Eur J Epidemiol
DOI 10.1007/s10654-017-0225-3
META-ANALYSIS
Alcohol consumption and dementia risk: a dose–response
meta-analysis of prospective studies
Wei Xu1 • Huifu Wang2 • Yu Wan2 • Chenchen Tan2
Jieqiong Li2 • Lan Tan1,2 • Jin-Tai Yu2
Received: 20 October 2016 / Accepted: 7 January 2017
Ó Springer Science+Business Media Dordrecht 2017
Abstract It is widely believed that light-to-moderate
alcohol intake may protect against dementia while excessive
drinking may instead increase the risk. Nonetheless, these
findings need cautious interpretations due to varying
methodologies and lack of standard definition, which hin-
dered our transferring into preventative practice. The
objective of this study is to investigate the potential dose–
response association between alcohol consumption and risk
of dementia. A systematic search was conducted in elec-
tronic databases to identify relevant studies. Risk estimates
were combined using a random-effect model. Eleven studies
with 73,330 participants and 4586 cases for all-cause
dementia (ACD), five studies with 52,715 participants and
1267 cases for Alzheimer’s dementia (AD) and four studies
with 49,535 participants and 542 cases for vascular dementia
were included. We observed a nonlinear association between
alcohol consumption and ACD risk (pnonlinearity \ 0.05). The
alcohol dose associated with lower risk of dementia was
confined to at most 12.5 g/day, with the risk hitting bottom
(RR & 0.9) at roughly 6 g/day. Of note, the ACD risk
seemed to be elevated (&10%) when the dose surpasses
Electronic supplementary material The online version of this
article (doi:10.1007/s10654-017-0225-3) contains supplementary
material, which is available to authorized users.
& Lan Tan
dr.tanlan@163.com
& Jin-Tai Yu
yu-jintai@163.com
1
College of Medicine and Pharmaceutics, Ocean University
of China, Qingdao, China
2
Department of Neurology, Qingdao Municipal Hospital,
Qingdao University, Qingdao, China
•
certain levels: 23 drinks/week or 38 g/day. For the alcohol
type, recommendation for wine is prioritized. The subgroup
analysis further indicated that the effect of alcohol may be
greater in younger adults (\60 years old) with regard to
fighting against dementia. Modest alcohol consumption
(B12.5 g/day) is associated with a reduced risk of dementia
with 6 g/day of alcohol conferring a lower risk than other
levels while excessive drinking (C38 g/day) may instead
elevate the risk.
Keywords Alcohol
Dementia
Meta-analysis

Dose–response
Introduction
Dementia is one common neurodegenerative disease with
complicated etiology underpinned by genetic and envi-
ronmental components. In the past decade, the prevention
of dementia has been increasingly catching attention of the
field. As a globally consumed beverage, alcohol is one of
numerous modifiable risk factors of dementia. It is widely
believed that light-to-moderate alcohol intake may protect
against dementia while excessive drinking may instead
increase the risk for adults of all ages, which belief stems
from previous epidemiological findings [1–8]. Nonetheless,
these findings need cautious interpretation because of
varying methodologies and lack of standard definition
(such as heterogeneous alcohol categorization and unit) [9],
which hindered our transferring into preventative practice
against dementia. Therefore, this study is aimed to better
quantify the association between alcohol use and dementia
to figure out the dose range defining the ‘‘light-to-moder-
ate’’ as well as ‘‘excessive’’ alcohol intake that significantly
influence risk of dementia.
123

