mendelian randomization: alcohol does not decrease cardiovasc dz

A large-scale study using mendelian randomization analysis found that there was no protective effect from even small amounts of alcohol on cardiovascular disease (see alcohol intake cardiovasc mendelian random JAMA2022 in dropbox, or doi:10.1001/jamanetworkopen.2022.3849). Many thanks to Tim Naimi for his input on this....

    -- mendelian randomization (MR) basically uses specific genetic variations called single nucleotide polymorphisms (SNPs) that are strongly associated with a specific behavior (eg alcohol consumption), with the assumption that these SNPs/genetic variations are randomly represented in the population, to see if there is a causal effect on an outcome (cardiovascular diseases) from observational data. 

        -- so, this technique represents a relatively easy-to-do analysis, sort of "randomized controlled trial" (since these SNPs are assumed to be randomly represented in the population), and pretty much eliminates reverse causation (the possibility that cardiovascular disease leads people to drink).

    -- an extension of this model is non-linear mendelian randomization, which divides the population into strata with different levels of exposures (alcohol, in this case), then analyzes each stratum to get stratum-specific estimates 

    --  see commentary section below for more details of these techniques

 

Details: 

-- 371,463 individuals were accessed through the UK Biobank from 2006 to 2010, with follow-up until 2016

-- mean age 57, 46% men, BMI 27, BP 140/82

-- hypertension in 33%, coronary artery disease in 7%, MI 4%, stroke 2% heart failure 2%, atrial fibrillation 4%

-- drinking groups:

    -- abstainers: 0 drinks/wk

    -- light drinkers: >0 to 8.4 drinks/wk

    -- moderate drinkers: 8.4-15.4 drinks/wk

    -- heavy drinkers: 15.4-24.5 drinks/wk

    -- abusive drinkers (their term, not mine): >24.5 drinks/wk

-- mean alcohol consumption: 9.2 standard drinks per week 

    -- light drinkers: mean consumption 4.9 drinks/week: 38% beer, 29% red wine, 24% champagne/white wine, 6% spirits, 3% fortified wine

    -- heavy drinkers: mean consumption 21 drinks/week: 38% beer, 24% red wine, 28% champagne/white wine, 7% spirits, 2% fortified wine

 

-- main outcomes: association between alcohol consumption and cardiovascular diseases, including hypertension, coronary artery disease (CAD), MI, stroke, heart failure, and atrial fibrillation

    -- they also compared results from a standard epidemiologic evaluation and the results from MR in the same cohort

 

Results: 

-- light-to-moderate alcohol consumption, in the epidemiologic component of this study, was associated with healthier lifestyles (diet, exercise, smoking, etc), as has been found in prior studies (eg see http://gmodestmedblogs.blogspot.com/2023/04/alcohol-consumption-small-amounts-not.html); statistical adjustment for these healthier habits attenuated the cardioprotective epidemiologic associations associated with modest alcohol intake 

-- mendelian randomization found that for those people with genetically predicted alcohol consumption (ie those with SNPs associated with higher likelihood of alcohol consumption) had: 

    -- hypertension: 1.3 fold increased risk,  p<0.001 

    -- coronary artery disease: 1.4 fold higher risk, p=0.006 

       -- overall interpretation: “light alcohol intake was associated with minimal increases in cardiovascular risk, whereas heavier consumption was associated with exponential increases in risk of both clinical and subclinical cardiovascular disease". But no cardioprotective effect of alcohol

 

Graphically: 

 -- epidemiologic association between alcohol consumption and hypertension or coronary artery disease, demonstrating the "J-shaped" curve (though adjusting for 6 lifestyle factors made these results insignificant): 

 

 

 

-- Genetic Association (mandelian randomization results) of alcohol consumption with hypertension or coronary artery disease demonstrating no more "J-shape": 

 

-- they also found very strong associations between MR estimates for the association between alcohol consumption (vs life-long abstainers) and: hypertension (28% increase), CAD (38% increase), MI (37% increase), stroke (26% increase), heart failure (39% increase), and atrial fibrillation (24% increase)

    -- and, in light and moderate drinkers, a 1-drink/day increase in allele score (combination of all weighted SNPs) stratified by the amount of alcohol consumed was associated with a 30% increased odds ratio of hypertension (however, the association with hypertension by observational studies is for drinking at least for those drinking 2 or more alcohol/d: eg see https://pubmed.ncbi.nlm.nih.gov/16208131/ ) and increasing to 70% in moderate drinkers. And these numbers increased with increasing alcohol consumption exponentially, and these increases were even stronger association for CAD.

