain (FMRIB) Centre, using a 3 Tesla Siemens Verio scanner (2012-15).

T1-weighted and diffusion tensor (DTI) 3T MRI sequences were used for these

analyses.30

For full technical details, see supplementary materials. In brief, relationships between

alcohol use and grey matter were examined initially using voxel-based morphometry, an

objective method to compare grey matter density between individuals in each voxel

(smallest distinguishable image volume) of the structural image. Hippocampal volumes

(adjusted for total intracranial volume) were additionally extracted, using an automated

segmentation/registration tool, for each subject for subsequent analyses. Automated

segmentation of the amygdalae was less reliable in this sample so extracted volumes

were not used in this analysis. Hippocampal atrophy was defined independently

according to visual rating (Scheltens score)31 by three clinicians, who reached a

consensus.

Diffusion tensor images indicate the directional preference of water diffusion in neural

tissue and allow inferences about the structural integrity of white matter tracts. In

healthy myelinated fibres diffusion is restricted perpendicular to the longitudinal axis of

the fibre, i.e. it is anisotropic. Voxelwise statistical analysis of diffusion tensor data

(fractional anisotropy (FA), axial diffusivity (AD), radial diffusivity (RD) and mean

diffusivity (MD)) was carried out using Tract-Based Spatial Statistics (TBSS).32 A

Generalised Linear Model (GLM) was applied to the VBM and TBSS analyses using

permutation-based non-parametric testing, correcting for multiple comparisons across

space (threshold-free cluster enhancement, tfce).

Page 10 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

10

Outcomes

Primary outcomes in the VBM analysis were continuous measures of grey matter

density, and in the TBSS analysis white matter integrity (fractional anisotropy, mean,

radial and axial diffusivity).

Visual ratings of hippocampal atrophy were dichotomised into atrophy vs. no atrophy

based on 0/1 on the (4 point) Scheltens scale to reflect clinical usage (“abnormal” vs.

“normal”).31

Hippocampal volume (%ICV) was used as a continuous variable in a multiple linear

regression analysis.

As cognitive outcomes we used decline on short-term memory, semantic and lexical

fluency (see below).

Statistical analysis

All analyses were done with R,33 unless otherwise stated.

Differences between included and excluded subjects, i.e. representativeness of included

participants, were examined using t-tests of means (continuous variables) or chi-square

tests (categorical). Subject characteristics for included subjects were summarised by

mean (with standard deviation) for continuous variables, and as percentage of the

sample for categorical variables split by safe vs. unsafe average alcohol use averaged

over all phases, on the basis of UK contemporary (pre-2016) guidelines. Significant

Page 11 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

11

differences between safe and unsafe drinkers were tested using t-tests of means for

continuous variables and chi-square tests for categorical variables. Weekly consumption

of alcohol (units and grams) was described using means, standard deviations, medians

and interquartile ranges, when data were not normally distributed.

Alcohol trends over time were examined using mixed effects modelling, with time from

study baseline (Phase 1) as the independent variable and alcohol consumption (units

per week) as the dependent variable. This method accounts for missing data, and

correlation of repeated measures (in this case alcohol use). Intercepts (baseline

consumption) and slopes (trends over study) were calculated for each participant.

Bivariate Person’s correlations were calculated between slopes/intercepts of alcohol

consumption and: age, sex, premorbid IQ, education, social class, Framingham Risk

Score (a composite measure including smoking, cardiovascular disease or diabetes,

cardiovascular medication), current psychotropic medications (number), lifetime

history of Major Depressive Disorder (SCID, binarised yes(2)/no(1)), exercise

frequency, club attendance, voluntary work, visits with friends and family. Correlations

were deemed significant if p<0.05 on two-tailed tests.

Mean alcohol consumption (units per week) across all study phases was included as an

independent variable in voxel-based morphometry (grey matter density as dependent

variable) and TBSS analyses (FA/MD/RD/AD as dependent variable). Voxelwise, a

Generalised Linear Model (GLM) was applied using permutation-based non-parametric

testing (randomise)34, correcting for multiple comparisons across space (threshold-free

cluster enhancement, tfce). Results were judged significant, if adjusted p<0.05.

Two post-hoc tests were used to confirm the associations between alcohol consumption

and hippocampal size following the VBM analysis. First, logistic regression was used to

Page 12 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

12

calculate odds ratios for left and right hippocampal atrophy versus no atrophy (visual

atrophy ratings based on a cut-off of 0/1 on the Scheltens’ scale),31 given average alcohol

consumption across study phases. The latter was categorised as abstinent (<1 unit,

reference group), 1 to <7 units, 14 to <21 units, 21 to <30 units and >30 units per week.

Second, multiple linear regression with hippocampal volume (extracted from FIRST, ICV-

adjusted) as the dependent variable and alcohol consumption as an independent

variable was performed.

In all analyses, the following potential confounding variables (identified from

knowledge of the literature) were included as independent variables: age, sex,

premorbid IQ, education, social class, Framingham Risk Score (a composite measure

including smoking, cardiovascular disease or diabetes, cardiovascular medication),

current psychotropic medications (number), lifetime history of Major Depressive

Disorder (SCID, binarised as yes (2)/no (1)), exercise frequency, club attendance,

voluntary work, visits with friends and family. In the subset with data on personality

traits (n=179), analyses were additionally adjusted for impulsiveness.

Changes in cognition over time (short-term verbal recall, lexical and semantic fluency)

were modelled using mixed effects models in a similar manner to alcohol, with test

score as the dependent variable and time from baseline (Phase 1) as the independent

variable. Slopes (trends over time) were calculated for each subject for further analyses,

which included binarisation into “cognitive improvers” (positive slope) vs. “cognitive

decliners” (negative slope). Independent t tests were used to identify significant

differences in improvers and decliners with respect to sociodemographic and clinical

variables.

Page 13 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

13

Structural equation modelling (SEM) (Amos 22 for Windows) was used for hypothesis

testing and to generate fit statistics for models of relationships between alcohol use,

brain measures and cognitive decline. SEM allows simultaneous analysis of multiple

variables in one model, can handle incomplete data and time series with auto-correlated

errors. The hypothesised underlying structure of the model was constructed following

the VBM and TBSS analyses, with average alcohol consumption as an exogenous

variable, hippocampal volume, corpus callosum MD (generally the most sensitive

measure of loss of white matter integrity), and decline in lexical fluency (slope from

mixed effects model) included as endogenous variables (using latent variables to

account for measurement error). Covariance of alcohol with sex and IQ, and between

brain measures was modelled. The model was improved by iteratively eliminating paths

with p>0.1, and monitoring of the successive improvement of the model fits statistics

(chi square, Comparative Fit Index, Root Mean Square Error of Approximation and the

Tucker-Lewis Index) until the most parsimonious model was identified.

Patient involvement

Participants are from the Whitehall II cohort. No patients were involved in setting the

research question or the outcome measures, nor were they involved in the design,

recruitment to, or conduct of the study. No patients were asked to advise on

interpretation or writing up of results. Results were disseminated to the study

participants in abstract format and as presentations at the 30th anniversary day for the

Whitehall II cohort. Participants are thanked in the acknowledgements.

Page 14 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

14

Figure 1: Flow chart of participants included in analysis.

Page 15 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

15

Figure 2: Frequency distribution of alcohol consumption on average across the study by sex.

