Sex‐Specific Associations Between Alcohol Consumption and Incidence of Hypertension: A Systematic Review and Meta‐Analysis of Cohort Studies

Background Although it is well established that heavy alcohol consumption increases the risk of hypertension, the risk associated with low levels of alcohol intake in men and women is unclear. Methods and Results We searched Medline and Embase for original cohort studies on the association between average alcohol consumption and incidence of hypertension in people without hypertension. Random‐effects meta‐analyses and metaregressions were conducted. Data from 20 articles with 361 254 participants (125 907 men and 235 347 women) and 90 160 incident cases of hypertension (32 426 men and 57 734 women) were included. In people drinking 1 to 2 drinks/day (12 g of pure ethanol per drink), incidence of hypertension differed between men and women (relative riskwomen vs men=0.79; 95% confidence interval, 0.67–0.93). In men, the risk for hypertension in comparison with abstainers was relative risk=1.19 (1.07–1.31; I2=59%), 1.51 (1.30–1.76), and 1.74 (1.35–2.24) for consumption of 1 to 2, 3 to 4, and 5 or more standard drinks per day, respectively. In women, there was no increased risk for 1 to 2 drinks/day (relative risk=0.94; 0.88–1.01; I2=73%), and an increased risk for consumption beyond this level (relative risk=1.42; 1.22–1.66). Conclusions Any alcohol consumption was associated with an increase in the risk for hypertension in men. In women, there was no risk increase for consumption of 1 to 2 drinks/day and an increased risk for higher consumption levels. We did not find evidence for a protective effect of alcohol consumption in women, contrary to earlier meta‐analyses.

Hypertension is largely a by-product of modern lifestyle factors such as lack of physical activity, 4 unhealthy diet (in particular, salt intake 4 ), or consumption of alcohol. 5 In fact, some guidelines for clinical management, including those from the National Institute for Health and Care Excellence, stipulate that all patients undergoing assessment or treatment for hypertension should receive initial and periodic lifestyle advice, which includes ascertaining their level of alcohol consumption and encouraging a reduced intake if they drink hazardously or heavily. 6,7 The American Heart Association guidelines for the prevention and treatment of high BP recommend limiting daily alcohol intake to 2 or less drinks for men and 1 or less drinks for women. 8 The relationship between alcohol consumption and hypertension was first reported by Lian 9 researching French soldiers serving in World War II. He found a dose-response association with a 4-fold increase between drinkers with the lowest (up to 2 L of wine per day) and highest (>3 L of wine per day plus aperitifs) levels of consumption. Numerous studies since then have confirmed the association between heavy drinking and development of hypertension. 10 However, the association between light-to-moderate drinking and hypertension is still disputed, 11,12 despite a number of meta-analyses 13,14 and countless reviews (overview of recent reviews 15 ). The association may also depend on sex, which could be related to differential alcohol metabolism 16 or drinking patterns. 17 In part because there have been a number of studies since the last systematic review including a meta-analysis, 14 and in part because the techniques for conducting metaanalyses have expanded considerably in recent years, 18 we conducted a systematic review and meta-analysis with the explicit aim to restricting our review to studies above a high-quality threshold and to explore potential influencing factors by stratification by sex and metaregression. This review intends not only to produce improved sex-specific estimates for comparative risk assessments within the Global Burden of Disease studies 19 and for the Global Status Reports of the World Health Organization, 20 but also provide much needed evidence for hypertension-specific drinking guidelines.

Methods
All data are from publicly available sources.

Search Strategy and Selection Criteria
Following the meta-analysis of observational studies in epidemiology checklist, 21 we conducted a systematic electronic literature search using Medline and Embase from inception to April 3, 2017 for keywords and MeSH terms relating to alcohol consumption, hypertension, and observational studies (Table S1). Additionally, we searched reference lists of identified articles and published meta-analyses and reviews. Inclusion criteria were as follows: 1. Full-text article with original cohort data (including nested case-control studies) examining the association between total alcohol consumption and incidence of hypertension. 2. Participants with hypertension at baseline were excluded. 3. Analyses were adjusted or matched for age at baseline. 4. Incidence for at least 2 quantitatively defined categories of average alcohol consumption in addition to nondrinkers, or incidence for former drinkers in relation to lifetime abstainers were reported. 5. Results were sex specific.
For a continuous nonlinear dose-response meta-analysis, results for at least 3 drinking groups in addition to nondrinkers had to be reported. We did not apply language restrictions. Authors were contacted for clarification and missing data. Two reviewers independently excluded articles based on title and abstract or full text, and abstracted the data. Any discrepancies were resolved in consultation with a third reviewer.

