Alcohol, Disease, and the Politics of Health Data

Here’s a simple question: How much alcohol can adults safely consume?

A study published two weeks ago provides a simple answer: None. Since 2023, this has been the World Health Organization’s position, and scientists increasingly concur.

A study published last week suggests instead that up to one drink per day causes no health risks.

These results of these studies are quite similar. Both show that beyond a certain amount, the more adults drink, the greater their risk of many diseases.

A key difference between the studies is in how to define that cut-off. Practically speaking, the discrepancy is important. If you routinely have a beer after work, or a glass of wine with dinner, are you putting your health at risk?

In this newsletter I’ll be discussing these studies and gleaning some practical guidance from the findings.

A personal note

This newsletter was hard to write, because I ended up arguing against my own biases.

Being the son of an alcoholic, I have a fairly grim perspective on drinking. I would love to tell you that there’s no safe amount of consumption. However, I don’t think that’s what the evidence shows.

I also have a political bias of relevance: I’m a moderate Democrat and angry that the Trump administration, in effect, suppressed the findings of the second study. Even so, I don’t consider the administration’s perspective on alcohol data to be completely misguided.

Sometimes the data just doesn’t say what you want it to say.

Why is this topic important?

Much has been made about the decline in drinking in the U.S. Often emphasized are the most recent Gallup findings, which show that we’ve reached a 90-year low:

In fact, as you can see from the figure, the actual changes aren’t dramatic. 54% of Americans still report drinking on occasion, and alcohol remains our most widely used psychoactive drug.

Meanwhile, the generational differences are minimal (50% of 18 to 34-year-olds report drinking, as compared to 56% of older individuals), and, among certain groups, such as pregnant women, alcohol consumption seems to be rising.

So, what public health message should be shared with that slight majority of us who drink?

The new studies, which present re-analyses of published data, show how granularly modern statistics allow this question to be addressed.

A burden of proof study

On June 1, in Nature Health, Xiaochen Dai and colleagues at University of Washington and Harvard published a re-analysis of 843 studies.

You read that right: 843 studies. Literally every peer-reviewed observational study published between 1961 and 2023 that explored the relationship between alcohol intake and each of 20 health outcomes.

Dai and colleagues took a highly conservative, “burden of proof” approach to their re-analysis. For any one health outcome, they calculated the smallest increase in risk that could be inferred from the data. (They also calculated the smallest protective effects, if any, that could be discerned.) In the Appendix, I explain how this was done; the takeaway is that risk estimates were highly conservative.

What is a “drink”?

In this study, one drink was defined as 10 grams of alcohol. This corresponds to:

• 8.5 ounces of regular beer (5% alcohol content). In the U.S., a typical bottle or can of beer is 12 ounces.

• 1 ounce of spirits (40% alcohol). Google sources tell me that American bartenders typically pour 1.5 ounces into a 2-ounce shot glass.

• 3.5 ounces of wine (12% alcohol content). To illustrate, I asked ChatGPT to generate an image of 3.5 ounces of wine in a standard glass. Then, I uploaded the image to Claude and asked Claude to guess the amount. (Claude guessed 4 ounces. Close enough.)

Some key findings

All analyses used non-drinkers as a basis of comparison.

1. Any amount of alcohol intake greater than zero increased the risk of 15 disease outcomes.

These outcomes included a range of cancers (breast, colorectal, esophageal, laryngeal, lip and oral, pharyngeal, liver, stomach, pancreatic and prostate), as well as cirrhosis, pancreatitis, atrial fibrillation, and more.

Here you can see why the study supports a “no safe amount” conclusion.

2. Light drinking showed protective effects, or at least created no risk, for 5 disease outcomes.

Traditional wisdom, as well as some studies, suggest that light drinking is beneficial. In the present study, drinking up to a certain amount – an inflection point – reduced or had no impact on the risk of the following outcomes:

Can we trust the findings?

I trust the overall dose-response trend showing that the more people drink, the greater their risk of various diseases. This is well-established.

However, I don’t think we can trust the “no safe amount” conclusion. Here are two reasons why not:

1. Low-quality evidence influenced the results.

For each health outcome, data from as few as 7 or as many as 164 studies were combined. One problem with combining studies in this way is that the bad gets mixed in with the good.

By “bad”, I’m not referring to studies that I personally dislike. I mean studies that any expert would view skeptically, because they’re published in the least trustworthy journals and contain obvious, unsalvageable flaws.

For instance, some of the 164 studies on ischemic heart disease fail to include adjustments for risk factors such as smoking. (People who drink more are more likely to smoke and engage in other unhealthy behaviors. If researchers don’t adjust for those behaviors, we can’t know how much drinking per se affects risk.)

I didn’t have time to review 164 studies, but it only took me about 10 minutes to identify two in which the risk ratios were extremely low (less than 1.08) and almost surely influenced the findings.

