International Journal of Impotence Research (2007) 19, 343–352
& 2007 Nature Publishing Group All rights reserved 0955-9930/07 $30.00
www.nature.com/ijir
REVIEW
Alcohol consumption and erectile dysfunction: meta-analysis
of population-based studies
JYW Cheng, EML Ng, RYL Chen and JSN Ko
Department of Psychiatry, Li Ka Shing Faculty of Medicine, University of Hong Kong, Hong Kong, China
Alcohol is long regarded as a risk factor for erectile dysfunction (ED), but epidemiological evidence
has been equivocal. We aimed to investigate the ED risk associated with various levels of alcohol
consumption by meta-analysis. We searched for population-based studies on ED through Medline,
PubMed, PsychInfo, and scanned through reference lists. Eleven cross-sectional studies were
included and analyzed with random effects model. We reviewed the results from one crosssectional study and two cohort studies. Regular alcohol consumption was negatively associated
with ED (odds ratio (OR) ¼ 0.79; 99% confidence interval (CI), 0.67–0.92; Po0.001). Consumption of
8 or more drinks/week significantly reduced the risk of ED (OR ¼ 0.85; 99% CI, 0.73–0.99; P ¼ 0.007),
but consumption of less alcohol (1–7 drinks/week) was not significant (OR ¼ 0.73; 99% CI, 0.44, 1.20;
P ¼ 0.101). Begg’s test and Egger’s test detected no significant publication bias. Our estimates (in
sensitivity analyses) were rendered nonsignificant when International Index of Erectile Function
definition was used and when statistical adjustment was made only for age. Meta-analysis of crosssectional studies yielded a protective association of alcohol on ED, but the two cohort studies did
not demonstrate any significant findings for alcohol consumption. More research is needed to
confirm whether alcohol is protective or is unrelated to ED development.
International Journal of Impotence Research (2007) 19, 343–352; doi:10.1038/sj.ijir.3901556;
published online 31 May 2007
Introduction
Alcohol is long regarded as a risk factor for erectile
dysfunction (ED), whether in textbooks,1,2 review
articles3,4 or clinical teachings. It was a long-held
empirical observation that acute alcohol intoxication increases sexual desire but inhibits sexual
performance (Macbeth, Act 2, Scene 3, ‘What three
things does drink especially provokey lechery, sir,
it provokes, and unprovokes; it provokes the desire,
but it takes away the performance’). The scientific
explanation is that alcohol is a central nervous
system depressant2, but it also leads to disinhibition
and increases sexual desire.
Epidemiological studies have demonstrated numerous risk factors for ED, such as age,5 variables
related to diabetes,6 depression,7 hypertension6 and
smoking.8 With more accumulating evidence, the
risk of alcohol consumption in ED seemed more
Correspondence: Dr JYW Cheng, Department of Psychiatry,
Li Ka Shing Faculty of Medicine, University of Hong Kong,
Hong Kong, China.
E-mail: [email protected]
Received 24 July 2006; revised 9 January 2007; accepted
24 February 2007; published online 31 May 2007
equivocal. ED has also been said to be the harbinger
of cardiovascular events.9 Endothelial dysfunction
has been hypothesized to result from cardiovascular
risk factors such as hypertension or diabetes, which
in turn leads to ED, myocardial infarction and
stroke.10 As evidence accumulates to support
the similarity in cardiovascular risk factor profiles11
of ED and cardiovascular diseases, it was fair
to hypothesize that alcohol may demonstrate a
J-shaped (or inverse) relationship with ED, with
moderate alcohol consumption conferring the lowest risk of ED. Furthermore, considering the scale of
the studies done on cardiovascular diseases and that
of ED, it was not surprising that perhaps ED studies
were simply not large enough to delineate the risk
of ED associated with alcohol consumption.
It was the aim of this paper to use meta-analysis
to pool the risk of regular alcohol consumption
associated with ED.
Methods
We searched for population-based studies on ED
published between 1990 to April 2006. We searched
through Medline, PubMed, PsychInfo, and scanned
through reference lists. We used keywords to search
Alcohol consumption and erectile dysfunction
JYW Cheng et al
344
for our disease of interest (for example, ED, impotence, sexual dysfunction) and the type of studies
that were of relevance (for example, prevalence,
epidemiology, incidence), and used the ‘AND’ search
string to narrow down the search to overlapping
studies. Whole texts were retrieved and read.