Page 16 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

16

Safe drinkers

N=428 Mean (SD) or

n (%)

Unsafe drinkers

N=99 Mean (SD) or n (%)

Continuous variables:

mean difference (95%

CI)

Categorical variables:

chi squared (degrees

of freedom, p value)

Age at start (years) 43.0 (5.4) 42.8 (5.1) 0.2 (0.0 to 1.4)

Sex, n male (%) 339 (79.2%) 85 (85.9%) 2.3 (1, 0.1)

Marital status (% married) 308 (72.0%) 81 (81.8%) 5.3 (5, 0.4)

Full-time education (years) 14.5 (3.3) 14.9 (2.9) -0.4 (-1.0 to 0.3)

Full Scale Intelligence Quotient (estimated from TOPF1)*

117.4 (10.6) 120.0 (8.3) -2.6 (-4.5 to -0.7)

Social class 1, 2, 3 & 4 (%)$ 62, 328, 34, 4

(14.5%, 76.6%,

7.9%, 0.9%)

21, 76, 2, 0 (21.2%,

76.8%, 2.0%, 0%) 7.4 (3, 0.6)

Smoker (%) 11 (2.6%) 11 (11.1%) 14.7 (1, <0.001)

Systolic blood pressure 140.3 (17.8) 143.3 (16.7) -3.0 (-6.9 to 0.8)

Diastolic blood pressure 76.3 (10.5) 77.7 (10.9) -1.4 (-3.7 to 0.9)

Framingham Stroke Risk total score (%)*

11.4 (8.3) 12.8 (8.3) -1.4 (-3.2 to 0.4)

History of Major Depressive

disorder (%)*5

79 (18.5%) 16 (16.2%) 0.3 (1, 0.6)

Social visits (weekly)* 4.4 (3.2) 3.9 (3.1) 0.5 (-0.3 to 1.2)

Psychotropic medication (%)* 62 (14.5%) 12 (12.1%) 5.9 (4, 0.2)

MoCA3 total (/30)* 27.1 (2.4) 27.1(2.1) 0.0 (-0.5 to 0.5)

MMSE baseline total (/30)4 28.8 (1.1) 28.9 (1.2) -0.1 (-0.3 to 0.2)

1Test of Premorbid Function, 2Centre for Epidemiological Studies Depression Scale,3 Montreal Cognitive Assessment, 4Phase 7 (n=389), 5Structured Clinical Interview for DSM IV (SCID), $Social class based on occupation at phase 3: 1=professional, 2=managerial, 3=skilled non-manual, 4=skilled manual, *Time of Scan

Table 1: Baseline (Phase 1 unless otherwise indicated) summary characteristics of 527 participants (unless marked) included in analysis by safe (<14 units weekly women, <21 units weekly men) alcohol consumption, defined by contemporaneous (pre-2016) UK Department of Health guidelines, on average over study duration. Bold results significant as defined by p<0.05 on t test of means (continuous variables) and chi-square test (categorical variables).

Page 17 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

17

CROSS-SECTIONAL MEASURES

Cognitive test Mean/ median*

(SD)

Unstandardised coefficient

Change in test score with each

10 unit increase in

weekly alcohol consumption

P value 95% CI

MoCA

(adjusted for

education)

28.0*

(2.4)

-0.04 -0.02 0.7 -0.2 to

0.01

Semantic

fluency

22.0

(5.7)

0.1 0.1 0.8 -0.04 to

0.5

Lexical fluency 15.7

(4.5)

-0.1 -0.2 0.7 -0.5 to 0.3

TMT A

(sec)

29.0*

(11.4)

-0.2 -0.2 0.7 -1.0 to 0.9

TMT B

(sec)

59.0*

(36.8)

0.2 0.5 0.3 -1.0 to 5.0

Rey copy 32.0*

(4.0)

-0.04 -0.1 0.8 -0.4 to 0.3

Rey immediate 15.5

(6.5)

-0.3 -0.4 0.3 -1.0 to

0.03

Rey delay 15.2

(6.3)

-0.4 -0.7 0.1 -1.0 to 0.1

HVLT total

recall

27.5

(4.8)

0.4 0.8 0.06 -0.02 to

0.7

HVLT delayed

recall

10.0*

(2.8)

0.08 0.3 0.5 -0.2 to 0.3

BNT 59.0

(4.7)

0.4 0.9 0.02 0.1 to 0.7

Digit span total 30.1

(5.8)

0.2 0.3 0.5 -0.3 to 0.6

Digit

substitution

test

62.1

(13.7)

-0.6 -0.5 0.3 -2.0 to 0.5

LONGITUDINAL MEASURES

Short-term

memory

-0.02

(0.03)

-0.0004 -0.2 0.7 -0.003.to

0.002

Lexical fluency -0.1

(0.05)

-0.004 -1.0 0.04 -0.008 to -

1.4 Semantic

fluency

-0.05

(0.05)

-0.002 -0.4 0.4 -0.007 to

0.003 *Median was used to describe the central tendency of cognitive test scores where the distribution

was not normal.

Table 2: Effects of alcohol consumption on cognitive function and ~ trajectory. Results of regression analyses, with cross-sectional cognitive test performance or longitudinal cognitive test slope as the dependent variable, and average alcohol across the study as an independent variable. Adjusted for: age, sex, education, FSIQ; MoCA=Montreal Cognitive Assessment, TMT=Trail making test, Rey=complex figure task, HVLT=Hopkins Verbal Learning Test, BNT=Boston naming test (associated with IQ, hence positive association).

Page 18 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

Confidential: For Review Only

18

Men Women All

R Hippocampal atrophy

(vs. none)

L Hippocampal atrophy

(vs. none)

R Hippocampal atrophy

(vs. none)

L Hippocampal atrophy (vs.

none

R Hippocampal atrophy (vs.

none) L Hippocampal atrophy (vs.

none)

Alcohol

(units

weekly)

N

(Total) OR

(95%

CI)

P N

(Total) OR (95%

CI) P N

(Total) OR (95%

CI) P N

(Total) OR (95%

CI) P N

(Total) OR

(95% CI)

P N

(Total) OR (95%

CI) P

0-<1 9 (22) 12 (22) 4 (15) 7 (15) 13 (37) 19

(37)

1 to <7 55 (99) 1.6

(0.6-

4.3)

0.4 69 (99) 1.7 (0.6-4.8) 0.3 12 (41) 1.1 (0.2-

5.6) 0.9 20 (41) 0.8 (0.2-3.6) 0.8 67

(140) 1.5 (0.7-

3.4) 0.3 89

(140) 1.3 (0.6-3.0) 0.5

7 to

<14 68

(132) 1.7

(0.6-

4.5)

0.3 85

(132) 1.5 (0.6-4.2) 0.4 19 (33) 3.1 (0.6-

16.6) 0.2 22 (33) 2.0 (0.4-

10.2) 0.4 87

(165) 2.0 (0.9-

4.4) 0.1 107

(165) 1.4 (0.6-3.2) 0.4

14 to

<21 57 (86) 3.2

(1.1-

9.3)

0.02 59 (86) 2.0 (0.7-5.7) 0.2 6 (11) 4.2 (0.6-

28.8) 0.1 9 (11) 6.2 (0.7-

55.2) 0.1 63 (97) 3.4 (1.4-

8.1) 0.007 68

(97) 1.9 (0.8-4.6) 0.1

21 to

<30 38 (54) 3.9

(1.3-

12.0)

0.02 39 (54) 2.2 (0.7-6.8) 0.2 1 (3) 1.1 (0.04-

26.9) 0.97 2 (3) 0.4 (0.4-

10.2) 0.4 39 (57) 3.6 (1.4-

9.6) 0.009 41

(57) 1.9 (0.7-4.9) 0.2

≥30 24 (31) 5.2

(1.4-

19.0)

0.01 27 (31) 6.3 (1.5-

27.0) 0.01 24 (31) 5.8 (1.8-

18.6) <0.001 27

(31) 5.7 (1.5-21.6) 0.01

Table 3: Adjusted ORs for left and right sided hippocampal atrophy on Scheltens visual rating score (reference based on abstainers), with average alcohol consumption (weekly units) 1- <7, 7- <14, 14- <21 and 21- <30, ≥30 units (abstinence <1 units as reference category). Significant (p<0.05) results in bold. N=527. Analyses adjusted for age, sex, premorbid IQ, education, social class, Framingham Risk Score, history of Major Depressive Disorder (SCID), exercise frequency, club attendance, social visits, current psychotropic medication. Numbers with hippocampal atrophy (N) and total numbers in drinking category (Total N) are also given.