Methods
Search strategy
Following the proposals by the Meta-analysis Of Observational
Studies in Epidemiology (MOOSE) Group and the PRISMA
2009 guidelines [10, 11], we searched electronic databases
(PubMed, EMBASE, and specific databases in Ovid) from
inception to October 7, 2016 using the following keywords,
among others: alcohol; ethanol; drink; dementia, Alzheimer,
prospective, cohort (full search terms available in the supple-
mentary Table 1). No restrictions were imposed. Bibliogra-
phies of eligible studies and relevant meta-analyses were hand-
searched for potential missing studies (Fig. 1).
Selection criteria
Studies were included if they simultaneously met the fol-
lowing criteria: (1) was prospective cohort or prospective
nested case–control studies; (2) investigated the association
between dementia (All-cause dementia [ACD] or Alzhei-
mer’s disease [AD] or vascular dementia [VD]) and alcohol
intake; (3) alcohol intake was categorized into C3 levels; (4)
reported adjusted level-specific relative risk (RR), 95% con-
fidence interval (CI) and, for the dose–response analysis, the
level-specific case number and person-years or sufficient data
for driving these numbers. Additionally, if multiple articles
were published based on the same cohort, we chose that with
longer follow-up or a larger sample size. Two investigators
independently made the inclusion decisions and any contro-
versies were resolved by consensus with a third reviewer.
Data extraction and quality evaluation
From each study, we extracted the first author, publication year,
cohort name, region, gender (male, female or combined), mean
age or age range, mean follow-up duration and rate, case
number, sample size and person-years stratified by alcohol
dose, diagnosis criteria, method of assessing alcohol intake,
alcohol dose and unit [including grams per day (g/day), drinks
per week (drinks/week), and frequency-times per week (times/
week)], adjusted confounders, and multivariable-adjusted
effects (RR and 95% CI) (Supplementary file-1). The study
quality was evaluated with the Newcastle-Ottawa Quality
Assessment Scale, which allowed a total score of up to 9 points
and only studies with C8 points were included in the present
meta-analysis.
Statistical analysis
Two types of meta-analyses were performed using random
effect model. Firstly, the alcohol intake was qualitatively
123
W. Xu et al.
defined as light (\7 drinks/week), light-to-moderate (\14
drinks/week), moderate (7–14 drinks/week), moderate-to-
heavy ([7 drinks/week), and heavy ([14 drinks/week)
based on previous meta-analyses [12–14]. Accordingly, we
combined the risk estimates reflective of the same category
in the same study using fixed model and then those in
different studies using random model. The combined
results for qualitative categories were compared to observe
the variation trend of dementia risk based on alcohol dose.
Next, a dose–response analysis was performed for alcohol
intake by amount (drinks/week and g/day) and frequency
(times/week). The method named two-stage generalized
least squares regression (GLST) for point and interval
estimation of relative risk for log-linear exposure–response
relations in meta-analyses of published ordinal categorical
exposure–response data has been proposed by Greenland
and Longnecker (called GL method) [15] and it has been
proved by Orsini et al. [16] that the differences between the
results of meta-analyses of summarized published data
using the GL method and a meta-analysis of individual
original data were negligible, even when there was evi-
dence of substantial confounding.
More specifically, at the first stage, we examined a
potential nonlinear association between alcohol intake and
dementia using study-specific restricted cubic spline mod-
els with four knots at fixed percentiles (5, 35, 65, and 95%)
of the exposure distribution. At the second stage, the study-
specific estimates (standard error) and the variance/co-
variance matrix that had been estimated within each study
were combined. The overall significance of the curve was
examined by testing the joint effect of the spline transfor-
mations. A nonlinear relationship was explored by testing
the null hypothesis that the regression coefficients of the
spline transformations were all equal to zero, as described
previously [17]. More details regarding the statistical
background was available in [15, 16].
As the reference category is supposed to be the least
exposure and the alcohol unit unified in quantitative anal-
ysis, data were transformed in advance. For studies wherein
the reference group was not the lowest category, we
regarded the lowest category as the reference and recal-
culate the effect size using the method by Orsini et al. [18].
Because the alcohol units in included studies were not
unified, we conducted separate quantitative analysis
according to the unit, including frequency (times/week)
and amount (g/day or drinks/week). Among the studies
included, Handing et al. [1] reported that the median fre-
quency for individuals drinking C12 g/day was about five
times a week, according to which we can conduct mutual
transformation between time/week and g/day. When a
study reported alcohol consumption in drinks/week, we
assumed that one drink contains 12 g of alcohol and
 Table 1 Characteristics of studies included in the dose–response analysis for dementia
N Author; year; Age Cohort Diagnosis Adjusted confounders Alcohol Stratified by alcohol category Alcohol Alcohol Multivariable-
region; (mean ± SD sample for criteria assessment unit exposure adjusted RR
Cohort or range) and analysis; Sample Incident Person- (95% CI)
name; study sex sample size case years
type (female%) at source;
baseline follow-up
time (mean or
range) and
rate

1 Paganini-Hill; 93 ± 2.6 547; DSM-IV Age, sex and education Questionnaire 208 87 624 Drinks/day Abstainers 1 (reference)
2016; USA; (90–103); community- criteria
The NA based; 3
90 ? Study; (mean); 96%
PCS
260 118 780 \2 drinks/day 0.97 (0.73–1.28)
79 41 147 C2 drinks/day 1.09 (0.75–1.58)
2 Langballe; 42–83 (born 40,435; ICD-10 Age, sex, years of Questionnaire 5688 182 56,652 Frequency Abstainers 1.15 (0.92–1.43)
2015; between population- education, hypertension,
Norway; 1905 and based; 9.96 obesity, smoking, and
HUNT1Q2 1946); C (mean); 27 symptoms of depression
study; PCS (51.2) (maximum);
98%
18,900 529 188,244 0 times/2 weeks, 1.12 (0.95–1.32)
but not abstainers
11,182 242 111,373 1–4 times/2 weeks 1 (reference)
2400 69 23,904 5 ? times/2 weeks 1.40 (1.07–1.84)
3 Handing; 54.2 ± 5.9 12,326; ICD Age, sex, education, Questionnaire 2873 518 28,730 g/day and 0 g/day 1.05 (1.00–1.11)
2015; (48.3–60.1); population- smoking status, physical frequency
Sweden; C (55.5) based; 43 activity, and time-
STR; PCS (maximum); varying diabetes and
Alcohol consumption and dementia risk: a dose–response meta-analysis of prospective studies

74% hypertension
5992 964 59,920 1.4 g/day 1 (reference)
2124 316 21,240 7.7 g/day 0.98 (0.92–1.04)
1037 122 10,370 16.1 g/day 1.10 (1.01–1.19)
300 38 3000 34.6 g/day 1.18 (1.01–1.36)
4 Zhou; 2014; 67.4 ± 4.7 2959; DSM-IV Age, BMI, education, Questionnaire 765 91 5355 Frequency Occasional 1 (reference)
China; (62.7–72.1); population- criteria APOE4, vascular risk drinking
Chong Qing M (0) based; 7; 93% factors
City; PCS
491 51 3437 Monthly drinking 0.87 (0.63–1.20)
402 41 2814 Weekly drinking 0.85 (0.60–1.21)
1301 174 9107 Daily drinking 1.12 (0.89–1.42)

123
 Table 1 continued
N Author; year; Age Cohort sample Diagnosis Adjusted confounders Alcohol Stratified by alcohol category Alcohol Alcohol exposure Multivariable-
region; Cohort (mean ± SD or for analysis; criteria assessment unit adjusted RR (95%
Sample Incident Person-