Studies were included if they were populationbased studies on ED, and provided age-adjusted ORs
for alcohol. If multivariate (including age-adjusted)
ORs were provided they were used for statistical
analysis in preference to the univariate age-adjusted
ORs. Since our preliminary search identified mainly
cross-sectional studies, we added the cross-sectional
design as an inclusion criteria.
No established quality assessment guidelines
were available on cross-sectional studies, and the
assessment instrument designed for prevalence
study by Prins et al.12 was adapted, expanded for
this study. We added to the quality assessment a
number of risk factors that were commonly adjusted
for in various population-based prevalence studies.
Note that the subscale score for ‘statistical adjustments for odds ratios’ was not the total number of
variables adjusted for in the original article, the
actual number adjusted for is found in Table 1 or
in the parentheses in Figures 1–3.
Two of the investigators (CJ and KJ) independently
rated and extracted data from each study, and any
disagreements were resolved through discussion
to obtain a consensus before including data in the
analysis.
Data were extracted according to a fixed protocol:
year of study, definition of ED, response rate, sample
size, OR, variables adjusted for OR. Two different
meta-analyses were formed, one that defined alcohol consumption as having two categories (yes or
no), and another that defined them as having three
categories (none, 1–7 drinks, 7 or more drinks). If an
included study used a three-category definition of
alcohol, a pooled estimate would be calculated for
the two categories, and the estimate would represent
‘above average alcohol consumption’ and be used
in the meta-analysis for the two-category analysis.
On the other hand, if a study used a two-category
definition, the resulting OR would not be included
in the meta-analysis for the three-category analysis.
Weighted average of ORs were calculated using
the inverse variance method for random effects
models. Adjusted ORs were converted to the
logarithmic scale for statistical calculation and
anti-logged for final results.
Representative data in the absence of publication
bias should (1) follow a symmetrical distribution
around the true value (for example, OR) and (2) the
variability of the data should increase as the
standard error of the data increases. Hence, when
representative data are plotted against their standard
errors, they should form a symmetrical funnel, and
any asymmetries in a funnel plot could be caused by
publication bias (unrepresentativeness). We used
International Journal of Impotence Research
Begg’s test and Egger’s test to test for publication
bias based on the symmetry assumption described.
Heterogeneity between studies and in the metaanalysis estimates was evaluated by the w2 test
(I2 indicates variation in OR attributable to heterogeneity and t2 the variance between the included
studies).
A wider 99% confidence interval (CI) was used for
each estimate to allow for multiple comparisons. All
statistical calculations were performed by STATA
version 8.0.
Results
Our initial search identified 221 articles, but only 11
cross-sectional studies13–22 documented the adjusted
OR for alcohol, another cross-sectional study23 and
a prospective cohort24 from the same study population provided adjusted relative risks (RR), one
cohort study25 provided adjusted OR. Only the 11
cross-sectional studies that provided adjusted
ORs will be included in the meta-analysis, since the
other two cohort studies were of different study
design, and the RR reported by the cross-sectional and
cohort studies were derived with a different statistical
method (Mantel–Haenszel rather than logistic regression). Relevant findings will be summarized below for
the three latter studies. An overview of the studies is
provided in Table 1. An overview of the quality
appraisal is shown in Tables 2 and 3. A significant
level of heterogeneity was found between studies
(Po0.001), and the random effects model was used
and its results are included below.
Two-category response model for alcohol
Regular alcohol consumption was negatively (protective effect) associated with ED (odds ratio (OR) ¼
0.79; 99% CI, 0.67–0.92; Po0.001) (heterogeneity
w2 ¼ 43.12, Po0.001; I2 ¼ 76.8%; t2-statistic ¼ 0.0243)
(Figure 1).
Three-category response model for alcohol
Having ‘1–7 drinks/week’ (OR ¼ 0.73; 99% CI, 0.44–
1.20; P ¼ 0.101) (heterogeneity w2 ¼ 44.34, Po0.001;
I2 ¼ 91.0%; t2-statistic ¼ 0.1607) (Figure 2) and
‘8 drinks or more/week’ (OR ¼ 0.85; 99% CI, 0.73–
0.99; P ¼ 0.007) (heterogeneity w2 ¼ 2.66, P ¼ 0.265;
I2 ¼ 24.7%; t2 statistic ¼ 0.0034) (Figure 3) were
negatively associated with ED.
Bias, influence and robustness
A funnel plot is shown in Figure 4. Begg’s test
(P ¼ 0.533) and Egger’s (P ¼ 0.157) test did not detect
significant publication bias.