Page 19 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

19

Figure 3: Voxel-based morphometry results (threshold-free cluster enhancement (tfce) corrected): significant (p<0·05, red/yellow) negative correlation between weekly alcohol units (average of all phases across study) and grey matter density. N=527. Adjusted for age, sex, education, premorbid FSIQ, social class, physical exercise, club attendance, social activity, Framingham Stroke Risk Score, psychotropic medication, and history of Major Depressive Disorder.

Page 20 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

20

Figure 4: Tract-based spatial statistics results (threshold-free cluster enhancement, tfce, corrected) showing negative correlation between average alcohol across study (all phases) and fractional anisotropy (top), radial diffusivity, mean diffusivity and axial diffusivity (bottom). N=511. Significant results (p<0·05) in red/yellow Adjusted for age, sex, education, premorbid FSIQ, social class, physical exercise, club attendance, social activity, Framingham Stroke Risk Score, psychotropic medication, and history of Major Depressive Disorder

Unstandardised B

(95% CI)*

Change in

hippocampal volume

(% intracranial

volume)*

P value

Unadjusted alcohol -0.06 (-0.08 to -0.03) -2.0 <0.001

Adjusted alcohol1 -0.04 (-0.06 to -0.02) -1.5 0.001

* For every 10 unit increase in alcohol weekly 1 Analyses adjusted for age, sex, premorbid IQ, education, social class, Framingham Risk Score, history of Major Depressive Disorder (SCID), exercise frequency, club attendance, social visits, current psychotropic medication.

Table 4: Multiple linear regression results, with hippocampal volume (% of ICV) as the dependent variable and average alcohol consumption as an independent variable.

Figure 5: Final parsimonious structural equation model (excluding covariance arrows for simplicity) illustrating relationships among alcohol consumption (average across study phases, units per week), hippocampal volume (average, %ICV), corpus callosum mean diffusivity (MD), decline in lexical fluency (slopes), and age. Values on arrows are standardized regression weights. Red arrows indicate inverse relationships and green positive relationships. Model explained 21% corpus callosum MD, 14% of hippocampal variance and 1% of lexical fluency decline variance (R2). Model fit: Chi-square=14.2, degrees of freedom=9, p=0.12, Root Mean Square Error of Approximation=0.03, Comparative Fit Index=0.98, Tucker-Lewis Index=0.95.

Page 21 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

21

Path Standardised regression

coefficient P value

Average alcohol to

hippocampal volume -0.20 <0.001

Average alcohol to corpus

callosum MD 0.11 0.005

Average alcohol to lexical

fluency decline -0.08 0.008

Corpus callosum MD to

lexical fluency decline -0.07 0.1

Age to hippocampal volume -0.31 <0.001

Age to corpus callosum MD 0.44 <0.001

Table 5: Regression coefficients for paths in the final structural equation model (figure 5) with their standard errors and p values.

Results

Participants/descriptive data

Sociodemographic, health and lifestyle data are reported for the 527 included

participants, separated into alcohol consumption groups (table 1). Twenty-three

subjects were excluded from the VBM and visual ratings analyses on the basis of

structural brain abnormalities, poor quality images or missing confounder data (figure

1). A further sixteen subjects were excluded from the TBSS analysis due to missing or

poor quality diffusion tensor images. Excluded subjects did not significantly differ from

included subjects on any of the reported characteristics (data available from author on

request). Subjects were slightly less educated, with higher blood pressure and lower

depressive symptoms compared to the larger Whitehall II cohort (see supplementary

Page 22 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

22

materials, table 1). Mean age was 43.0 years (SD=5·4) at the study start (supplementary

materials, table 2). Unsafe drinkers differed from safe drinkers only by having a higher

premorbid IQ, a higher percentage of men and more smokers (see table 1).

Median alcohol consumption across study phases (supplementary materials, table 3)

was 10.2 units (81.3 g) per week (SD=10.2). Baseline consumption (intercept in mixed

effects model) was higher in men (r=0.24, p<0.001). Weekly alcohol intake did not

significantly increase over the phases of the study for the group as a whole, (β=-0.006,

SE=0.02 p<0.7), but trends over time correlated with baseline intake (intercepts and

slopes correlated negatively (r= -0.4, 95%CI: 0.5 to 0.3), i.e. those drinking more at

baseline tended to lower their consumption more over the course of the study. Other

sociodemographic and clinical factors were not related to longitudinal trends in

consumption. Average alcohol use over the study was over “safe limits” in 13.6% women

and 20.0% men, as judged by pre-2016 UK guidelines (>21 units (168g)/week for men,

>14 units (112g)/week for women), and 40.3 % as judged by the 2016 revised UK

guidelines (>14 units (112g) per week for men and women) (see supplementary

materials for consumption data for single phases). Scores on the CAGE questionnaire

were below the sensitive screening cut-off of 228 for all participants at all Whitehall II

phases (see supplementary materials, table 4).

Alcohol and brain structure

Higher alcohol use was associated with reduced grey matter density, hippocampal

atrophy, and reduced white matter microstructural integrity.

Page 23 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

23

Grey matter

Average alcohol consumption over the study (units per week) was negatively correlated

with grey matter density in the VBM analyses, especially in hippocampi (figure 3), even

after adjustment for multiple potential confounders. Associations also extended

anteriorly into the amygdalae. Frontal regions were unaffected.

Increased alcohol consumption was also associated with increased odds of abnormally

rated hippocampal atrophy (defined as score >0 on Scheltens visual rating scale; table

3). This was a dose-dependent effect. The highest odds were in those drinking in excess

of 30 units per week (OR 5.8 95% CI: 1.8 to 18.6 p=<0.001), but odds of atrophy were

higher compared with abstinence even in those drinking at moderate levels of 7 to <14

units per week (OR 3.4 95% CI: 1.4 to 8.1 p=0.007). There was no protective effect (i.e.

reduced odds of atrophy) of light drinking (1- <7 units per week). Findings were similar

in sub-analyses of men alone, but not in the smaller subgroup of women. Associations

were stronger and beginning at lower alcohol consumptions for right-sided atrophy. No

significant association was found with trend in alcohol use over time (slopes from mixed

effects models).

Mean hippocampal volumes (raw and ICV-adjusted) were within the range cited in the

literature (see supplementary materials, table 5),35-37 and correlated with visual ratings

of hippocampal atrophy (Pearson’s r= -0.4 p<0.001). Consistent with VBM and visual

ratings findings, alcohol consumption independently predicted FIRST-extracted

hippocampal volume (%ICV) (see supplementary materials). Excluding the three highest

drinkers (>60 units weekly) did not substantially change the results (supplementary

materials, table 6). On the subset, for whom personality trait data were available from

Phase 1 (n=179), additionally adjusting the analysis for trait impulsivity did not alter the

Page 24 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

24

findings.

White matter

Participants, who on average across the study drank more alcohol, had abnormalities in

white matter microstructure (figure 4), reflected by reduced corpus callosum fractional

anisotropy, and increased radial, axial and mean diffusivity. These associations were

focused on the anterior corpus callosum (genu and anterior body, figure 4).

Alcohol and cognitive function

Scores on all three cognitive tests significantly declined over the study for the group as a

whole (short-term recall: B= -0.02 95% CI= -0.03 to -0.006; lexical fluency: B= -0.10

95% CI= -0.12 to -0.08; semantic fluency: B= -0.05 95% CI= -0.07 to -0.04). “Cognitive

improvers” on the short-term recall test had a significantly higher FSIQ than “cognitive

decliners” (see supplementary materials, table 8). Average alcohol consumption across

phases predicted decline in lexical fluency (slopes), independently of age, sex, education

and FSIQ (supplementary materials). Additional adjustment for corpus callosum mean

diffusivity rendered the association insignificant, although the effect size changed only

marginally. No other sociodemographic or clinical variable was associated with lexical

decline.

Page 25 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

25

Modelling alcohol consumption and brain structure and function

In order to see how alcohol consumption and the associated brain regions interacted

with cognitive decline, structural equation modelling was used. Hippocampal volume

and corpus callosum MD were included as exogenous variables. Age, sex and premorbid

FSIQ were also incorporated.