123
name; study range) and sex sample source; CI)
type (female%) at follow-up time size case years
baseline (mean or range)
and rate

5 Jarvenpaa; 49.4 ± 5.9 554; TELE Age, sex, education and Structured 80 36 2000 Drinks/ Abstainers 1 (reference)
2005; (43.5–55.3); population- score \ 16 binge drinking interview week
Finland; C (55) based; 25;
FTC; PCS 72%
139 43 3475 B3 1.0 (0.5–2.0)
66 16 1650 3–7 for women or 0.5 (0.1–1.4)
3–14 for men
16 6 400 More than a drink a 0.8 (0.2–3.3)
day for women
and more than 2
for men
6 Anttila; 2004; 48.3 ± 4.8 1018; DSM-IV Age, sex, education, follow Questionnaire 300 14 7140 Frequency Lifetime abstainers 0.91 (0.39–2.14)
Finland; (43.5–53.1); population- criteria up time, BMI, total
CAIDE C (62) based; 23 serum cholesterol, SBP,
study; PCS (mean); 70% DBP, smoking, history of
myocardial infarction,
and stroke.
423 17 9983 Less than once a 1 (reference)
month
295 17 6815 Several times a 1.44 (0.66–3.15)
month
7 Mukamal; 65 (?); C 746; DSM-IV Age, sex, race, APOE4, Questionnaire 274 151 1644 Drinks/ Abstainers 1 (reference)
2003; USA; (58.4) community- criteria education, income, week
CHS; NCCS based; 6 marital status, estrogen
(mean); NA replacement treatment,
smoking, DM-2, BMI,
total cholesterol level,
atrial fibrillation, heart
disease, stroke, TIA,
physical activity
126 53 756 \1 drinks/week 0.65 (0.41–1.02)
105 33 630 1–6 drinks/week 0.46 (0.27–0.77)
57 25 342 7–13 drinks/week 0.69 (0.37–1.31)
42 24 252 C14 drinks/week 1.22 (0.60–2.49)
8 Truelsen; 65 (?); C 1709; DSM-III-R Age, sex, education, stroke, Questionnaire 428 23 1284 Drinks/ \1 drinks/week 0.87 (0.46–1.64)
2002; (62.1) population- criteria cohabitation status, week
Denmark; based; 3 income, SBP, and current
CCHS; (1991–1994); smoking status
NCCS NA
W. Xu et al.
 Table 1 continued
N Author; year; Age Cohort Diagnosis Adjusted confounders Alcohol Stratified by alcohol category Alcohol Alcohol Multivariable-
region; (mean ± SD sample for criteria assessment unit exposure adjusted RR
Cohort name; or range) and analysis; Sample Incident Person- (95% CI)
study type sex sample size case years
(female%) at source;
baseline follow-up
time (mean
or range)
and rate

671 27 2013 1–7 drinks/week 1 (reference)

331 12 993 8–14 drinks/week 0.81 (0.39–1.72)
125 11 375 15–21 drinks/week 1.74 (0.74–4.07)
154 10 462 C22 drinks/week 1.29 (0.53–3.15)
9 Ruitenberg; 55 (?); C (59) 5395; DSM-III-R Age, sex, BMI, SBP, Questionnaire 1113 62 6678 Drinks/ None (including 1 (reference)
2002; population- criteria diabetes, smoking, and week less than twice a
Netherlands; based; 6 education. month)
RS; PCS (mean);
99.7%
1156 44 6936 \1 per week 0.82 (0.56–1.22)
1518 48 9108 [1 drink per week 0.75 (0.51–1.11)
but \1 per day
1443 38 8658 1–3 drinks per day 0.58 (0.38–0.90)
165 5 990 C4 drinks per day 1.00 (0.39–2.59)
10 Espeland; 65–79; W 4461; WHIMS Age, no. of years since Questionnaire 1345 24 5649 Drinks/day None 1 (reference)
2005; USA; (100) population- adjudication menopause, education,
WHIM based; 4.2 criteria ethnicity, family income,
Study; PCS (mean); smoking status, BMI,
NA hypertension status, prior
cardiovascular disease,
diabetes, statin use,
Alcohol consumption and dementia risk: a dose–response meta-analysis of prospective studies

aspirin use, and prior

hormone therapy.
2500 34 10,500 \1 drinks/day 0.76 (0.41–1.40)
616 3 2587 C1 drinks/day 0.28 (0.08–0.99)

NA not accessible, SBP systolic blood pressure, DBP diastolic blood pressure, BMI body mass index, PCS prospective cohort study, NCCS nested case–control study, CAIDE cardiovascular risk
factors, aging and dementia, CHS Cardiovascular Health Study, WHIM Women’s Health Initiative Memory, CCHS Copenhagen City Heart Study, RS Rotterdam Study, FTC Finnish Twin
Cohort, STP Swedish Twin Registry