-- an analysis of the Mass General Brigham Biobank of 30,716 participants also found a strong association between habitual alcohol intake and associations with both systolic and diastolic blood pressure, also in a non-linear quadratic fashion

 

 Commentary:

-- mendelian randomization (MR) basically uses specific single nucleotide polymorphisms (SNPs), which are genomic variants that occur naturally in the human population and can result in changes in amino acids or changes in gene expression; human DNA contains at least 5-10 million SNPs which account for 90% of genetic variations that have been found to be strongly associated with a specific behavior (eg several were found affecting alcohol consumption here). the assumption is that these SNPs/genetic variations are randomly represented in the population, and they can be analyzed to see if there is a causal effect on an outcome (cardiovascular diseases) from analyzing subsequent observational data. so, this technique represents an easy-to-do study, sort of like a "randomized controlled trial" (since the distribution of SNPs is random), when whole genome evaluations are available.

    -- this technology would largely eliminate reverse causation (when the direction of the association is the reverse, eg that cardiovascular disease leads people to drink), since these SNPs are present at birth and long before any clinical outcomes (or much alcohol consumption...) happens 

    -- also, given the strong predictive value of specific SNPs with actual alcohol consumption, MR is a better measure of lifetime alcohol consumption than can be achieved in cohort studies, which assess drinking typically only very few times (most only once)

    -- it should also be noted that most SNPs have unknown or unclear associations with any functional change. Some come and go (and are not passed on), and some pass on from generation to generation

-- the big assumptions here with MR are:

        1. there is a clear relation between the SNPs for alcohol and alcohol consumption.

            -- most SNPs chosen in this study were specifically related to the alcohol dehydrogenase gene that affects alcohol metabolism. other studies have found that people with 2 SNPs (eg: in homozygous Asian people, found quite commonly in the Asian populations) get flushing with they drink and therefore drink less; and about 7% of Europeans are heterozygous and drink 20% less alcohol, even within the "moderate" range): see  https://www.nature.com/articles/s41598-020-66048-z and https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5684860/

              -- ie, individuals with these SNPs at birth really do drink much less alcohol....

        2. this relationship is not mediated through a common confounder of the SNP and alcohol consumption: the SNP is independently related to alcohol consumption and there is no other factor leading to the connection between the SNP and consuming alcohol

           -- in this study, samples of SNPs associated with smoking, BMI, physical activity and red meat and vegetable intake were removed from the analysis: ie they attempted to focus on the most specific SNPs for alcohol and excluded any SNPs known to have an effect on other cardiovascular risk factors (though are there other confounders??, see below)

         3.  and there is no other pathway from the SNP to the outcome (alcohol consumption) other than the exposure to alcohol. (seems quite likely)

 

-- i have some concerns about MR, though i would appreciate comments from others more statistically/genetically savvy than I if i am off-base:

        1. are we certain we have the full set of SNPs related to alcohol consumption? maybe there are unknown others that are better predictors of either alcohol consumption or cardiac risk, or at least moderators of the association?

            -- there have been many SNPs identified that are associated with alcohol consumption. it does seem that the ones being targeting in this study are very specific target of alcohol dehydrogenase, though some are much more predictive than others in predicting alcohol consumption, presumably by affecting this enzyme. My concern here is that many proteins/enzymes derived from the genetic material in the body have multiple targets and functions and not just one (alcohol breakdown), but perhaps alcohol dehydrogenase affects other now unknown bodily functions, and those other functions may independently affect the cardiovascular outcomes??). 