Removing regression arrows from age, sex, FSIQ and hippocampal volume to lexical

fluency decline improved the model fit. Alcohol consumption independently predicted

decline in lexical fluency. The final parsimonious model explained 21% of corpus

callosum MD, 14% of right hippocampal volume, and 1% of lexical fluency decline

variance (see figure 5 & table 5), with good model fit (Chi square=14.2, degrees of

freedom=9, p=0.12, Root Mean Square Error of Approximation=0.03, Comparative Fit

Index=0.98, Tucker-Lewis Index=0.95). Alcohol consumption (in addition to age)

predicted smaller hippocampal volume, and greater corpus callosum MD. Through its

relationship with corpus callosum MD, and through a direct path, increased alcohol

consumption predicted faster decline of lexical fluency.

Discussion

Principle findings

We have found a previously uncharacterized dose-dependent relationship between

alcohol consumption over 30 years of follow-up and hippocampal atrophy, as well as

impaired white matter microstructure. Additionally, higher alcohol consumption

predicted greater cognitive decline. There was no evidence of a protective effect of light

Page 26 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

26

drinking on brain structure or function. The hippocampal findings were consistent

between the brain-wide voxel-based approach, automatically extracted volumes, and

clinical visual ratings of hippocampal atrophy. The relationship was dose dependent,

and increased odds of hippocampal atrophy were found even in moderate drinkers (14-

<21 units weekly in men). The association between alcohol consumption and white

matter microstructure in non-dependent drinkers is also novel and appeared to be

driven by greater radial relative to axial diffusivity.

Strengths and limitations

Strengths of this study are the 30-year longitudinal alcohol data and the detail of

available confounder data. Additional strengths include the availability of a large

number of MRI data and the advanced imaging analysis methods. Grey matter findings

were replicated using a voxel-based approach, automated hippocampal volumes, and

visual ratings. Visual atrophy ratings are known to correlate closely with automated

methods (own data) and are more applicable to clinical settings.38 In large neuroimaging

studies automatic segmentation is widespread.39,40 The automated approach we use

(FIRST) has been shown to give accurate and robust results.41

When interpreting these results, some caveats are necessary. Whilst the sample is

community dwelling, it may not be representative of the wider UK population. They are

mostly male, educated and middle-class. The hippocampal atrophy associations we

found in the total sample were replicated in men alone, but not in women. This may

reflect a lower power to detect an effect in women, in part due to the sample being

dominated by men (a reflection of the civil service gender disparity in 1980s), and in

part because few of the included women drank heavily. This is an observational study,

as alcohol use cannot be randomised. Observer bias may have influenced participants to

lead healthier lifestyles as they were enrolled in Whitehall II “Stress and Health” study.

Page 27 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

27

The alcohol use data are self-reported, and participants may have underestimated their

drinking, although the longitudinal rather than cross-sectional approach taken in other

reported studies may minimise this,26 and the percentage of subjects drinking “unsafely”

are comparable to other reported studies.42-44 We used the CAGE screening instrument

to identify alcohol dependence, as it is well validated.28,45 Some subjects reported

drinking high levels of alcohol whilst screening negative on the CAGE, indicating a

possible inclusion of subjects with an alcohol use disorder in the sample. However,

increased odds of hippocampal atrophy were found even in those drinking moderate

amounts. Although the alcohol and cognitive data was longitudinal, the analyses with

MRI measures were cross-sectional, raising the possibility that the associations between

brain structure and alcohol were the result of a confounding variable. Longitudinal

imaging over more than a couple of years adds further confounders as the physical

scanner and imaging sequences are unlikely to be the same due to developments in MRI

science, making results difficult to interpret. Whilst efforts have been made to control

for multiple potential sources of confounding, we cannot exclude the possibility of

residual confounding from unmeasured sources. However to produce the adjusted

associations we found, any uncontrolled confounders would need to be associated with

both alcohol consumption and risk of brain abnormalities and unrelated to the multiple

factors we controlled for. Finally, we cannot exclude the possibility, of unlikely face

validity, that those with hippocampal atrophy at study baseline were more likely to

drink more.

Comparison with other studies

On average, 20% of men and 14% of women were drinking above contemporary

guidelines (>21 units/ >14 units weekly, respectively). Other studies vary in reported

rates of heavy drinking, but our rates are comparable.42,43 Alcohol consumption may

Page 28 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

28

vary according to country, as highlighted by a study utilizing the WHO Global Alcohol

Database.44

Hippocampal atrophy is a sensitive and relatively specific marker of Alzheimer’s

disease,46 although it has been reported in chronic alcoholics.18,47 The brain regions most

vulnerable to alcohol abuse are said to be the frontal lobes.21 Higher but non-dependent

alcohol use was not associated with subsequent frontal brain atrophy or impaired

cognition in our sample. Only one study has reported hippocampal findings in non-

dependent drinkers. This used a manual tracing rather than voxel-based or visual rating

approach to estimate hippocampal size. They reported a protective effect of moderate

alcohol in comparison with abstinence, which conflicts with our results.18 However,

alcohol consumption was determined cross-sectionally, making it difficult to exclude

reverse causation. In contrast, due to the longitudinal cognitive component of our study

we were able to demonstrate an association between higher alcohol consumption and

cognitive decline. Additionally, a number of known hippocampal size confounders, such

as depression, were not controlled for in Den Heijer et al.48 Other studies in non-

dependent drinkers have reported either no effect 26,49,50 or a negative correlation with

global grey matter, but not hippocampal atrophy.17,51 In contrast with our first

hypothesis and some other studies,11,12,18,52 we observed no evidence of a protective

effect of light drinking compared with abstinence on brain structure or cognitive

function. Previous studies did not control for (premorbid) IQ,11,12 and only a few for

socioeconomic class.53-55 The observed protective effect may be due to confounding as

we and others found a positive association between alcohol intake and IQ.56 These

factors separately predict better performance on cognitive tests. Supporting our second

hypothesis, we found heavier alcohol consumption to be associated with adverse brain

outcomes. The biological mechanism for this is unclear. Ethanol and acetaldehyde (a

metabolite) are neurotoxic57 and cause reduced numbers58,59 and morphological

Page 29 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

29

changes in hippocampal neurones in animal models.60 Associated thiamine and folate

deficiency,61 repeated head trauma, cerebrovascular events, liver damage, and repeated

intoxication and withdrawal have also been implicated in more severe drinkers. The risk

of hippocampal atrophy appeared to be stronger and at lower levels of alcohol

consumption for the right side. More significant hippocampal atrophy on the right has

been described in those at higher risk of Alzheimer’s disease (asymptomatic ApoE4

homozygotes),62 as well as in those with mild cognitive impairment or Alzheimer’s

disease.63 We also found no structural laterality in associations with cognitive function.

The literature on this is scarce and conflicting. Stronger associations between right

hippocampal volume and visuo-spatial memory have been reported.64

The VBM analysis also showed associations between increased alcohol consumption and

reduced grey matter density in the amygdalae. This result could not be confirmed using

other methods, as automated segmentation of these regions was unreliable, and we are

unaware of any reliable visual atrophy rating scales. Amygdala atrophy has been

described in Alzheimer’s disease,65 and is implicated in preclinical models of alcohol

misuse,66 alcohol abuse relapse67 and in abstinent alcoholics,68 although others have

found no association with lower levels of consumption.50

In animals, radial diffusivity reflects differences in myelination.69,70 Prior studies have

highlighted the corpus callosum as an area affected in foetal alcohol syndrome,71 and in

chronic alcoholism in Marchiafava-Bignami disease.72,73 One study reported increased

mean diffusivity in the amygdala in a post-hoc analysis of female non-dependent

drinkers.25 We are not aware of any studies investigating white matter microstructural

changes in moderate drinkers using a data driven skeletonised tract approach to

diffusion tensor images, such as TBSS. Alternative voxel-wise methods may compromise