Clinical Perspective
What Is New?
• This is the first meta-analysis based on high-quality cohort studies on the relationship between different levels of alcohol consumption and risk for incident hypertension. • We investigated the risk for hypertension separately for men and women in people who did not have hypertension at baseline. • The risk for hypertension in former drinkers was similar to that of lifetime abstainers. • We found that, compared with nondrinkers, the risk for hypertension was increased at all levels of alcohol consumption in men. Contrary to earlier meta-analyses, we did not find a protective effect of low levels of alcohol consumption in women.
What Are the Clinical Implications?
• The findings support sex-specific drinking guidelines with regard to risk for hypertension. These guidelines may be revised to indicate the increased risk for any alcohol consumption in men. • Alcohol consumption should be assessed at the primary care level whenever there is elevated blood pressure. • Changing clinical practice promises to reduce substantial mortality and burden of disease associated with both alcohol consumption and hypertension.

Data Extraction
From all relevant articles, we extracted authors' names, year of publication, country, calendar year(s) of baseline examination, follow-up period, setting of the study, assessment of hypertension status, age (mean or median) at baseline, sex, number of observed incident hypertension cases among participants by drinking group, number of total participants by drinking group, specific adjustment or stratification for potential confounders, and adjusted measures of effect (relative risks [RRs], odds ratios, and hazard ratios) and their confidence intervals or standard errors. Risk estimates by sex and race/ethnicity were treated as independent samples. As a result, multiple articles and estimates from the same study [22][23][24] were included, but each case of incident hypertension was used only once in each of the analyses conducted. If necessary, effect sizes within studies were recalculated to contrast alcohol consumption categories against nondrinkers. 25 Because incidence of hypertension was not rare, we transformed odds ratios to RRs based on the formula described by Zhang and Yu. 26 Hazard ratios and RRs were treated as equivalent measures of risk.

Exposure and Outcome Assessment
Consolidating exposure measures across primary studies involved a 2-step process. First, among drinkers, we converted reported alcohol intake categories in primary studies into an average of pure alcohol in g/d using the midpoints (mean or median) of reported drinking group categories. For open-ended categories, we added three quarters of the second-highest category's range to the lower limit of the open-ended category of alcohol intake if the mean was not reported. Standard drinks vary by country, with 1 standard drink containing 8 to 14 g of pure alcohol. 27 We used reported conversion factors when standard drinks were the unit of measurement to convert all measures to g/d. Then, for reporting of our analyses, we considered categories with a mean of up to 12 g of pure ethanol as 1 standard drink for a global representation. Qualitative descriptions, such as "social" or "frequent" drinkers with no clear total alcohol intake in g/d, were excluded. Because of the changing definitions of hypertension over time, we defined hypertension status at baseline and incident cases of hypertension as defined in the primary studies (typically assessed as taking antihypertensive medications or as mean systolic BP at baseline >140 mm Hg).

Quality Assessment
Most quality scores are tailored for meta-analyses of randomized trials of interventions, [28][29][30] and many criteria do not apply to epidemiological studies examined in this study.
Additionally, quality score use in meta-analyses remains controversial. [31][32][33] As a result, study quality was incorporated by including quality components, such as study design, measurement of alcohol consumption and hypertension, adjustment for age, and sex-specific RRs, in the inclusion and exclusion criteria and further by investigating potential heterogeneity in metaregression models and several subgroup analyses. We used the most adjusted RR reported and the most comprehensive data available for each analysis, and gave priority to estimates where lifetime abstainers were used as the risk reference group.
In a formal risk of bias analysis, we used the Cochrane risk of bias tool for nonrandomized studies (ROBINS-I) 34 to assess risk of bias in primary studies. We rated the evidence for the association between alcohol consumption and incidence of hypertension based on the Grades of Recommendation, Assessment, Development and Evaluation system. 35

Statistical Analyses
In analyses using standard drinks as the exposure measure, RRs were pooled with inverse-variance weighting using DerSimonian-Laird random-effect models to allow for between-study heterogeneity. 36 Small-study bias was examined using Egger's regression-based test. 37 Variation in the effect size because of heterogeneity between-studies was quantified using the I 2 statistic. 38 Between-study heterogeneity was investigated with random-effects metaregressions. 39 Using studies that reported data for 4 or more alcohol intake groups, we conducted 2-stage restricted cubic spline regression in multivariate metaregression models taking into account the variance-covariance matrix for risk estimates derived from 1 reference group 40,41 to calculate continuous nonlinear dose-response curves for total alcohol consumption (g/d) in relation to abstainers. All meta-analytical analyses were conducted on the natural log scale of the RRs (and hazard ratios) in Stata statistical software version 14.2 (Stata LP, College Station, TX).