            -- perhaps there is a strong role for SNPs related to less publicized but quite powerful risk factors that might lead both to increased alcohol consumption and cardiac risk, eg SNPs leading to higher likelihood of increased perception of stress? or to anxiety disorders? Alcohol is a pretty common self-medication for both stress and anxiety, and both of them are risk factors for CAD: see http://gmodestmedblogs.blogspot.com/2022/01/stress-induced-cardiovascular-disease.html .  does the genetic SNP analysis targeting stress or anxiety overlap in any way with the alcohol dehydrogenase ones in this study? this was not tested

            -- and, what about the interactions between SNPs or baseline genetic material that moderate the associations?? some studies assess multiple genetic components to see if there is interplay. but do we know all of them??

       2. and what is the role of epi-genetics?? in this case it is not just the "random" genetic material delivered to the newborn, but target genes that are turned on or off later in life (eg, see:  https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2698894/ )

            -- epigenetic changes reflect the effects of the environment on the individual: "Several lifestyle factors have been identified that might modify epigenetic patterns, such as diet, obesity, physical activity, tobacco smoking, alcohol consumption, environmental pollutants, psychological stress, and working on night shifts" , see  "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3752894/#:~:text=Several%20lifestyle%20factors%20have%20been,and%20working%20on%20night%20shifts. This all means to me that MR does not completely rule out reverse causation: one might have SNPs associated with condition "X", but diet or exercise or smoking might turn on or off the related SNPs, or perhaps turn on/off some protective SNPs (or even baseline genetic pieces) that protect against condition "X"

            -- epigenetic changes can come and go in response to changes in behavior or the environment, though they can also be passed onto offspring, bypassing the egg and sperm. and, epigenetic changes would be missed by conventional DNA sequencing analysis used in MR. would epigenetic changes undercut the MR conclusions??

                -- eg, several epigenetic changes that can be passed on to children are very relevant to alcohol consumption: eg traumatic stress or drug/toxic exposure on brain function, and perhaps even the effect of prenatal eating patterns and the later child development/weight (see https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3521963/

        3. also, are there some other MR methodologic issues?

              -- but (to me) there are still potential concerns about its generalizability: who are the people in the Biobanks who volunteered to submit their DNA for analysis? Are they representative of the general population? If they are really skewed to educated white people, how well do their genomes reflect those of other people? Or people in very different countries and with very different cultures/exposures? And their epigenetic changes may be very different vs those in different environments, changing their SNP and other genetic functionality (eg see https://clinicalepigeneticsjournal.biomedcentral.com/articles/10.1186/1868-7083-4-1 ). A recent article found that genetic loci associated with anxiety had positive genetic correlations with depression and insomnia, but also with coronary artery disease (see https://www.nature.com/articles/s41380-019-0559-1 )

            -- and the clinical outcomes assessed have to be combined with the MR analyses to derive the associations. How robust is the questionnaire-derived information in assessing their alcohol exposure? We know that people tend to minimize reporting alcohol exposure. And perhaps other things (perhaps more so in the psychosocial arena, such as stress, anxiety, sleep disorders, history of trauma, etc)? The questionnaires do not necessarily establish valid objective representations. Does that affect the quality of the MR outcomes? And, even if people do report 100% accurately, how many points in time are necessary? Is one questionnaire about alcohol sufficient to derive alcohol consumption accurately? People do change their habits over time. How many evaluations are necessary? (which could vary by the variable assessed). And even items like diabetes: at the time of the questionnaire, some patients who later develop type 2 diabetes may respond negatively (many patients with type 2 diabetes are unaware of their diagnosis until they become symptomatic, perhaps many years later.) And we do know that sub-threshold glucose intolerance is associated with CAD

        -- but, that all being said, i am not sure how much an effect these concerns have over the big-picture issue: MR seems to be a very valuable tool to assess genetic predisposition to clinical events, largely avoids reverse causation, allows for observing very long-term effects of the genetic changes associated with behavioral or medical changes, these long-term changes also prevent some complex interactions between confounding factors (which typically develop later in life, eg the effects of hypertension and dyslipidemia have a more-than-linear association with cardiovascular disease), and, by mimicking RCTs, MR may thereby provide even more robust outcome data (especially with these very long-term outcomes since RCTs are often less than 1-2 years)

 

-- non-linear MR (NLMR), as noted above, breaks the observed group into strata of risk and assesses outcomes as localized average causal effect (LACE) estimates. this technique can pick up important differences related to different levels of exposure. 