Page 30 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

30

optimal analysis of multiple subjects, as there are alignment problems causing potential

difficulties with interpretation of voxel-wise statistics.74

Increased alcohol consumption was associated with a small but significant increase in

decline of lexical fluency over the study. Lexical fluency involves selecting and retrieving

information based on spelling (orthography), and has characteristically been associated

with frontal executive function,75 in contrast to semantic fluency which may depend

more on temporal lobe integrity.76 However, the distinction may not be as clear cut as

functional networks overlap.77 The inverse relationship between alcohol consumption

and lexical decline was perhaps unsurprising given the frontal predominance of the

negative associations with white matter integrity. We suggest two possibilities for the

lack of more widespread associations with cognition, particularly with semantic fluency

and short-term memory decline, given the structural brain findings (hippocampal

atrophy). First, there appears to be a practice effect over the study, i.e. at least some

subjects improve their performance after repeated testing. Therefore the trend in test

score may reflect learning ability of a subject, which obscures detection of cognitive

decline. Whilst only 2% of subjects were “cognitive improvers” (had a positive slope

indicated improving cognition over the study) for lexical fluency, 8% were for semantic

fluency, and 25% were for short-term memory. This suggests that certain tests were

more susceptible to practice effects. Furthermore, variables predicting the ability to

learn may be different from those protecting against cognitive impairment due to a

neurodegenerative process. “Cognitive improvers” on the short-term memory test had

significantly higher FSIQ than “cognitive decliners”. The correlation between heavier

alcohol consumption and a larger learning effect may be explained through confounding

for instance by IQ, which was associated with drinking. Adjustment for FSIQ alone may

be insufficient to remove the confounding effect if a third variable, for example diet,

Page 31 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

31

mediates the relationship between IQ and learning but is not in the model. Second, the

brain changes may reflect an intermediate phenotype and cognitive change is not yet

evident. It is now well documented that hippocampal atrophy precedes symptoms in

Alzheimer’s by several years,78 so a similar phenomenon in alcohol-related changes is

plausible.

Conclusions and policy implications

Prospective studies of the effects of alcohol use on the brain are few, and replication of

these findings will be important. Alcohol consumption for individuals was remarkably

stable across the study phases. This sample was therefore underpowered to detect

differences in those significantly changing their intake from others who drink

consistently. Larger investigations are needed to clarify whether there are graded risks

between a short versus long periods of higher alcohol consumption.

The finding that alcohol consumption in moderate quantities is associated with multiple

markers of abnormal brain structure and cognitive function has important potential

public health implications for a large sector of the population. By example, in our sample

nearly half of the men and a quarter of women were currently drinking in this range.

Additionally, drinking habits were remarkably stable over a 30-year period, suggesting

that risky drinking habits may be embarked upon in midlife. Recommended guidelines

for drinking have remained unchanged in the UK from 1987 until 2016. Our findings

support the recent reduction in UK safe limits and call into question the current US

guidelines which suggest that up to 24.5 units weekly is safe for men, as we found

increased odds of hippocampal atrophy at just 14-21 units weekly, and we found no

support for a protective effect of light consumption on brain structure. Alcohol may

represent a modifiable risk factor for cognitive impairment, and primary prevention

interventions targeted to later life may be too late.

Page 32 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

32

Acknowledgements

We would like to thank the Whitehall II cohort participants for their time.

Contributions

Author Planning of Study

Data acquisition

Data analysis

Contributed to paper

Anya Topiwala X X X X

Charlotte L. Allan X X X X

Vyara Valkanova X X

Enikő Zsoldos X X

Nicola Filippini X X X

Claire Sexton X X X

Abda Mahmood X X

Peggy Fooks X X

Archana Singh-Manoux

X X

Clare E. Mackay X X

Mika Kivimäki X X

Klaus P. Ebmeier X X X

Competing interests All authors have completed the ICMJE uniform disclosure form at

www.icmje.org/coi_disclosure.pdf and declare: grant support for the submitted work is

detailed above; no financial relationships with any organisations that might have an

interest in the submitted work in the previous three years; no other relationships or

activities that could appear to have influenced the submitted work.

Ethical approval

Predicting MRI abnormalities with longitudinal data of the Whitehall II sub- study”

(MSD/IDREC/C1/2011/71). IDREC,Research Services,University of

Oxford,Wellington Square, Oxford, OX1 2JD

Data sharing

Data sharing policy referenced on: https://www.psych.ox.ac.uk/research/neurobiology-

of-ageing/research-projects-1/whitehall-oxford - Reference to Whitehall II Data sharing policy here: http://www.ucl.ac.uk/whitehallII/data-sharing.

Transparency

Dr Anya Topiwala, Profs Klaus P Ebmeier and Mika Kivimäki (the manuscript's

guarantors) affirm that the manuscript is an honest, accurate, and transparent account

of the study being reported; that no important aspects of the study have been omitted;

Page 33 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

33

and that any discrepancies from the study as planned (and, if relevant, registered) have

been explained.

A CC BY licence is required.

References:

1. Organization WH. Global status report on alcohol and health-2014: World Health

Organization; 2014.

2. Connor JP, Haber PS, Hall WD. Alcohol use disorders. The Lancet 2015.

3. Leon DA, McCambridge J. Liver cirrhosis mortality rates in Britain from 1950 to 2002: an

analysis of routine data. The Lancet 2006; 367(9504): 52-6.

4. Government H. The Government's Alcohol Strategy. 2012.

5. Stampfer MJ, Colditz GA, Willett WC, Speizer FE, Hennekens CH. A prospective study of

moderate alcohol consumption and the risk of coronary disease and stroke in women. New

England Journal of Medicine 1988; 319(5): 267-73.

6. Rimm EB, Williams P, Fosher K, Criqui M, Stampfer MJ. Moderate alcohol intake and

lower risk of coronary heart disease: meta-analysis of effects on lipids and haemostatic factors.

Bmj 1999; 319(7224): 1523-8.

7. Cao Y, Willett WC, Rimm EB, Stampfer MJ, Giovannucci EL. Light to moderate intake of

alcohol, drinking patterns, and risk of cancer: results from two prospective US cohort studies.

2015.

8. Agriculture USDoHaHSaUSDo. 2015-2020 Dietary Guidelines for Americans. 8th Edition.

2015.

9. Bagnardi V, Rota M, Botteri E, et al. Alcohol consumption and site-specific cancer risk: a

comprehensive dose–response meta-analysis. British journal of cancer 2015; 112(3): 580-93.

10. Association AP. Diagnostic and statistical manual of mental disorders (4th ed., text rev.).

2000.

11. Mukamal KJ, Kuller LH, Fitzpatrick AL, Longstreth Jr W, Mittleman MA, Siscovick DS.

Prospective study of alcohol consumption and risk of dementia in older adults. Jama 2003;

289(11): 1405-13.

12. Ruitenberg A, van Swieten JC, Witteman JC, et al. Alcohol consumption and risk of

dementia: the Rotterdam Study. The Lancet 2002; 359(9303): 281-6.

13. Cleophas T. Wine, beer and spirits and the risk of myocardial infarction: a systematic

review. Biomedicine & pharmacotherapy 1999; 53(9): 417-23.

14. Berger K, Ajani UA, Kase CS, et al. Light-to-moderate alcohol consumption and the risk of

stroke among US male physicians. New England Journal of Medicine 1999; 341(21): 1557-64.

15. Gu Y, Scarmeas N. Diet and neuroimaging markers of cerebrovascular disease. Current

nutrition reports 2013; 2(2): 81-9.

16. Paul CA, Au R, Fredman L, et al. Association of alcohol consumption with brain volume in

the Framingham study. Archives of neurology 2008; 65(10): 1363-7.

17. Ding J, Eigenbrodt ML, Mosley TH, et al. Alcohol Intake and Cerebral Abnormalities on

Magnetic Resonance Imaging in a Community-Based Population of Middle-Aged Adults The

Atherosclerosis Risk in Communities (ARIC) Study. Stroke 2004; 35(1): 16-21.

18. den Heijer T, Vermeer SE, van Dijk EJ, et al. Alcohol intake in relation to brain magnetic

resonance imaging findings in older persons without dementia. The American journal of clinical

nutrition 2004; 80(4): 992-7.

19. Sachdev PS, Chen X, Wen W, Anstry KJ. Light to moderate alcohol use is associated with

increased cortical gray matter in middle-aged men: a voxel-based morphometric study.