Influence analysis for the random effects model
did not demonstrate any large changes to the
Table 1 An overview of the eligible studies
Reference Author, year
Country
Definition of ED
Sample
size
Age range
13
Shaeer et al.,
2003
Pakistan
Egypt
Nigeria
Single question derived from NIH consensus: ‘able to get and
keep an erection good enough for satisfactory intercourse.’
1814
14
Moreira et al.,
2002
Brazil
Single question derived from NIH consensus: ‘able to get and
keep an erection adequate for satisfactory intercourse.’
602
15
Akkus et al.,
2002
Turkey
Single question: ‘able to get and keep an erection adequate for
sexual intercourse.’
16
Bai et al., 2004 China
5
Nicolosi et al.,
2003
17
Millett et al.,
2006
18
Variables
adjusted for
ORalcohol
Alcohol
consumption
ORalcohol
(95% CI)
35–70
Yes/no
0.53 (0.39–0.74)
1
40–70
Yes/no
0.54 (0.37–0.87)
1
1982
40–70 þ
Yes/no
0.55 (0.36–0.85)
10
IIEF-5
2203
p30–X70
Yes/no
0.69 (0.56–0.84)
1
Brazil
Italy
Japan
Malaysia
Single question: ‘able to get and keep an erection adequate for
satisfactory sexual intercourse.’
2417
40–70
None
1–7/week
X8/week
1
0.74 (0.53–1.02)
0.73 (0.53–0.99)
9
Australia
Single question: ‘During the past year has there been a period
of one month or more when you trouble keeping an erection
when you wanted to?’
8367
16–59
Non-drinker
1–4/day
4 þ /day
1
0.36 (0.28–0.45)
0.89 (0.81–1.34)
6
Parazzini et al., Italy
2000
Single question: ‘ability to achieve and maintain an erection
sufficient for satisfactory sexual performance.’
2010
10–470
None
1–7
8–14
15–21
X22
19
Mak et al., 2002 Belgium
IIEF-15
20
Fedele et al.,
2000
40–70
Single question: ‘ability to achieve and maintain an erection
for satisfactory sexual performance.’
9756
p45–X66
Cho et al., 2003 Korea
IIEF-5
3501
20–X60
22
Terai et al., 2004 Japan
IIEF-5
2084
p39–X60
None
1–7
8–14
15–21
X22
1
0.56 (0.37–0.84)
1.29 (0.88–1.88)
0.9
0.9
1.0
0.9
1
(0.8–1.1)
(0.8–1.0)
(0.7–1.2)
(0.8–1.0)
3
1
2
Non or
1
light(o3/week)
Moderate (3–4/ 0.91 (0.70–1.18)
week)
Heavy (44/
1.36 (0.95–1.96)
week)
5
Yes/no
1
1.70 (0.96–2.99)
Abbreviations: CI, confidence interval; ED, erectile dysfunction; IIEF, International Index of Erectile Function; OR, odds ratio.
345
International Journal of Impotence Research
21
o20 g/day
20–40 g/day
X40 g/day
1
(0.6–1.5)
(0.6–1.3)
(0.5–1.3)
(0.5–1.1)
Alcohol consumption and erectile dysfunction
JYW Cheng et al
Italy
799
1.0
0.9
0.8
0.7
Alcohol consumption and erectile dysfunction
JYW Cheng et al
346
Figure 1 Forest plot for regular alcohol consumption (yes/no).
Figure 2 Forest plot for consuming 1–7 drinks/day.
summary estimates due to the omission of any
one study.
Sensitivity analysis (Figure 5) demonstrated
statistical significance for both random effects model
(OR ¼ 0.79; 99% CI, 0.67–0.92; Po0.001) and fixed
effects model (OR ¼ 0.86; 99% CI, 0.81–0.91;
Po0.001), but they did not differ significantly
(P ¼ 0.2117). Only the self-reported single question
ED definition yielded a statistically significant
summary estimate (OR ¼ 0.73; 99% CI, 0.61–0.88;
Po0.001), but not the International Index of Erectile
Function (IIEF) definition (OR ¼ 0.95; 99% CI, 0.65–
1.40; P ¼ 0.739), and they did not differ significantly
(P ¼ 0.1057). Summary estimates for sample sizes less
than 2000 (OR ¼ 0.62; 99% CI, 0.43–0.88; P ¼ 0.001)
International Journal of Impotence Research
and more than 2000 (OR ¼ 0.85; 99% CI, 0.73–0.99;
P ¼ 0.007) differed significantly (P ¼ 0.0400). Only
statistical adjustments for ‘age and other variables’
demonstrated statistically significant OR (OR ¼ 0.84;
99% CI, 0.72–0.97; P ¼ 0.002), but not age adjustment
alone (OR ¼ 0.74; 99% CI, 0.50–1.09; P ¼ 0.047), and
the estimates did not differ significantly.