123

Fig. 1 Flow diagram of the literature search and studies selection process
converted it into g/day or vice versa, as proposed by pre-
vious meta-analyses [12, 14, 19–22].
For those eligible for dose–response analysis, the median
or mean alcohol intake for each category was assigned to
each corresponding RR. When unavailable, we assigned the
midpoint of the upper and lower boundaries in each category
as the mean PA level. For studies with an open-ended upper
boundary (e.g. [2 drinks/day); we multiplied the reported
upper boundary by 1.25 and used this value (17.5 drinks/
week in the example). The heterogeneity among studies was
assessed by Q test and I2 statistic [23]. Heterogeneity was
confirmed with a significance level of p \ 0.05 and I2 values
123
W. Xu et al.
with cutoff of 30 and 50% are considered to indicate low,
moderate and high heterogeneity, respectively. Publication
bias was evaluated using the Egger test, and where statisti-
cally significant bias was found, the trim and fill method was
used to adjust it. Additionally, we conducted subgroup
analysis according to age (\60 and [60) because the influ-
ence of alcohol might vary depending on the age [9]. Finally,
we explored the association stratified by APOE4 status and
alcohol type (wine, beer and liquor). All statistical analyses
were conducted using Stata version 12.0 (StataCorp, College
Station, TX, USA), with two-tailed p \ 0.05 for statistical
significance.
Alcohol consumption and dementia risk: a dose–response meta-analysis of prospective studies
Fig. 2 Results of the qualitative analysis for all-cause dementia (ACD), Alzheimer’s dementia (AD) and vascular dmentia (VD). A nonlinear
relationship was revealed for ACD, AD and VD
Results
Figure 1 shows the results of literature searching and
selection. A total of 3806 papers were found after dedu-
plication. 3772 were excluded after reviewing the titles and
abstracts, leaving 34 papers with full-text available.
Another 14 potential papers were further identified from
the reference of relevant reviews. Finally, 18 were further
excluded (Supplementary Table 2) and a total of sixteen
studies (11 for ACD, 5 for AD and 4 for VD) were included
for qualitative analysis and fifteen studies for quantitative
analysis [10 for ACD (Table 1), 5 for AD and 4 for VD].
Description of studies included in qualitative
and quantitative synthesis
In qualitative synthesis, eleven studies (seven in Europe,
three in USA, and one in China) with 73,330 participants
for ACD, five studies (three in Europe and one in USA and
one in China) with 52,715 participants for AD and four
studies (two in Europe, one in USA and one in China) with
49,535 participants for VD were included, respectively.
During follow-up (3–25 years for ACD, 3–9.96 years for
AD, and 6–9.96 years for VD), 4586, 1267 and 542 non-
demented participants at baseline were diagnosed as ACD,
AD and VD, respectively. In dose–response analysis, ten
studies (seven for amount and four for frequency of alcohol
intake) with 70,150 participants for ACD (Table 1), four
studies (two for amount and two for frequency) with
49,535 participants for AD, and four studies (two for
amount and two for frequency) with 49,535 participants for
VD were included. During follow-up (3.9–28.3 years for
ACD), 4369 non-demented participants at baseline were
diagnosed as ACD. The reporting quality of the included
studies was high based on NOS rating. Two studies [24, 25]
were excluded because of lower quality (Supplementary
Table 3).
Qualitative analysis
In qualitative analysis, we observed a trend indicating a
U-shaped relationship between alcohol consumption and
dementia (Fig. 2): An inverse association of ACD with
light, light-to-moderate or moderate drinking was revealed
while the summary RR for heavier dose marched upwards
despite with no statistical significance (Fig. 2a) Similar
results were also indicated for AD (Fig. 2b). However, we
did not identify significant association for VD despite with
a nonlinear shape (Fig. 2c). Egger’s regression test pro-
vided no evidence of substantial publication bias.
Further, we conducted the qualitative analysis for ACD
according to the alcohol type (wine [1, 2, 6, 26–28], beer
[2, 6, 26–28] and liquor [1, 2, 26–28]) (Table 2). The
exposure can be categorized into ‘‘current versus never
drinker’’, ‘‘light-to-moderate drinker versus none’’, and
‘‘highest versus lowest’’. It seemed that the protective
effects from alcohol only existed for wine (alcoholic bev-
erage made from grapes or other fruits/vegetables) con-
sumption: current drinker (RR 0.67; 95% CI 0.48–0.94) or
light-to-moderate drinker (RR 0.58; 95% CI 0.39–0.87).
Further, we observed an elevated risk for highest versus
lowest consumption of beer (RR 1.84; 95% CI 1.01–3.34).
We cannot explore the quantitative relationship for alcohol
type or the qualitative association between alcohol type and
other dementia types because of the constrained data.
Also, we explored whether the association between
alcohol intake and ACD was influenced by APOE4 status
(Table 3). A total of three studies [2, 3, 29] reported the
effect size stratified by APOE4 status, among which the
case number in study by Kivipelto et al. [29] was too small
and thus excluded. We found a roughly 40% reduction of
dementia risk for light drinking or light-to-moderate
drinking for APOE4 carriers. Otherwise, the non-signifi-
cant association for APOE4 non-carriers should be inter-
preted with caution due to a massive heterogeneity.
123
 W. Xu et al.

Table 2 the association between all-cause dementia and alcohol intake, stratified by alcohol type
Alcohol type Exposure Number of studies included Combined results p value I2 (%)

Wine Current versus none 2 0.67 (0.48–0.94) 0.2 39

Light-to-moderate drinker 3 0.58 (0.39–0.87) 0.196 38.6
Highest versus lowest 4 1.01 (1.00–1.02) 0.583 0.0
Beer Current versus none 2 1.04 (0.78–1.40) 0.291 10.3
Light-to-moderate drinker 3 1.59 (0.75–3.41) 0.003 82.6
Highest versus lowest 3 1.84 (1.01–3.34) 0.979 0.0
Liquor/spirits Current versus none 1 1.16 (0.80–1.69) NA NA
Light-to-moderate drinker 3 0.93 (0.74–1.18) 0.759 0.0
Highest versus lowest 3 1.16 (0.73–1.84) 0.931 0.0