Literature Search and Study Characteristics
Of 3771 identified references, 465 were reviewed in full text. In total, data from 20 reports from 18 studies were used in the analysis (Figure 1). Nine reports were from the United States, 22-24,42-47 4 from Japan, 48-51 2 from China, 52,53 and 1 each from Germany, 54 South Korea, 55 Finland, 56 Turkey, 57 and Thailand 58 (Table). Several reports from the Nurses' Health Study 42,44-46 were included, but any 1 case of incident hypertension was included only once in any particular analysis. Overall, data from 361 254 participants (125 907 men and 235 347 women) and 90 160 incident cases of hypertension (32 426 men and 57 734 women) were analyzed. Mean age at baseline among men ranged from 25 to 57 years with a weighted mean of 47.1 years (me-dian=50 years), with mean follow-up duration of 5.3 years (median=4; range, 3.9-20.0). In women, mean age ranged from 25 to 60 years with a weighted mean of 46.7 years (median=54), with mean follow-up duration of 7.3 years (median=4; range, 3.9-20.0). Most studies were well adjusted for potential confounders; 1 study was adjusted only for age. 55

Meta-Analyses
The pooled RR among former drinkers 22,24,50,53,58 in comparison with lifetime abstainers was 1.03 (95% CI, 0.89-1.20), with virtually no differences between men and women ( Figure S1). Any alcohol consumption increased the risk for hypertension compared with abstainers in men ( Figure 2 and Figure S2). In women, there was no observed risk increase for consumption of 1 or 2 drinks/day in comparison with abstainers, and an increased risk beyond this level with a pooled RR=1.42 (95% CI, 1.22-1.66; I 2 =88%) for consumption of 3 or more drinks per day ( Figure 3 and Figure S3). Because we included 2 studies from the Nurses' Health Study 42,45 with different follow-up periods of the same participants, we ran a sensitivity analysis including only 1. 42 The results compared to Figure 3 were almost identical (1-2 drinks/day: pooled RR=0.95; 95% CI, 0.88-1.03). Different adjustment for potential confounders in regression models in primary studies resulted in little changes in RRs for different levels of alcohol consumption. In men, data for alcohol consumption beyond 75 g/d were only available from Asian countries ( Figure 4). There were no data for women consuming more than 75 g/d ( Figure 5). Two studies 53,55 were judged to be of serious risk of bias, 1 of low risk, and 16 of moderate risk of bias mainly because of the observational study design and 1-time measurement of alcohol consumption (Table S2). Thirteen studies used clinical measurements of BP to determine incidence of hypertension. [22][23][24][48][49][50][51][52][53][54][55][56][57] Similar relationships were found when we excluded studies with potential serious risk of bias and that relied on self-reported incidence of hypertension ( Figures S4 and S5).
Heterogeneity was high in most analyses, and we conducted metaregressions to investigate potential sources of heterogeneity in drinkers of 1 to 2 drinks/day. The difference between men and women was statistically significant (RR women vs men =0.79; 95% CI, 0.67-0.95; P=0.012; proportion of heterogeneity explained: 69%). There was no significant difference between studies from Asian and non-Asian countries in men (RR non-Asian vs Asian =0.93; 95% CI, 0.71-1.21; P=0.55) or women (RR non-Asian vs Asian =0.92; 95% CI, 0.76-1.11; P=0.33). Three studies from the United States presented results stratified by race (black versus white). In men, people of white origin had nominally higher risk for incidence of hypertension compared with people of black origin (RR Black vs White =0.63; 95% CI, 0.08-4.93; P=0.44), whereas in women the opposite was observed (RR Black vs White =1.31; 95% CI, 0.66-2.63; P=0.38). However, the number of incident hypertension cases was low and CIs were wide, indicating low statistical power to detect significant differences. Three studies presented results by age groups. 46,49,51 Total references identified (4947): Articles excluded based on title and abstract with minimal uncertainty (n=3306) Full-text articles excluded with reason (n=445): Not a cohort study (n=67) Reported outcome was not incidence of HTN (n=51) No alcohol risk data reported (n=227) Less than 2 categories of drinking in comparison to abstainers (n=78) Results were not age-adjusted (n=8) Participants were not normotensive (n=7) Results were not sex-specific (n=7) Prospective studies included in metaanalysis (n=20):   We found no evidence for small-study bias in men or women consuming 1 to 2 drinks/day in visual inspection of funnel plots (Figures S6 and S7)    at a time revealed no meaningful differences in effects ( Figures S8 and S9). Given the observational nature of the studies included, we rated the evidence for a causal effect of high alcohol consumption (3 or more drinks/day) on incidence of hypertension as moderate. However, evidence from randomized controlled trials 59 support a causal effect with higher confidence. Regarding alcohol consumption of 1 to 2 drinks/day, our findings indicate effect modification by sex and no protective association. Evidence from randomized controlled trials for this level of alcohol consumption is limited and thus we judge the quality of the evidence as moderate.