Cross-sectional and cohort studies with RR (HPFS)
The Health Professionals Follow-up Study
(HPFS)23,24 provided much evidence on the influence of lifestyle factors on the development of ED.
The HPFS cross-sectional study23 involved 31 742
men aged 53–90 and was probably the largest cross-
Alcohol consumption and erectile dysfunction
JYW Cheng et al
347
Figure 3 Forest plot for consuming 8 or more drinks/day.
Table 2 Year published and quality assessment of selected studies
Internal validity
Reference Year a b c d e
no.
Shaeer
Moreira
Akkus
Bai
Nicolosi
Millett
Parazzini
Mak
Fedele
Cho
Terai
13
14
15
16
5
17
18
19
20
21
22
2003
2002
2002
2004
2003
2006
2000
2002
2000
2003
2004
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
External validity
Statistical adjustments for odds ratio
f
g h i Sum j
k l M n o p q r Sum s
t
u
v
w
x
y
z
Aa Sum
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
7
7
7
7
7
4
5
8
6
8
8
þ
þ
þ
þ
þ
þ
sectional study on ED to date. The multivariateadjusted RR for ED were decreased with moderate
levels of alcohol consumption. The RRs were 1.0
(0.9–1.2), 0.9 (0.8–1.0), 0.8 (0.7–1.0) and 1.0 (0.8–
1.2) for 0.1–4.9, 5.0–14.9, 15–29.9, X30.0 g/day of
alcohol consumption respectively (using 0 g/day as
reference), after adjustments for comorbidity, medication, smoking status, physical activity, television
watching, body mass index (BMI) and other factors.
The HPFS prospective cohort study24 demonstrated the independent effects of physical activity
(RR 0.7, 95% CI, 0.7–0.8), obesity (multivariate RR
1.9, 95% CI, 1.6–2.2) and smoking (RR 1.5, 95% CI,
1.3–1.7) on the development of ED. Around 51 529
health professional men were recruited at baseline,
after inclusion of those who were healthy at baseline
and exclusion of those lost to follow-up, 22 086 men
were computed in the analysis. No significant
difference in risk of developing ED was found in
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
6
7
5
4
4
5
6
5
5
6
5
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
þ
1
1
8
1
7
5
3
1
2
4
1
all categories of alcohol consumption: multivariate
adjusted RR 1.0 (0.9–1.1), 1.0 (0.9–1.1), 1.0 (0.9–1.1)
and 1.1 (1.0–1.2), for 0.1–4.9, 5.0–14.9, 15–29.9,
X30.0 g/day of alcohol consumption respectively
(using 0 g/day as reference). Statistical adjustments
were made for age, marital status, smoking, alcohol
and BMI.
ED was defined as having poor or very poor ability
in the previous 3 months to have and maintain an
erection adequate for intercourse. Mantel–Haenszel
statistics was used in both papers to calculate RR,
rather than logistic regression.
Cohort study (MMAS)
The Massachusetts Male Aging Study (MMAS)25
involved a baseline cohort of 1709 men, but the
analysis was restricted to only 513 men without ED
at baseline. The adjusted incidence for ED was 16%
International Journal of Impotence Research
Alcohol consumption and erectile dysfunction
JYW Cheng et al
348
Table 3 Criteria for the methodology quality assessment of prevalence studies. (Adapted and expanded from Prins et al.)
Internal validity:
Objective
a) Was there a clearly focused objective?
Data collection
b) Were the data prospectively collected?
Measurement instrument
c) Was the measurement instrument validated?
d) Was the period covered by the measurement instrument specified?
Definition of variables
e) Was the definition of the outcome variable stated?
Reported prevalences
f) Was age-specific prevalence reported?
g) Were severity-related prevalences reported?
Causal analysis
h) Were possible statistical correlates of outcome variable reported?
i) Were alternative explanations considered?
External validity
Study population
j) Was there randomization involved in the selection of study subjects?
k) Did the recruitment of study subjects make them representative of the target population or patient population they claim to study?
Eligibility criteria
l) Was the age range specified?
m) Were inclusion and exclusion criteria specified?
Response rate
n) Was the response rateb at least 70%?
o) Were information given about the non-responders that allow inference to be made on the representativeness of the study population?
Description of study period
p) Was the study period specified?