Table 3 The association

Status Exposure Combined results p value I2 (%)
between all-cause dementia
(ACD) and alcohol intake, APOE4 (?) Light 0.59 (0.41–0.84) 0.904 0.0
stratified by APOE4 status
according to two studies [2, 3] Light-to-moderate 0.60 (0.44–0.80) 0.548 0.0
Moderate 0.71 (0.31–1.60) 0.232 29.9
APOE4 (-) Light 0.70 (0.32–1.55) 0.005 87.3
Light-to-moderate 0.67 (0.38–1.16) 0.023 80.7
Moderate 0.57 (0.35–0.95) 0.715 0.0

Dose–response analysis
Among the 10 prospective studies included in the dose–
response analysis for ACD, 6 studies [2–4, 28, 30, 31] used
drinks/week, 3 studies used times/week [5, 7, 32] and only
one study [1] used g/day as alcohol unit. Consistent with
the qualitative analysis, we uncovered a significantly
nonlinear association between alcohol consumption and
ACD risk (Fig. 3) (pfor non-linearity \ 0.05). Several cutoffs
were of great value especially in defining the alcohol dose
associated with the dementia’s risk. Specifically, we found
the alcohol dose associated with lower risk was roughly
situated between 0 and 7.5 drinks/week (Fig. 3a) or
12.5 g/day (Fig. 3b) or 2 times/week (Fig. 3c) with the
lowest risk (RR & 0.9) corresponding to roughly 4 drinks/
week (Fig. 3a), 6 g/day (Fig. 3b), and 1 times/week
(Fig. 3c). On the other hand, the ACD risk was signifi-
cantly elevated when the alcohol consumed surpasses
certain cutoff: 23 drinks/week (Fig. 3a) or 38 g/day
(Fig. 3b). The subgroup analysis further indicated that it
might be favorable for adults aged \60 years old to drink
less than 6 drinks/week (Fig. 4a) while for those aged
[60 years old, the safe drinking frequency for dementia
prevention was B2 times/week (pfor nonlinearity \ 0.05).
Among the four studies eligible for dose–response
analysis for AD, two studies [2, 3] reported drinks/week
while another two studies [5, 7] reported times/week. We
thus cannot conduct the quantitative analysis because
number of eligible studies is small. The quantitative
analysis for VD was hindered for similar reason. In spite
of this, some clues can still be implied: Mukamal et al. [2]
reported elders aged [65 years old who consumed alco-
hol B6 drinks/week would have magnificent lower risk of
AD. This amount is similar to that concluded by the
present analysis for ACD, which is reasonable because
AD is the most common type of ACD. Otherwise,
Ruitenberg et al. [3] proposed one to three drinks per day
as the optimal dose against VD. Zhou et al. [7] found that
daily drinking (frequency) can markedly increase the risk
of AD or VD.
The influence of ‘‘optimal’’ dose range for dementia
on the overall health
As mentioned above, alcohol consumption was also asso-
ciated with risk of diseases other than dementia. From the
perspective of overall health, assessing whether the optimal
dose range (B12.5 g/day) for dementia is still protective
for other diseases is pretty necessary. Thus, we searched
the PubMed database using the keywords ‘‘dose–response’’
AND ‘‘meta-analysis’’ AND ‘‘alcohol’’, and found eleven
studies exploring dose–response relationship with other
diseases, including heart disease, DM-2, cancer, stroke, and
hypertension. By comparison, we identified drinking
B12.5 g/day was still protective for all diseases above
except some cancers, which is also the major source of
dispute concerning whether alcohol consumption should be
regarded as a safe approach to preventing dementia.
(Supplementary Table 3).

123
Alcohol consumption and dementia risk: a dose–response meta-analysis of prospective studies
Fig. 3 Dose–response meta-analyses for all-cause dementia (ACD)
and alcohol intake by drinks/week (a), grams/day (b) and times/week
(c). Modest alcohol consumption (B7.5 drinks/week or B12.5 g/day
or B2 times/week) is associated with a reduced risk of dementia with
4 drinks/week, 6 g/day or 1 time/week of alcohol conferring a lower
Fig. 4 Dose–response meta-analyses for all-cause dementia (ACD)
and alcohol intake in population aged \60 and [60 years old. The
effect of alcohol may be greater in the adults (\60 years old) (a) than
Discussion
Our results revealed an optimal dose range of alcohol
consumption (B12.5 g/day), which was associated with a
lower risk of dementia, and also raised the alarm that
excessive drinking would possibly advance the risk. It is
justifiable for the inconsistence between findings based on
amount (g/day or drinks/week) and that based on frequency
(times/week) as they are not closely related and we believe
that the former is more informative. On the other hand,
though the lowest effect size is modest (RR around 0.9), it
was noteworthy that the estimate by relative risk only
indicated the average level for the study population
included in the present analysis and it thus may be various
for specific individual. Also, the effect size may vary
risk than other levels while heavy drinking (C23 drinks/week or
C38 g/day) would significantly elevate the risk. Black solid lines
represent relative risk and black dashed lines represent 95%
confidence intervals
in older adults (b) with regard to fighting against dementia. Black
solid lines represent relative risk and black dashed lines represent
95% confidence intervals
depending upon characteristics of individuals, such as age
(Fig. 4), gender, and alcohol type consumed, etc.
Alcohol consumption may have both neurotoxic and
neuroprotective effects [9], depending upon its dose con-
sumed or its type [33]. Light-to-moderate drinking has
been associated with improved lipid profile [34], larger
total brain volume [35], less severe white matter lesions
and brain infarcts [36], lower risk of metabolic syndrome
[37], reduced arterial stiffness [38–41], and lowered
inflammation markers [42, 43], all of which may contribute
to a lower risk of dementia [8]. On the other hand, it has
been showed that for heavy drinkers ([30 g/day), an
increment of 10 g/day increases systolic blood pressure on
average by 1–2 mmHg and diastolic blood pressure by
1 mmHg. Also, it has been estimated that about 7–11% of
123