    -- For example, several MR analyses of vitamin D have found that vitamin D supplementation provided no benefit for cardiovascular disease (CVD) risk, but a non-linear MR done of the UK Biobank group of 44,519 CVD cases and 251,269 controls did find that 25(OH)D levels had an "L-shaped" relationship between genetically-predicted serum 25(OH)D and CVD risk, where CVD risk initially decreased steeply with increasing concentrations of 25(OH)D but levelled off at 50 nmol/L, or 20 ng/mL), see vit d and CHD stroke mortality and random LancetDM 2021 in dropbox, or https://www.thelancet.com/action/showPdf?pii=S2213-8587%2821%2900263-1 or https://pubmed.ncbi.nlm.nih.gov/35093020/#:~:text=Mendelian%20randomization%20indicates%20that%20increased,CI%200.60%20to%200.90%5D). or https://pubmed.ncbi.nlm.nih.gov/36279545/  . it is estimated that correction of 25(OH)D levels from below 20 ng/mL would be associated with a 4.4% reduction in CVD incidence.

    -- in the case of the alcohol study above, they also found that quadratic models fit better than linear ones: increased alcohol consumption was associated with exponential increases in disease risk (even when comparing light and moderate drinkers)

 

-- this was an impressive MR study, as compared to some of the older MR studies, in that it removed from evaluation SNPs independently associated with risk factors (smoking, BMI, physical activity, red meat intake, overall health rating, CRP, total cholesterol level), making a direct association between SNP (and actual alcohol consumption) much more likely

-- and the divergent findings of the J-shaped curve in many epidemiologic studies but not in MR strongly supports the much more currently accepted paradigm that light-to-moderate alcohol consumption should not be considered cardioprotective (and, is not so good for cancer risk either: http://gmodestmedblogs.blogspot.com/2019/04/a-bottle-of-wine-week-and-cancer-risk.html )

 

so, this study found several important things:

    -- their epidemiologic study did find a J-shaped curve with cardioprotection at lower levels of alcohol consumption. But they also found increased cardiac risk factors in the non-drinking group that, when controlled for, attenuated this epidemiologic association with light-to-moderate drinkers

    -- the human genetic data found a causal association between alcohol intake and the risk of hypertension and CAD that started to increase with even modest alcohol consumption (eg a bit less than 7 drinks/week, such as a not so full glass of wine with dinner). and the risk and cardiovascular disease increased exponentially at higher levels of alcohol intake

        -- increasing from 0-7 drinks/week to 7-14 drinks/week was associated with increase cardiovascular risk in both men and women 

    -- this study confirms the conclusions of the previous blog (http://gmodestmedblogs.blogspot.com/2023/04/alcohol-consumption-small-amounts-not.html), though i should note that my title for that blog (alcohol consumption: small amounts assoc with increased mortality) was inaccurate and has been re-titled alcohol consumption: small amounts is not assoc with decreased mortality

bottom line: one of the important findings in this study is that consuming 2 drinks of alcohol/day, typically considered low-risk in men, is perhaps not so low a risk....

 

geoff

-----------------------------------

If you would like to be on the regular email list for upcoming blogs, please contact me at gmodest@bidmc.harvard.edu

  

to get access to all of the blogs (2 options):

 

1. go to http://gmodestmedblogs.blogspot.com/ to see the blogs in reverse chronological order

  -- click on 3 parallel lines top left, if you want to see blogs by category, then click on "labels" and choose a category​

  -- or you can just click on the magnifying glass on top right, then type in a name in the search box and get all the blogs with that name in them

 

2. go to https://bucommunitymed.wpengine.com/, a website from the Community Medicine section at Boston Medical Center.  This site does have a very searchable and accessible list of my blogs and is the easiest to view blogs and displays more at a time.

 

please feel free to circulate this to others. also, if you send me their emails (gmodest@bidmc.harvard.edu), i can add them to the list