Description of study population
q) Were demographic characterisitics of the study population given?a
Implications
r) Were implications of the results considered? (policy makers, clinicians, community)
Statistical adjustments
1) Was the odds ratio
2) Was the odds ratio
3) Was the odds ratio
4) Was the odds ratio
5) Was the odds ratio
6) Was the odds ratio
7) Was the odds ratio
8) Was the odds ratio
9) Was the odds ratio
for odds ratio
adjusted for age?
adjusted for education?
adjusted for income/occupation/socioeconomic status?
adjusted for depression?
adjusted for diabetes?
adjusted for hypertension?
adjusted for heart diseases?
adjusted for smoking?
adjusted for alcohol?
a
Two or more of: (i) age distribution, (ii) relevant comorbidity, (iii) lifestyle factors (for example, smoking), (iv) socioeconomic data (for
example, income, educational level, marital status).
b
Treated as completion rate for quality appraisal.
(95% CI, 12–22), 16% (11–23) and 15% (8–24) in
those with o1 drink/day, 1–3 drinks/day and X4
drinks/day of alcohol consumption respectively.
This incidence figure was adjusted for age, active
and passive smoking, overweight, hypertension,
physical activity, cholesterol, fat intake, testosterone, depression and antihypertensive medication
intake. It is also found that the OR (adjusted for the
International Journal of Impotence Research
same variables) for ED was 0.95 (95% CI, 0.54–1.67)
and 0.87 (0.41–1.86) in those with 1–3 drinks/day
and X4 drinks/day of alcohol consumption respectively, using o1 drink/day as reference, although
the result was not statistically significant.
ED was defined according to NIH consensus as
not ‘being able to get and keep an erection that is
rigid enough for satisfactory sexual activity.’
Alcohol consumption and erectile dysfunction
JYW Cheng et al
349
Figure 4 Funnel plot for publication bias.
Figure 5 Sensitivity analysis. Solid line indicates OR ¼ 1.
Discussion
A handful of studies26–28 have documented the
harmful effects of chronic alcohol consumption on
sexual functioning but hardly any study found
potential beneficial effects of alcohol on ED. We
have demonstrated through meta-analysis the possible beneficial effects of alcohol on ED.
The relationship between alcohol and ED was
complex from the data: consuming 1–7 drinks/week
appeared to confer the lowest risk (OR ¼ 0.73;
P ¼ 0.101) but it was not statistically significant,
and only 8 or more drinks/week was significant
(OR ¼ 0.85; P ¼ 0.007). It appears that alcohol consumption, much similar to its relationship with
cardiac survival, is related to sexual function in
a J-shaped manner, with moderate consumption
conferring highest protection and higher consumption conferring less benefits.
Considering that ED and heart diseases share
similar cardiovascular risk factors, and the wellknown chronic cytotoxic effects of alcohol on
general health,29 hepatic function30 and immune
function,31 general health might be a mediator
between the association of high alcohol consumption and ED. This J-shaped relationship finding
might explain why studies have shown harmful
effects of heavy alcoholism on sexual function, that
is alcoholism entails excessive drinking and carries
the increased risk of ED at the tail of the J curve.
However, caution has to be exercised in the extrapolation, as our results did not show a statistically
significant OR for consuming 1–7 drinks/week, and
did not demonstrate any harmful effects of alcohol
on sexual function. In fact, three large studies9,16,18
have demonstrated progressively smaller ORs of ED
for increasing levels of alcohol consumption.
We attempted to identify the cutoff level of
alcohol consumption where risks outweigh benefits,
by identifying the ‘number of drinks/week’ that has
an OR closest to 1. To our surprise, the category ‘8 or
International Journal of Impotence Research
Alcohol consumption and erectile dysfunction
JYW Cheng et al
350
more drinks/week’ generated an OR that barely
missed the unity and was statistically significant.
Although we were unable to pinpoint the level of
alcohol consumption where risks (OR41) outweigh
benefits (ORo1), we were at least able to say that
‘8 or more drinks/week’ is likely to be a cutoff
where the OR for ED becomes less than 1. Whether
consuming more or less alcohol would yield a
smaller OR was indeterminate.