hypertension (C160/95 mmHg) in men are due to alcohol
intake C40 g/day while the percentage is smaller for
women because of their much lower alcohol intake [44].
According to our previous study [8], hypertension is a
significant risk factor for dementia with unknown mecha-
nisms [45]. It is thus reasonable to infer that heavy drinker
may be more vulnerable to suffer from hypertension [46],
which would elevate the dementia’s risk. Similarly,
excessive alcohol consumption was associated with higher
risk of multiple comorbidities other than dementia, such as
type-2 diabetes mellitus [13, 47] and cardiovascular events
[14, 19, 48, 49], all of which might elevate the risk of
dementia [8].
Different types of beverage may confer different effects
and our study indicated that modest wine consumption
might help decrease the dementia risk. There are several
potential mechanisms for protection against dementia by
wine, such as a greater cardiovascular benefit due to higher
amount of polyphenolic compounds like resveratrol (trans-
3,4,5-trihydroxystilbene) which was related to maintaining
endothelial function and antioxidant effects [50, 51].
Otherwise, it has been suggested that drinking beer, but not
wine or liquor, associates with lower burden of amyloid
beta aggregation in the brain. Our analysis did not reveal an
inverse association between light-to-moderate beer drink-
ing and dementia’s risk, yet with a high heterogeneity
(82.6%). Nevertheless, Excessive drinking either wine or
beer was positively associated with dementia’s risk
(Table 2).
The primary strength of our meta-analysis lies in its
dose–response design, which provides better quantification
of the associations between specified amounts of alcohol
and risk of dementia. A dose–response meta-analysis
should be the first choice when performing a systematic
review, rather than summarizing the risk estimates based
merely on the comparison of the extreme categories
(highest versus lowest), which could vary magnificently
among studies. Another strength is that only prospective
studies were included. The prospective design minimized
the potential influence of recall and selection bias, which
might be an inevitable issue in retrospective design.
There are several limitations in our study. First, the
dose–response modeling was based on limited, aggregated
epidemiological data and thus warrants further validation
using more advanced approach, such as Mendelian ran-
domization. Second, the influence of alcohol may vary
depending on the amount consumed, duration of drinking
or abstinence, alcohol type, drinking pattern or frequency,
gender, environmental factors, and the genetic predisposi-
tion of the individual [9]. We performed qualitative but not
quantitative analysis for most factors above because of the
heterogonous methodologies or incomplete data. Another
limitation is the possible misclassification of alcohol
123
W. Xu et al.
consumption which was inevitable considering that alcohol
dose consumed was self-reported. Fourth, we failed to
conduct the dose–response analysis for AD and VD due to
the restriction of data. Nonetheless, the conclusions in our
study may also apply to AD given that it accounted for a
vast majority in dementia. Fifth, we cannot exclude the
potential influences of including former drinkers, who may
quit drinking due to underlying diseases and have a high
risk of dementia, in the reference group due to data
restrictions. A majority of researchers seemed to have
neglected this important issue, which warrants more
attention in the future study. Sixth, in considerations of the
inconsistence of the adjusted confounders in studies
included, we cannot exclude the potentially spurious
inverse association caused by some confounders.
Nonetheless, it is noteworthy that all studies controlled for
age, sex and education and a majority of them controlled
for others, such as APOE4 and cardiovascular conditions,
which would lower the risk to some extent.
Conclusion
In summary, Modest alcohol consumption (B12.5 g/day) is
associated with a reduced risk of dementia with 6 g/d of
alcohol conferring a lower risk than other levels while
heavy drinking (C23 drinks/week or C38 g/day) would
significantly elevate the risk.
Authors’ contributions WX, LT and JY designed the study. WX,
HW, YW and JL performed the literature search, the selection of
eligible articles and the data extraction. WX and HW analyzed the
data. WX, HW, JL and JY were involved in data interpretation. WX
and JY wrote the first draft of the manuscript. All authors approved
the final version of the submitted manuscript.
Compliance with ethical standards
Conflict of interest The authors declare that they have no competing
interests.
References
1. Handing EP, Andel R, Kadlecova P, Gatz M, Pedersen NL.
Midlife alcohol consumption and risk of dementia over 43 years
of follow-up: a population-based study from the Swedish Twin
Registry. J Gerontol Ser A Biol Sci Med Sci.
2015;70(10):1248–54.
2. Mukamal KJ, Kuller LH, Fitzpatrick AL, Longstreth WT Jr,
Mittleman MA, Siscovick DS. Prospective study of alcohol
consumption and risk of dementia in older adults. J Am Med
Assoc. 2003;289(11):1405–13.
3. Ruitenberg A, van Swieten JC, Witteman JC, et al. Alcohol
consumption and risk of dementia: the Rotterdam Study. Lancet.
2002;359(9303):281–6.
Alcohol consumption and dementia risk: a dose–response meta-analysis of prospective studies
4. Espeland MA, Gu L, Masaki KH, et al. Association between
reported alcohol intake and cognition: results from the Women’s
Health Initiative Memory Study. Am J Epidemiol.
2005;161(3):228–38. doi:10.1093/aje/kwi043.
5. Langballe EM, Ask H, Holmen J, et al. Alcohol consumption and
risk of dementia up to 27 years later in a large, population-based
sample: the HUNT study, Norway. Eur J Epidemiol.
2015;30(9):1049–56.
6. Weyerer S, Schaufele M, Wiese B, et al. Current alcohol con-
sumption and its relationship to incident dementia: results from a
3-year follow-up study among primary care attenders aged
75 years and older. Age Ageing. 2011;40(4):456–63. doi:10.
1093/ageing/afr007.
7. Zhou S, Zhou R, Zhong T, Li R, Tan J, Zhou H. Association of