Discussion
In high-quality cohort studies, we found that the association between average alcohol consumption of 1 to 2 drinks/day and risk of hypertension was modified by sex, with men showing an increased risk, whereas women showed no different risk compared with abstainers. Alcohol intake beyond 2 drinks/day was consistently associated with increased incidence of hypertension in both men and women.
Before discussing these results and their implications, we would like to point out some limitations. Conclusions of every meta-analysis are determined by the quality of the original . .  studies. This meta-analysis is based on cohort studies, and thus this study type does not allow conclusions about causality. 60 However, as indicated above, analogue doseresponse relationships for alcohol reduction on reduction of BP and hypertension based on trial data point to a causal effect of level of alcohol consumption on risk of hypertension, 59 and this reasoning is corroborated by plausible biological pathways. 10 Second, all alcohol assessments were based on subjective measurements, which may entail bias. 61 However, most reviews come to the conclusion that subjective measurement of alcohol is reliable, 62 even though there is some bias of underestimating true consumption, for example, sales data show higher consumption compared to survey data. 61,63 Thus, although the overall dose-response relationship would not be affected, there may be some misestimation of the RRs of the levels of drinking (eg, previous work 64 ). Finally, although patterns of drinking, in particular irregular heavy drinking occasions, have been shown to impact on BP and risk of hypertension, 65 we could not find enough cohort studies meeting our inclusion criteria to quantify this effect by meta-analysis. Despite these limitations, the results are consistent in showing a dose-response relationship between level of consumption and risk of hypertension based on observational data, corroborated by randomized controlled trials. For men, there seems to be no lower threshold, whereas for women, the dose-response seems to emerge only beyond 2 drinks a day. For both sexes, no protective effect could be found (and was not expected given the biological pathways 10 ). What could explain the differences between men and women? One explanation could be the difference in heavy drinking occasions within an overall average intake of alcohol of less than 2 drinks. An average of 2 drinks/day could be achieved by actually drinking 2 drinks every day, or by drinking 7 drinks each on Saturday and Sunday. The latter has different effects on blood pressure 66 and thus on risk of hypertension. Further research (both observational and experimental) is necessary, however, to ascertain the effects of pattern of drinking (including peak blood alcohol level) on hypertension and thus the repeated plea to include more measures on patterns of drinking into all epidemiological work. 67,68 Implications Clinicians are faced with a dilemma. On the one side, low-level drinking has been associated with less risk for ischemic heart disease 69,70 ; on the other side, the risk for hypertension seems increased, at least in men. In order to side with caution, patients should be advised to drink as little as possible for many reasons, including increased risk of cancer and injury, to name a few, 15 and the risk of hypertension should be added to the list of diseases where no alcohol consumption is safe. This may require a change in drinking advice in current guidelines for prevention and treatment of hypertension, which state that men should limit their alcohol intake to 2 drinks or less, a level which we found to be associated with increased risk for hypertension. Additional research with stronger, more-experimental study design may help in answering outstanding questions on cardiovascular risk from low levels of drinking.
Other implications of this research are clear: Alcohol consumption should be assessed at the primary care level whenever there is elevated BP. 71 Unfortunately, despite some guidelines recommending such an approach, it is rarely followed in clinical practice (for the example of the 6 largest European Union countries 72 ). Efforts should be made to change  clinical practice, given that alcohol-induced hypertension is both preventable and reversible, 10,59 and there are effective and cost-effective interventions to reduce alcohol consumption level in primary care. 73,74 Changing clinical practice promises to reduce substantial mortality and burden of disease associated with both alcohol consumption and hypertension. 10,75

Sources of Funding
Research reported in this publication was supported by the National Institute on Alcohol Abuse and Alcoholism (NIAAA) of the National Institutes of Health under Award Number R21AA023521 to Roerecke. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The sponsor of the study (NIAAA) had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The authors collected the data and had full access to all of the data in the study. The authors also had final responsibility for the decision to submit the study results for publication.

Disclosures
Roerecke and Rehm report grants from the National Institutes of Health (NIH), the National Institute on Alcohol Abuse and Alcoholism (NIAAA), during the conduct of the study. Rehm reports grants and honoraria from Lundbeck outside of this work (modest relationship). The remaining authors have no disclosures to report.   Funnel plot with pseudo 95% confidence limits Figure S7. Funnel plot for 1-2 drinks/day alcohol intake at baseline compared to abstainers in women, 1989-2017 Horizontal axis shows study effects (logRR), vertical axis shows study precision (standard error of RR). Each dot represents an individual study. Vertical line shows pooled effect (random-effect model). Funnel plot with pseudo 95% confidence limits Figure S8. Influence of omitting a single study for 1-2 drinks/day alcohol intake at baseline compared to abstainers in men, 1989-2017 Figure S9. Influence of omitting a single study for 1-2 drinks/day alcohol intake at baseline compared to abstainers in women, 1989-2017