The results from the two cohort studies complicated the picture even further: they did not show
any significant effects of alcohol consumption on
ED. The cross-sectional analysis of the HPFS data
demonstrated a protective association of alcohol on
ED, much in a J-shaped manner, but after follow-up
of the subjects the cohort analysis did not find any
significant associations between alcohol consumption and ED. Since the level of evidence from a
cohort study is generally higher than from a crosssectional study, due to less confounding and recall
bias, the evidence from the HPFS cohort should
weigh more heavily in our analysis. This suggests
that the apparent protective association of alcohol
consumption on ED was probably due to confounding (since recall, selection, observer and volunteer
biases were less likely in the selected populationbased cross-sectional studies). However, the sensitivity analysis demonstrated a significant protective
association of alcohol consumption with higher
level of statistical adjustments (less confounding),
not with age adjustment alone (more confounding),
and suggests that with better statistical adjustments
(reduction of confounding) the demonstration of
protective association might be possible. The sensitivity analysis on study sample size found that
smaller studies demonstrated larger effects than
larger studies, and might suggest that the estimate
obtained from larger studies would be more reliable
(effect towards null), although their estimates did
not differ significantly. The relative importance of
each study with regard to their sample sizes was
taken care of by the differential weighting of the
random effects model.
A major limitation of previous population-based
studies was their relatively small sample sizes, and
we sought to overcome this sample size problem by
pooling data across studies, although we suspected
that the OR for ‘1–7 drinks/week’ was not statistically significant due to inadequate sample size.
The meta-analyses had four limitations, (1)
confounding, (2) definitions of ED and alcohol
consumption, (3) weak causal inference and
(4) heterogeneity.
Limitation 1
Confounding can reduce the internal validity of a
study, and since ED is a multifactorial disorder, the
association between alcohol and ED as demonstrated could be confounded by some other factors
International Journal of Impotence Research
not adjusted for. First, the effect size for alcohol
consumption was not large, and the significant
effect could be due to residual confounding. Second,
we obtained pooled estimates based on ORs that
were adjusted for different numbers of variables
from different studies, and therefore each included
study had a different level of confounding. Our
minimum requirement for inclusion was that the
ORs had to be age-adjusted, other than this if more
variables were adjusted for the better. However
referring to Figure 1, if confounding was a problem,
one would find ORs that were adjusted for more
variables to lie closer to unity (that is, no effect) and
those that were only age-adjusted to be closer to zero
(that is, protective association), but this did not
appear to be the case.
Limitation 2
Alcohol consumption was assessed in many ways,
one by grams of alcohol, and most by the number of
drinks/week, but the categories were different. We
took a conservative approach in pooling the data, for
example, if the categories in Millett et al.’s17 study
were none/1–4/X4 drinks/week, we excluded the
OR for the ‘open-ended’ X4 drinks/week and only
included the OR for 1–4 drinks/week into our final
1–7 drinks/week category, and likewise we only
included the OR for 3–4 drinks/week from Cho
et al.’s21 study in the 1–7 drinks/week category. We
believe that this conservative approach in pooling
the results should ensure the reliability of our
results.
The included studies used two broad definitions
of ED, one that was based on IIEF-5 or IIEF-15, and
the other that was based on a single self-reporting
question. Sensitivity analysis showed that the
summary estimates based on each definition were
different, and only the self-reported one was
significant (OR ¼ 0.73; 99% CI, 0.61–0.88;
Po0.001) but not the IIEF one (OR ¼ 0.95; 99% CI,
0.65–1.40; P ¼ 0.739), and the latter was not statistically significant. This finding raises serious doubts
to the use of single question self-reported composite
measures of ED, which many studies worldwide
used, since it appears that the summary estimate for
each definition was different. However, since the
estimates did not differ significantly (P ¼ 0.1057),
this difference could be due to chance.
Limitation 3
Causal inference from the cross-sectional design
is weak. The alternatives for better causal inference
are either a cohort study or a randomized controlled
trial, but in either case few studies have been
done. We sought to support our meta-analysis
results with the research findings from cohort
studies (MMAS and HPFS), but the cohorts did not
demonstrate any significant relationships between
Alcohol consumption and erectile dysfunction
JYW Cheng et al
alcohol consumption and risk of ED. The crosssectional HPFS supported our meta-analysis results
of the cross-sectional studies in demonstrating a
significant protective association, but the cohort
HPFS did not, and this suggests that the protective
association demonstrated in cross-sectional studies
might have resulted from confounding. Another
possible explanation was that men who developed
ED opted not to drink alcohol, whereas men without
ED continued to drink alcohol. As a result, an
apparent ‘harmful effect’ of not drinking alcohol was
found, which was the inverse of protective association of drinking alcohol in terms of ORs. Since the
cohort study design has better causal inference, the
results from the two cohort studies suggest that
alcohol neither causes nor prevents ED from developing. Also, the statistical association was found
after pooling a large number of studies, and may
have limited biological significance for an individual drinker.