smoking and alcohol drinking with dementia risk among elderly
men in China. Curr Alzheimer Res. 2014;11(9):899–907.
8. Xu W, Tan L, Wang HF, et al. Meta-analysis of modifiable risk
factors for Alzheimer’s disease. J Neurol Neurosurg Psychiatry.
2015;86(12):1299–306. doi:10.1136/jnnp-2015-310548.
9. Kim JW, Lee DY, Lee BC, et al. Alcohol and cognition in the
elderly: a review. Psychiatry Investig. 2012;9(1):8–16. doi:10.
4306/pi.2012.9.1.8.
10. Moher D, Liberati A, Tetzlaff J, Altman DG, Group P. Preferred
reporting items for systematic reviews and meta-analyses: the
PRISMA statement. Int J Surg. 2010;8(5):336–41. doi:10.1016/j.
ijsu.2010.02.007.
11. Stroup DF, Berlin JA, Morton SC, et al. Meta-analysis of
observational studies in epidemiology: a proposal for reporting.
Meta-analysis Of Observational Studies in Epidemiology
(MOOSE) group. JAMA. 2000;283(15):2008–12.
12. Wang YT, Gou YW, Jin WW, Xiao M, Fang HY. Association
between alcohol intake and the risk of pancreatic cancer: a dose–
response meta-analysis of cohort studies. BMC Cancer.
2016;16:212. doi:10.1186/s12885-016-2241-1.
13. Li XH, Yu FF, Zhou YH, He J. Association between alcohol
consumption and the risk of incident type 2 diabetes: a systematic
review and dose–response meta-analysis. Am J Clin Nutr.
2016;103(3):818–29. doi:10.3945/ajcn.115.114389.
14. Larsson SC, Orsini N, Wolk A. Alcohol consumption and risk of
heart failure: a dose–response meta-analysis of prospective
studies. Eur J Heart Fail. 2015;17(4):367–73. doi:10.1002/ejhf.
228.
15. Greenland S, Longnecker MP. Methods for trend estimation from
summarized dose–response data, with applications to meta-anal-
ysis. Am J Epidemiol. 1992;135(11):1301–9.
16. Orsini N, Li R, Wolk A, Khudyakov P, Spiegelman D. Meta-
analysis for linear and nonlinear dose–response relations: exam-
ples, an evaluation of approximations, and software. Am J Epi-
demiol. 2012;175(1):66–73. doi:10.1093/aje/kwr265.
17. Xu W, Tan L, Wang HF, et al. Education and risk of dementia:
dose–response meta-analysis of prospective cohort studies. Mol
Neurobiol. 2016;53(5):3113–23. doi:10.1007/s12035-015-9211-
5.
18. Orsini N. From floated to conventional confidence intervals for
the relative risks based on published dose–response data. Comput
Methods Progr Biomed. 2010;98(1):90–3. doi:10.1016/j.cmpb.
2009.11.005.
19. Yang Y, Liu DC, Wang QM, et al. Alcohol consumption and risk
of coronary artery disease: a dose–response meta-analysis of
prospective studies. Nutrition. 2016;32(6):637–44. doi:10.1016/j.
nut.2015.11.013.
20. Bagnardi V, Rota M, Botteri E, et al. Alcohol consumption and
site-specific cancer risk: a comprehensive dose–response meta-
analysis. Br J Cancer. 2015;112(3):580–93. doi:10.1038/bjc.
2014.579.
21. Larsson SC, Drca N, Wolk A. Alcohol consumption and risk of
atrial fibrillation: a prospective study and dose–response meta-
analysis. J Am Coll Cardiol. 2014;64(3):281–9. doi:10.1016/j.
jacc.2014.03.048.
22. Fedirko V, Tramacere I, Bagnardi V, et al. Alcohol drinking and
colorectal cancer risk: an overall and dose–response meta-anal-
ysis of published studies. Ann Oncol. 2011;22(9):1958–72.
doi:10.1093/annonc/mdq653.
23. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-
analysis. Stat Med. 2002;21(11):1539–58. doi:10.1002/sim.1186.
24. Lindsay J, Laurin D, Verreault R, et al. Risk factors for Alzhei-
mer’s disease: a prospective analysis from the Canadian Study of
Health and Aging. Am J Epidemiol. 2002;156(5):445–53.
25. Lemeshow S, Letenneur L, Dartigues JF, Lafont S, Orgogozo JM,
Commenges D. Illustration of analysis taking into account com-
plex survey considerations: the association between wine con-
sumption and dementia in the PAQUID study. Personnes Ages
Quid. Am J Epidemiol. 1998;148(3):298–306.
26. Mehlig K, Skoog I, Guo X, et al. Alcoholic beverages and inci-
dence of dementia: 34-year follow-up of the prospective popu-
lation study of women in Goteborg. Am J Epidemiol.
2008;167(6):684–91.
27. Deng J, Zhou DH, Li J, Wang YJ, Gao C, Chen M. A 2-year
follow-up study of alcohol consumption and risk of dementia.
Clin Neurol Neurosurg. 2006;108(4):378–83.
28. Truelsen T, Thudium D, Gronbaek M. Amount and type of
alcohol and risk of dementia: the Copenhagen City Heart Study.
Neurology. 2002;59(9):1313–9.
29. Kivipelto M, Rovio S, Ngandu T, et al. Apolipoprotein E epsilon4
magnifies lifestyle risks for dementia: a population-based study.
J Cell Mol Med. 2008;12(6B):2762–71. doi:10.1111/j.1582-4934.
2008.00296.x.
30. Paganini-Hill A, Kawas CH, Corrada MM. Lifestyle factors and
dementia in the oldest-old: the 90? study. Alzheimer Dis Assoc
Disord. 2016;30(1):21–6.
31. Jarvenpaa T, Rinne JO, Koskenvuo M, Raiha I, Kaprio J. Binge
drinking in midlife and dementia risk. Epidemiology.
2005;16(6):766–71.
32. Anttila T, Helkala EL, Viitanen M, et al. Alcohol drinking in
middle age and subsequent risk of mild cognitive impairment and
dementia in old age: a prospective population based study. BMJ.
2004;329(7465):539.
33. Kok EH, Karppinen TT, Luoto T, Alafuzoff I, Karhunen PJ. Beer
drinking associates with lower burden of amyloid beta aggrega-
tion in the brain: Helsinki sudden death series. Alcohol Clin Exp
Res. 2016;40(7):1473–8.
34. Vu KN, Ballantyne CM, Hoogeveen RC, et al. Causal role of
alcohol consumption in an improved lipid profile: the
atherosclerosis risk in communities (ARIC) study. PloS ONE.
2016;11(2):e0148765.
35. Gu Y, Scarmeas N, Short EE, et al. Alcohol intake and brain structure
in a multiethnic elderly cohort. Clin Nutr. 2014;33(4):662–7.
36. 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. Am J Clin Nutr. 2004;80(4):992–7.
37. Fujita N, Takei Y. Alcohol consumption and metabolic syn-
drome. Hepatol Res. 2011;41(4):287–95.
38. Mattace-Raso FU, van der Cammen TJ, van den Elzen AP, et al.
Moderate alcohol consumption is associated with reduced arterial
stiffness in older adults: the Rotterdam study. J Gerontol Ser A
Biol Sci Med Sci. 2005;60(11):1479–83.
39. Hougaku H, Fleg JL, Lakatta EG, et al. Effect of light-to-mod-
erate alcohol consumption on age-associated arterial stiffening.
Am J Cardiol. 2005;95(8):1006–10. doi:10.1016/j.amjcard.2004.
12.051.
123