Psychiatry Research: Neuroimaging 2008; 163(1): 61-9.

20. Gu Y, Scarmeas N, Short EE, et al. Alcohol intake and brain structure in a multiethnic

elderly cohort. Clinical Nutrition 2014; 33(4): 662-7.

21. Kubota M, Nakazaki S, Hirai S, Saeki N, Yamaura A, Kusaka T. Alcohol consumption and

frontal lobe shrinkage: study of 1432 non-alcoholic subjects. Journal of Neurology, Neurosurgery

& Psychiatry 2001; 71(1): 104-6.

Page 34 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

34

22. de Bruin EA, Pol HEH, Schnack HG, et al. Focal brain matter differences associated with

lifetime alcohol intake and visual attention in male but not in female non-alcohol-dependent

drinkers. Neuroimage 2005; 26(2): 536-45.

23. Anstey KJ, Mack HA, Cherbuin N. Alcohol consumption as a risk factor for dementia and

cognitive decline: meta-analysis of prospective studies. The American journal of Geriatric

psychiatry 2009; 17(7): 542-55.

24. Fukuda K, Yuzuriha T, Kinukawa N, et al. Alcohol intake and quantitative MRI findings

among community dwelling Japanese subjects. Journal of the neurological sciences 2009; 278(1):

30-4.

25. Sasaki H, Abe O, Yamasue H, et al. Structural and diffusional brain abnormality related to

relatively low level alcohol consumption. Neuroimage 2009; 46(2): 505-10.

26. Poikolainen K. Underestimation of recalled alcohol intake in relation to actual

consumption. British journal of addiction 1985; 80(2): 215-6.

27. Marmot MG, Stansfeld S, Patel C, et al. Health inequalities among British civil servants:

the Whitehall II study. The Lancet 1991; 337(8754): 1387-93.

28. Mayfield D, McLeod G, Hall P. The CAGE questionnaire: validation of a new alcoholism

screening instrument. American journal of psychiatry 1974; 131(10): 1121-3.

29. Department of Health. How to keep health risks from drinking alcohol to a low level:

public consultation on proposed new guidelines. London: Williams Lea for the Department of

Health; 2016.

30. Filippini N, Zsoldos E, Haapakoski R, et al. Study protocol: the Whitehall II imaging sub-

study. BMC psychiatry 2014; 14(1): 159.

31. Scheltens P, Leys D, Barkhof F, et al. Atrophy of medial temporal lobes on MRI in"

probable" Alzheimer's disease and normal ageing: diagnostic value and neuropsychological

correlates. Journal of Neurology, Neurosurgery & Psychiatry 1992; 55(10): 967-72.

32. Smith SM, Jenkinson M, Johansen-Berg H, et al. Tract-based spatial statistics: voxelwise

analysis of multi-subject diffusion data. Neuroimage 2006; 31(4): 1487-505.

33. Team RC. R: A Language and Environment for Statistical Computing. 2015.

34. Winkler AM, Ridgway GR, Webster MA, Smith SM, Nichols TE. Permutation inference for

the general linear model. Neuroimage 2014; 92: 381-97.

35. Brown ES, Hughes CW, McColl R, Peshock R, King KS, Rush AJ. Association of depressive

symptoms with hippocampal volume in 1936 adults. Neuropsychopharmacology 2014; 39(3):

770-9.

36. through Meta ENIG. Common variants at 12q14 and 12q24 are associated with

hippocampal volume. Nature genetics 2012; 44(5): 545-51.

37. Thomas AG, Dennis A, Rawlings NB, et al. Multi-modal characterization of rapid anterior

hippocampal volume increase associated with aerobic exercise. NeuroImage 2016; 131: 162-70.

38. Harper L, Fumagalli GG, Barkhof F, et al. MRI visual rating scales in the diagnosis of

dementia: evaluation in 184 post-mortem confirmed cases. Brain 2016: aww005.

39. Stein JL, Medland SE, Vasquez AA, et al. Identification of common variants associated

with human hippocampal and intracranial volumes. Nature genetics 2012; 44(5): 552-61.

40. Hibar DP, Stein JL, Renteria ME, et al. Common genetic variants influence human

subcortical brain structures. Nature 2015; 520(7546): 224-9.

41. Patenaude B, Smith SM, Kennedy DN, Jenkinson M. A Bayesian model of shape and

appearance for subcortical brain segmentation. Neuroimage 2011; 56(3): 907-22.

42. Djoussé L, Levy D, Benjamin EJ, et al. Long-term alcohol consumption and the risk of

atrial fibrillation in the Framingham Study. The American journal of cardiology 2004; 93(6): 710-

3.

43. Fuchs CS, Stampfer MJ, Colditz GA, et al. Alcohol consumption and mortality among

women. New England Journal of Medicine 1995; 332(19): 1245-50.

44. Rehm J, rgen u, Rehn N, et al. The global distribution of average volume of alcohol

consumption and patterns of drinking. European addiction research 2003; 9(4): 147-56.

45. Beresford TP, Blow FC, Singer K, Hill E, Lucey MR. Comparison of CAGE questionnaire

and computer-assisted laboratory profiles in screening for covert alcoholism. The Lancet 1990;

336(8713): 482-5.

46. McKhann GM, Knopman DS, Chertkow H, et al. The diagnosis of dementia due to

Alzheimer's disease: recommendations from the National Institute on Aging-Alzheimer's

Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement

2011; 7(3): 263-9.

Page 35 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

35

47. Sullivan EV, Marsh L, Mathalon DH, Lim KO, Pfefferbaum A. Anterior hippocampal

volume deficits in nonamnesic, aging chronic alcoholics. Alcoholism: Clinical and Experimental

Research 1995; 19(1): 110-22.

48. Videbech P, Ravnkilde B. Hippocampal volume and depression: a meta-analysis of MRI

studies. American Journal of Psychiatry 2015.

49. Preti A, Muscio C, Boccardi M, Lorenzi M, de Girolamo G, Frisoni G. Impact of alcohol

consumption in healthy adults: A magnetic resonance imaging investigation. Psychiatry Research:

Neuroimaging 2014; 224(2): 96-103.

50. Anstey KJ, Jorm AF, Réglade-Meslin C, et al. Weekly alcohol consumption, brain atrophy,

and white matter hyperintensities in a community-based sample aged 60 to 64 years.

Psychosomatic Medicine 2006; 68(5): 778-85.

51. Mukamal KJ, Longstreth Jr W, Mittleman MA, Crum RM, Siscovick DS. Alcohol

consumption and subclinical findings on magnetic resonance imaging of the brain in older adults

the cardiovascular health study. Stroke 2001; 32(9): 1939-46.

52. Stampfer MJ, Kang JH, Chen J, Cherry R, Grodstein F. Effects of moderate alcohol

consumption on cognitive function in women. New England Journal of Medicine 2005; 352(3):

245-53.

53. Britton A, Singh-Manoux A, Marmot M. Alcohol consumption and cognitive function in

the Whitehall II Study. American Journal of Epidemiology 2004; 160(3): 240-7.

54. Sabia S, Elbaz A, Britton A, et al. Alcohol consumption and cognitive decline in early old

age. Neurology 2014; 82(4): 332-9.

55. McGuire LC, Ajani UA, Ford ES. Cognitive functioning in late life: the impact of moderate

alcohol consumption. Annals of epidemiology 2007; 17(2): 93-9.

56. Touvier M, Druesne-Pecollo N, Kesse-Guyot E, et al. Demographic, socioeconomic,

disease history, dietary and lifestyle cancer risk factors associated with alcohol consumption.

International Journal of Cancer 2014; 134(2): 445-59.

57. Arendt T, Allen Y, Sinden J, et al. Cholinergic-rich brain transplants reverse alcohol-

induced memory deficits. Nature 1988; 332(6163): 448-50.

58. Bengoechea O, Gonzalo L. Effects of alcoholization on the rat hippocampus. Neuroscience

letters 1991; 123(1): 112-4.