Limitation 4
We found significant heterogeneity in two of the
three meta-analysis estimates. In general, there can
be many sources of heterogeneity in meta-analysis,
and we can only surmise the sources in this study,
which may include varying number of controlled
confounders, different definitions of ED and alcohol
consumption and the diverse populations included.
We used the random effects model that is preferred
to the fixed effects model when significant heterogeneity exists.
The geographical regions covered in this metaanalysis include South America (Brazil), Europe
(Italy, Belgium), Africa (Egypt, Nigeria), Middle East
(Turkey) and Asia-Pacific (Australia, China, Japan,
Korea, Malaysia, Pakistan). Therefore our results
have limited generalizability to other regions (for
example, North America).
Two reasons might have propagated the myth that
alcohol consumption is a risk factor for ED. First,
that alcohol consumption enhanced sexual desire
but impaired sexual performance is perhaps a shortlived effect of alcohol and will not cause ED permanently. Second, that severe alcoholism impairs
sexual functions may be an extreme example and
is confounded by underlying deterioration of general health, and unless it is excessive it is unlikely to
cause ED permanently. More research has to be done
to assess the association between acute (we did not
investigate) and chronic (which we investigated)
alcohol consumption and development of ED, particularly using large-scale cohorts since randomized
controlled trials may be unethical.
This is the first study that systematically reviewed
and meta-analyzed the association between alcohol
consumption and ED. Our meta-analysis found a
significant protective association of regular
(chronic) alcohol consumption on ED in cross-
sectional studies, in particular for the consumption
of 8 or more drinks/week. Evidence from large
cohort studies suggests that regular alcohol consumption is not significantly associated with ED
development. Therefore this study has demonstrated, in the least, that chronic alcohol consumption is not a risk factor for ED. We hope to demonstrate with the results of this study that the
association between alcohol consumption and development of ED might not be as straightforward as
it seemed, and the undue popularity of alcohol
being labeled as a risk factor for ED was probably
unjustified since there was little research evidence
to support it.
351
References
1 Rosen RC, Goldstein I. Sexual dysfunction. In: Warrell DA,
Cox TM, Firth JD, Benz EJ (eds), Oxford Textbook of Medicine,
4th edn Oxford University Press: Oxford, 2003.
2 Sadock VA. Normal human sexuality and sexual dysfunctions.
In: Sadock BJ, Sadock VA (eds), Kaplan & Sadock’s Comprehensive Textbook of Psychiatry, 8th edn Lippincott Williams
& Wilkins: Philadelphia, PA, 2005.
3 Carbone Jr DJ, Seftel AD. Erectile dysfunction. Diagnosis and
treatment in older men. Geriatrics 2002; 57: 18–24.
4 Hutter Jr AM. Role of the cardiologist: clinical aspects of
managing erectile dysfunction. Clin Cardiol 2004; 27(Suppl 1):
I3–I7.
5 Nicolosi A, Moreira Jr ED, Shirai M, Bin Mohd Tambi MI,
Glasser DB. Epidemiology of erectile dysfunction in four
countries: cross-national study of the prevalence and correlates of erectile dysfunction. Urology 2003; 61: 201–206.
6 Seftel AD, Sun P, Swindle R. The prevalence of hypertension,
hyperlipidemia, diabetes mellitus and depression in men with
erectile dysfunction. J Urol 2004; 171(6 Part 1): 2341–2345.
7 Seidman SN. The aging male: androgens, erectile dysfunction,
and depression. J Clin Psychiatry 2003; 64(Suppl 10): 31–37.
8 Lam TH, Abdullah AS, Ho LM, Yip AW, Fan S. Smoking and
sexual dysfunction in Chinese males: findings from men’s
health survey. Int J Impot Res 2006; 18: 364–369.
9 Thompson IM, Tangen CM, Goodman PJ, Probstfield JL,
Moinpour CM, Coltman CA. Erectile dysfunction and subsequent cardiovascular disease. JAMA 2005; 294: 2996–3002.
10 Le NA. Inflammation, oxidative stress, and atherosclerosis.
Curr Opin Lipidol 2004; 15: 227–229.
11 Thomas GN, Tomlinson B, Abdullah AS, Yeung VT, Chan JC,
Wong KS. Association of erectile dysfunction with cardiovascular risk factors and increasing existing vascular disease in
male chinese type 2 diabetic patients. Diabetes Care 2005; 28:
2051–2053.