40. Damiani IT, Gagliardi RJ, Scaff M. The influence of ethanol in
alcoholic beverages in extracranial carotid arteries atherosclero-
sis. Arq Neuropsiquiatr. 2004;62(4):1022–6.
41. Nakanishi N, Kawashimo H, Nakamura K, et al. Association of
alcohol consumption with increase in aortic stiffness: a 9-year
longitudinal study in middle-aged Japanese men. Ind Health.
2001;39(1):24–8.
42. Shai I, Rimm EB, Schulze MB, Rifai N, Stampfer MJ, Hu FB.
Moderate alcohol intake and markers of inflammation and
endothelial dysfunction among diabetic men. Diabetologia.
2004;47(10):1760–7.
43. Estruch R, Sacanella E, Badia E, et al. Different effects of red
wine and gin consumption on inflammatory biomarkers of
atherosclerosis: a prospective randomized crossover trial. Effects
of wine on inflammatory markers. Atherosclerosis.
2004;175(1):117–23.
44. Keil U, Swales JD, Grobbee DE. Alcohol intake and its relation
to hypertension. Cardiovasc Risk Factors. 1993;3(4):189–200.
45. Querfurth HW, LaFerla FM. Alzheimer’s disease. N Engl J Med.
2010;362(4):329–44. doi:10.1056/NEJMra0909142.
46. Taylor B, Irving HM, Baliunas D, et al. Alcohol and hyperten-
sion: gender differences in dose-response relationships deter-
mined through systematic review and meta-analysis. Addiction.
2009;104(12):1981–90. doi:10.1111/j.1360-0443.2009.02694.x.
123
W. Xu et al.
47. Knott C, Bell S, Britton A. Alcohol consumption and the risk of
type 2 diabetes: a systematic review and dose–response meta-
analysis of more than 1.9 million individuals from 38 observa-
tional studies. Diabetes Care. 2015;38(9):1804–12. doi:10.2337/
dc15-0710.
48. Mostofsky E, Chahal HS, Mukamal KJ, Rimm EB, Mittleman
MA. Alcohol and immediate risk of cardiovascular events: a
systematic review and dose–response meta-analysis. Circulation.
2016;133(10):979–87. doi:10.1161/CIRCULATIONAHA.115.
019743.
49. Zhang C, Qin YY, Chen Q, et al. Alcohol intake and risk of
stroke: a dose–response meta-analysis of prospective studies. Int J
Cardiol. 2014;174(3):669–77. doi:10.1016/j.ijcard.2014.04.225.
50. Riccioni G, Gammone MA, Tettamanti G, Bergante S, Pluchi-
notta FR, D’Orazio N. Resveratrol and anti-atherogenic effects.
Int J Food Sci Nutr. 2015;66(6):603–10. doi:10.3109/09637486.
2015.1077796.
51. Rakici O, Kiziltepe U, Coskun B, Aslamaci S, Akar F. Effects of
resveratrol on vascular tone and endothelial function of human
saphenous vein and internal mammary artery. Int J Cardiol.
2005;105(2):209–15. doi:10.1016/j.ijcard.2005.01.013.