59. Cadete-Leite A, Tavares M, Pacheco M, Volk B, Paula-Barbosa M. Hippocampal mossy

fiber-CA3 synapses after chronic alcohol consumption and withdrawal. Alcohol 1989; 6(4): 303-

10.

60. McMullen PA, Saint-Cyr JA, Carlen PL. Morphological alterations in rat CA1 hippocampal

pyramidal cell dendrites resulting from chronic ethanol consumption and withdrawal. Journal of

comparative neurology 1984; 225(1): 111-8.

61. Langlais PJ, Savage LM. Thiamine deficiency in rats produces cognitive and memory

deficits on spatial tasks that correlate with tissue loss in diencephalon, cortex and white matter.

Behavioural brain research 1995; 68(1): 75-89.

62. Tohgi H, Takahashi S, Kato E, et al. Reduced size of right hippocampus in 39-to 80-year-

old normal subjects carrying the apolipoprotein E ϵ4 allele. Neuroscience letters 1997; 236(1):

21-4.

63. Shi F, Liu B, Zhou Y, Yu C, Jiang T. Hippocampal volume and asymmetry in mild cognitive

impairment and Alzheimer's disease: Meta-analyses of MRI studies. Hippocampus 2009; 19(11):

1055-64.

64. Soininen HS, Partanen K, Pitkänen A, et al. Volumetric MRI analysis of the amygdala and

the hippocampus in subjects with age-associated memory impairment Correlation to visual and

verbal memory. Neurology 1994; 44(9): 1660-.

65. Cuénod C-A, Denys A, Michot J-L, et al. Amygdala atrophy in Alzheimer's disease: an in

vivo magnetic resonance imaging study. Archives of neurology 1993; 50(9): 941-5.

66. Koob GF. Neuroadaptive mechanisms of addiction: studies on the extended amygdala.

European Neuropsychopharmacology 2003; 13(6): 442-52.

67. Wrase J, Makris N, Braus DF, et al. Amygdala volume associated with alcohol abuse

relapse and craving. American Journal of Psychiatry 2008.

68. Fein G, Landman B, Tran H, et al. Brain atrophy in long-term abstinent alcoholics who

demonstrate impairment on a simulated gambling task. Neuroimage 2006; 32(3): 1465-71.

69. Budde MD, Kim JH, Liang HF, et al. Toward accurate diagnosis of white matter pathology

using diffusion tensor imaging. Magnetic resonance in medicine 2007; 57(4): 688-95.

Page 36 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

36

70. Song S-K, Sun S-W, Ramsbottom MJ, Chang C, Russell J, Cross AH. Dysmyelination

revealed through MRI as increased radial (but unchanged axial) diffusion of water. Neuroimage

2002; 17(3): 1429-36.

71. Bookstein FL, Streissguth AP, Sampson PD, Connor PD, Barr HM. Corpus callosum shape

and neuropsychological deficits in adult males with heavy fetal alcohol exposure. Neuroimage

2002; 15(1): 233-51.

72. Pfefferbaum A, Lim KO, Desmond JE, Sullivan EV. Thinning of the corpus callosum in

older alcoholic men: a magnetic resonance imaging study. Alcoholism: Clinical and Experimental

Research 1996; 20(4): 752-7.

73. Harper C, Kril J, Daly J. Does a" moderate" alcohol intake damage the brain? Journal of

Neurology, Neurosurgery & Psychiatry 1988; 51(7): 909-13.

74. Smith SM, Jenkinson M, Johansen-Berg H, et al. Tract-based spatial statistics: voxelwise

analysis of multi-subject diffusion data. Neuroimage 2006; 31(4): 1487-505.

75. Birn RM, Kenworthy L, Case L, et al. Neural systems supporting lexical search guided by

letter and semantic category cues: a self-paced overt response fMRI study of verbal fluency.

Neuroimage 2010; 49(1): 1099-107.

76. Baldo JV, Schwartz S, Wilkins D, Dronkers NF. Role of frontal versus temporal cortex in

verbal fluency as revealed by voxel-based lesion symptom mapping. Journal of the International

Neuropsychological Society 2006; 12(06): 896-900.

77. Costafreda SG, Fu CH, Lee L, Everitt B, Brammer MJ, David AS. A systematic review and

quantitative appraisal of fMRI studies of verbal fluency: role of the left inferior frontal gyrus.

Human brain mapping 2006; 27(10): 799-810.

78. Jack CR, Knopman DS, Jagust WJ, et al. Tracking pathophysiological processes in

Alzheimer's disease: an updated hypothetical model of dynamic biomarkers. The Lancet

Neurology 2013; 12(2): 207-16.

Page 37 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Page 38 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Page 39 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Page 40 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Page 41 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Page 42 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Page 43 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Page 44 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Page 45 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Page 46 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Page 47 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Page 48 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Averagealcohol

Hippocampalvolume

Age

Lexicalfluencydecline-0.20

-0.31

0.44

-0.07

CorpuscallosumMD

0.11

-0.08

Page 49 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nlySupplementary methods

MRI analysis

Tissue segmentation

T1-weighted images were processed using FSL tools (FMRIB Software Library,

www.fmrib.ox.ac.uk/fsl) and ‘fsl_anat (Beta version)’

(http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/fsl_anat). This reorients images to standard (MNI) space,

corrects for bias field, registers the images to standard space (using linear FLIRT25,26 and non-

linear FNIRT27 registration), and extracts whole brain volumes (‘BET’)

28. FMRIB's

Automated Segmentation Tool (FAST) allowed extraction of measures of total grey matter

(GM), white matter (WM) and cerebrospinal fluid (CSF). GM and WM volumes were

adjusted for total intracranial volume.

Voxel-based morphometry (VBM)

Relationships between alcohol use and grey matter were examined initially using voxel-

based morphometry, an objective method to compare grey matter density between

individuals in each voxel (smallest distinguishable image volume) of the structural

image. Structural data were analysed with FSL-VBM 29

http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSLVBM), an optimised VBM protocol 30 carried out with

FSL tools 31. First, structural images were brain-extracted and grey matter-segmented before

being registered to the MNI 152 standard space using non-linear registration 27. The resulting

images were averaged and flipped along the x-axis to create a left-right symmetric, study-

specific grey matter template. Second, all native grey matter images were non-linearly

registered to this study-specific template and "modulated" to correct for local expansion (or

contraction) due to the non-linear component of the spatial transformation. The modulated

grey matter images were then smoothed with an isotropic Gaussian kernel with a sigma of 3

mm. Finally, voxelwise, a Generalised Linear Model (GLM) was applied using permutation-

based non-parametric testing, correcting for multiple comparisons across space (threshold-

Page 50 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nlyfree cluster enhancement, tfce).

Visual rating of hippocampal atrophy

Structural T1 scans were assessed independently, by three researchers for hippocampal

atrophy (HA) using the Scheltens scale according to the width of the choroid fissure, width of

the temporal horn, and height of the hippocampus (0-4).32 Raters remained blind to all other

participant data. Intra- (on a random 10% of 208 scans) and inter-rater reliability (n=208) for

visual rating scores was high (intra-class correlation coefficients: 0.8 to 0.9 and 0.7 to 0.9,

respectively). Left and right hippocampal atrophy was defined independently according

to visual rating (Scheltens score33 >0) by three clinicians, who reached a consensus.

Hippocampal volumes

Hippocampal volumes were calculated using FIRST (http://fsl.fmrib.ox.ac.

uk/fsl/fslwiki/FIRST)34 an automated model-based segmentation/registration tool in a two-

stage process – first using all subcortical masks, and second a hippocampal mask only. Both

were visually inspected to check optimal segmentation. Extracted hippocampal volumes were

corrected for intracranial volume and averaged across left and right sides.

Diffusion tensor imaging

Diffusion tensor imaging (DTI) measures the directional preference of water diffusion in

neural tissue and allows inferences about the structural integrity of white matter tracts.