12 Prins J, Blanker MH, Bohnen AM, Thomas S, Bosch JL.
Prevalence of erectile dysfunction: a systematic review of
population-based studies. Int J Impot Res 2002; 14: 422–432.
13 Shaeer KZ, Osegbe DN, Siddiqui SH, Razzaque A, Glasser DB,
Jaguste V. Prevalence of erectile dysfunction and its correlates among men attending primary care clinics in three
countries: Pakistan, Egypt, and Nigeria. Int J Impot Res 2003;
15(Suppl 1): S8–S14.
14 Moreira Jr ED, Lisboa Lobo CF, Villa M, Nicolosi A, Glasser
DB. Prevalence and correlates of erectile dysfunction in
Salvador, northeastern Brazil: a population-based study. Int J
Impot Res 2002; 14(Suppl 2): S3–S9.
15 Akkus E, Kadioglu A, Esen A, Doran S, Ergen A, Anafarta K
et al. Prevalence and correlates of erectile dysfunction in
Turkey: a population-based study. Eur Urol 2002; 41: 298–304.
16 Bai Q, Xu QQ, Jiang H, Zhang WL, Wang XH, Zhu JC.
Prevalence and risk factors of erectile dysfunction in three
International Journal of Impotence Research
Alcohol consumption and erectile dysfunction
JYW Cheng et al
352
17
18
19
20
21
22
23
cities of China: a community-based study. Asian J Androl
2004; 6: 343–348.
Millett C, Wen LM, Rissel C, Smith A, Richters J, Grulich A
et al. Smoking and erectile dysfunction: findings from a
representative sample of Australian men. Tob Control 2006;
15: 136–139.
Parazzini F, Menchini Fabris F, Bortolotti A, Calabro A,
Chatenoud L, Colli E et al. Frequency and determinants of
erectile dysfunction in Italy. Eur Urol 2000; 37: 43–49.
Mak R, De Backer G, Kornitzer M, De Meyer JM. Prevalence
and correlates of erectile dysfunction in a population-based
study in Belgium. Eur Urol 2002; 41: 132–138.
Fedele D, Bortolotti A, Coscelli C, Santeusanio F, Chatenoud
L, Colli E et al. Erectile dysfunction in type 1 and type 2
diabetics in Italy. On behalf of Gruppo Italiano Studio Deficit
Erettile nei Diabetici. Int J Epidemiol 2000; 29: 524–531.
Cho BL, Kim YS, Choi YS, Hong MH, Seo HG, Lee SY
et al. Prevalence and risk factors for erectile dysfunction in
primary care: results of a Korean study. Int J Impot Res 2003;
15: 323–328.
Terai A, Ichioka K, Matsui Y, Yoshimura K. Association of
lower urinary tract symptoms with erectile dysfunction in
Japanese men. Urology 2004; 64: 132–136.
Bacon CG, Mittleman MA, Kawachi I, Giovannucci E, Glasser
DB, Rimm EB. Sexual function in men older than 50 years of
International Journal of Impotence Research
24
25
26
27
28
29
30
31
age: results from the health professionals follow-up study.
Ann Intern Med 2003; 139: 161–168.
Bacon CG, Mittleman MA, Kawachi I, Giovannucci E, Glasser
DB, Rimm EB. A prospective study of risk factors for erectile
dysfunction. J Urol 2006; 176: 217–221.
Feldman HA, Johannes CB, Derby CA, Kleinman KP, Mohr BA,
Araujo AB et al. Erectile dysfunction and coronary risk factors:
prospective results from the Massachusetts male aging study.
Prev Med 2000; 30: 328–338.
Schiavi RC. Chronic alcoholism and male sexual dysfunction.
J Sex Marital Ther 1990; 16: 23–33.
Smith DE. Alcoholism, recovery, and sexual dysfunction.
Psychiatr Med 1985; 3: 163–172.
Okulate G, Olayinka O, Dogunro AS. Erectile dysfunction:
prevalence and relationship to depression, alcohol abuse and
panic disorder. Gen Hosp Psychiatry 2003; 25: 209–213.
Hart CL, Smith GD, Hole DJ, Hawthorne VM. Alcohol consumption and mortality from all causes, coronary heart disease, and
stroke: results from a prospective cohort study of scottish men
with 21 years of follow-up. BMJ 1999; 318: 1725–1729.
Lieber CS. Alcoholic liver disease: new insights in pathogenesis lead to new treatments. J Hepatol 2000; 32(1 Suppl):
113–128.
Happel KI, Nelson S. Alcohol, immunosuppression, and the
lung. Proc Am Thorac Soc 2005; 2: 428–432.