In healthy myelinated fibres diffusion is restricted perpendicular to the longitudinal axis

of the fibre, i.e. it is anisotropic. Voxelwise statistical analysis of fractional anisotropy

(FA), axial diffusivity (AD), radial diffusivity (RD) and mean diffusivity (MD) data was

carried out using Tract-Based Spatial Statistics (TBSS).35 This involves non-linear

registration followed by projection onto an alignment-invariant tract representation

Page 51 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly(the “mean FA skeleton”). This avoids alignment problems for multiple subjects and

avoids arbitrariness of spatial smoothing extent, improving the sensitivity, objectivity

and interpretability of analysis of multi-subject diffusion imaging studies.36 Multiple

diffusion indices were analysed to allow a richer investigation of localised connectivity

related changes. AD describes diffusion parallel to, and RD perpendicular to the to the

principal fibre direction. MD is the apparent diffusion coefficient averaged over all

directions. FA reflects AD in relation to RD and is widely used as a marker of tract

integrity.37 38 Diffusion images were corrected for head movement and eddy currents

(eddy_correct) and brain masks generated using BET. Fractional anisotropy, mean

diffusivity, axial diffusivity and radial diffusivity maps were generated using DTIFit

(http://fsl.fmrib.ox.ac.uk/fsl/fdt) that fits a diffusion tensor model at each voxel. Tract-

based spatial statistics (TBSS) were used in a 4-stage process. Pre-processing prepared

images for registration to standard space. Mean fractional anisotropy (FA), diffusivity

(MD), radial diffusivity (RD), axial diffusivity (AD) and skeletonized FA, MD, RD and AD

images were created, and thresholded. Lastly each FA, MD, RD and AD image was

projected onto the relevant skeleton and ‘randomize’ used for statistical analyses. Masks

of specific white matter tracts were created using ICBM-DTI-81 white-matter labels

atlas39 and used to extract mean white matter integrity indices.

Page 52 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nlySupplementary results

Variable MRI Sample Phase 11 Participants

N=6306

Mean difference (95%

CI)/chi square (degrees of

freedom, p value)

N Mean/ % S.D. N Mean/% S.D.

Age [years] 527 69.6 5.3 6306 69.8 5.9 -0.2 (-0.7 to 0.3)

Sex

Female

Male

527

103

424

19.5%

80.5%

6306

1947

4459

29.3%

70.7%

27.5 (1, <0.001)

Full time education*

[years]

527 14.6 3.2 5101 15.1 4.2 -0.5 (-0.9 to -0.1)

CES-D* 527 5.0 5.8 5855 7.3 7.6 -2.3 (-3.0 to -1.6)

Systolic BP [mmHg]* 527 140.8 17.6 5652 127.8 16.5 13.0 (11.5 to 14.5)

Diastolic BP [mmHg]* 527 76.5 10.6 5652 70.8 9.9 5.7 (4.8 to 6.6)

Table 1: Comparison of imaging sample with Phase 11.

Men Women

Phase Age (mean,

s.d.)

% ‘unsafe’ drinkers

(Pre-2016 guidelines)

% ‘unsafe’ drinkers

(Post-2016 guidelines)

% ‘unsafe’ (Pre- = Post-

2016 guidelines)

1 43.0 (5.4) 17.1 32.1 11.9

3 48.2 (5.4) 17.9 32.1 14.4

5 53.1 (8.8) 28.5 44.6 17.2

7 59.5 (5.3) 22.0 40.0 16.5

9 64.4 (5.3) 18.4 34.0 13.7

11 68.5 (5.4) 14.6 28.5 7.8

Time of

Scan

69.6 (5.3) 25.6 47.8 27.0

Average 20.0 40.3 13.6

Table 2: Percent (male and female separately) drinking over safe weekly limits, as defined

by pre-2016 (>14 units/112g women, >21 units/168g men), and post-2016 guidelines

(>14 units/112g)

Page 53 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nlyPhase Mean Standard deviation Median Interquartile range

1 11.7 (93.9) 12.4 (98.9) 8.0 (64.0) 13.0 (104.0)

3 11.4 (91.6) 11.4 (91.1) 8.0 (64.0) 13.0 (104.0)

5 15.0 (120.1) 14.8 (118.8) 11.0 (88.0) 16.0 (128.0)

7 13.0 (104.0) 12.0 (96.0) 10.0 (80.0) 14.0 (120.0)

9 11.2 (89.3) 10.3 (82.7) 9.0 (72.0) 14.0 (112.0)

11 10.3 (82.4) 10.0 (80.1) 8.0 (64.0) 12.0 (96.0)

Time of Scan 15.0 (117.7) 10.2 (118.2) 12.0 (91.6) 16.9 (135.2)

Average 12.5 (99.7) 10.2 (81.5) 10.2 (81.3) 12.6 (96.8)

Table 3: Summary of alcohol units (grams) consumed per week at each study phase for

included subjects (n=527).

CAGE score (%)

Phase 0 1 2/3

3 88.6 11.4 0

5 89·2 10.8 0

7 88.7 11.3 0

9 88.5 11.5 0

11 88.8 11.2 0

Table 4: Summary of CAGE (screen for alcohol dependence) scores.

Mean (S.D.)

Right hippocampal volume (unadjusted, mm3) 3474 (433)

Left hippocampal volume (unadjusted, mm3) 3368 (444)

Right hippocampal volume (adjusted as % of ICV) 2.42 (0.30)

Left hippocampal volume (adjusted as % of ICV) 2.35 (0.32)

Table 5: Mean (S.D.) raw and adjusted hippocampal volumes as percentage of intracranial

volume (%ICV) extracted using FIRST for the sample.

Page 54 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

Figure 1: Partial regression plot from multiple linear regression with hippocampal volume

(% of ICV) as the dependent variable and average alcohol consumption as an independent

variable, corrected for age, sex, education, premorbid FSIQ, social class, Framingham Risk

Score, history of Major Depressive Disorder (SCID), exercise frequency, club attendance,

social visits, current psychotropic medication.

Unstandardised B (95% CI) Change in

hippocampal

volume (%

intracranial

volume)*

P value

Alcohol adjusted1 (all

cases)

-0.004 (-0.006 to -0.003) -0.15 0.001

Alcohol adjusted1

(excluding highest 3

drinkers)

-0.004 (-0.006 to -0.001) -0.13 0.003

* For every 10 unit increase in alcohol weekly

1 Analyses adjusted for age, sex, premorbid IQ, education, social class, Framingham Risk Score, smoking, history

of Major Depressive Disorder (SCID), exercise frequency, club attendance, social visits, current psychotropic

medication.

Table 6: Outlier analysis of multiple linear regression results, with hippocampal volume

(% of ICV) as the dependent variable and average alcohol consumption as an independent

variable.

Page 55 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nlyTest Beta (p value) 95% CI

Average alcohol1 -0.1 (0.03) -0.001 to 0.00

Average alcohol2 -0.08 (0.09) -0.001 to 0.00

1 Adjusted age, sex, education, FSIQ 2 As in 1 and additionally corpus callosum MD

Table 7: Results from multiple linear regression analysis, with decline in lexical fluency

(slopes, Phase 3 to time of scan) as the dependent variable, and average alcohol

consumption (weekly units) as an independent variable.

“Cognitive improvers”

Mean (S.D.)

“Cognitive decliners”

Mean (S.D.)

Mean difference (95%

CI)

FSIQ 120.5 (8.7) 117.0 (10.6) -3.6 (-5.6 to -1.6)

Table 8: Significant results from a t test of means between “cognitive improvers” (defined

as those with positive slopes on short-term memory tests over time) and “cognitive

decliners” (defined as those with negative slopes on short-term memory tests over time).

A. Semantic decline over study

Page 56 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

B. Lexical decline

C. Memory decline over study

Page 57 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


nly

D. Hippocampal volume (%ICV)

Figure 2: Partial regression plots for multiple regression analyses with the following as

dependent variable: A – semantic fluency decline over study (slopes from mixed effects

model), B – lexical fluency decline over study, C – memory decline over study, D –

hippocampal volume (extracted from FIRST and as %ICV).

Page 58 of 58


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

confidential: for review only - bmj€¦ · confidential: for review only 5 introduction alcohol...

Documents