In sum, the “no safe amount” conclusion may be driven by low-quality research, though this remains a somewhat speculative possibility. Here’s a more serious concern:

2. The lowest levels of alcohol intake were extrapolated.

For instance, the UK Biobank alone is the source for more than a hundred observational studies on the health impacts of drinking. In UK Biobank surveys, the lowest amount of alcohol consumption above zero is labeled as “1 or 2” drinks per day.

If health risks are greater at “1 or 2” drinks per day as compared with zero, we can’t conclude there’s no safe amount of alcohol intake. Perhaps 1 drink per day would be fine, but 2 is harmful. We can’t know.

In fact, the newest evidence, which I’ll discuss now, suggests that up to 1 drink per day does not increase risk. I’ll discuss this study briefly, and then talk a bit about political attempts to suppress the findings.

The Alcohol Intake and Health Study

This study, published on June 8 in Journal of Studies on Alcohol and Drugs, was led by Sinead George (University of Toronto) and a team of researchers from the U.S. and UK.

As in the Burden of Proof study, George and colleagues analyzed data from existing publications, but their emphasis was on quality over quantity:

• For each health outcome, a panel of experts voted on the highest quality meta-analysis.

• For each meta-analysis chosen, self-reported drinking estimates were adjusted upward by roughly 10%, under the assumption that people routinely underreport how much they drink. (Whether or not you agree, the result is more conservative analyses.)

• Within each meta-analysis, comparison groups consisted of lifelong abstainers. (Simply comparing drinkers to non-drinkers can be misleading; for example, the latter may include people who drank for a long time and then quit.)

(More on the statistical methodology in the Appendix.)

Here one “drink” was defined as 13.6 grams of alcohol, or roughly the equivalent of one 12-ounce can or bottle of beer, one 1.5-ounce shot, or one 5-ounce glass of wine.

(ChatGPT generated this picture; Claude AI guessed correctly that it depicts 5 ounces of wine.)

I didn’t use AI for anything else in this post, by the way. It’s troubling that I feel the need to say that.

Some key findings

1. Starting at around one drink per day, the more alcohol people consume, the greater their lifetime risk of alcohol-related mortality.

For instance, at one drink per day, the increased risk of mortality was 1 in 1,000. The risk increased to about 1 in 25 at two drinks per day.

2. As people exceeded one drink per day, their lifetime risk of various diseases increased.

The dose-response patterns were similar to those that Dai and colleagues reported. However, the actual risk ratios can’t be directly compared (see Appendix for details).

3. Drinking was related to increased lifetime risk of injuries (fatal motor vehicle crashes, suicide attempts, and interpersonal violence).

For instance, each 0.02% increase in blood alcohol concentration – roughly the impact of one drink on an average adult – increased the odds of a fatal crash by 74%.

Can we trust the findings?

Here again, I trust the overall pattern: The more people drink, the greater their risk of alcohol-related mortality, diseases, and injuries.

I also agree that the lowest levels of alcohol intake don’t increase health risks among most adults. This study and others support this conclusion. But I’m not convinced that exactly one drink per day is the cut-off.

Why not? Simply put, alcohol intake has never been measured very accurately in this literature.

Surveys usually ask participants about their drinking behavior “on a typical day”, or “over the past month”, or across some other period of time. “Drinks” are defined qualitatively (e.g., “small glass of wine”) as well as quantitatively (e.g., in ounces). Participants respond to these questions once during the course of the study, or once every several years. Based on their responses, drinking behavior over a period of decades is inferred.

The crudity of these methods should be obvious. Do you measure how many ounces of alcohol you drink? Does your drinking behavior stay the same across the decades? (If you’re familiar with Mendelian Randomization studies, you know that they don’t solve the problem but, so to speak, relocate it.)

These are a few of the reasons why I believe we can trust general trends in the data, but not specific cut-off values. How much alcohol can be consumed without incurring health risks? More than zero, but probably less than two drinks per day (depending on alcohol content, body weight, metabolism, etc.).

I should stress that I’m not referring to everyone here. For some people, there truly is no safe amount of alcohol intake. Experts mostly or completely agree that this includes children and adolescents, pregnant women, people taking certain medications, people with certain medical conditions, and people with a history of alcohol use disorder.

The politicization of health data: A case study

George and colleagues’ Alcohol Intake and Health Study was commissioned by the Biden administration to help inform revisions to the 2020-2025 Dietary Guidelines for Americans. The DGA is an authoritative guide jointly produced by HHS and the USDA.

What happened instead is that a competing report was prioritized. This report, created by panelists appointed by the National Academy of Sciences, Engineering, and Medicine (NASEM), found that as many as 2 drinks per day for men and 1 for women actually serve a protective function, lowering the risk of most diseases other than breast cancer.

It turns out that a number of panelists on the NASEM report have financial ties to the alcohol industry. The report itself is deeply flawed. Meanwhile, industry representatives badgered the Biden and second Trump administrations with concerns about the Alcohol Intake and Health Study. Further details can be found here and here.

How did the current administration respond?

Put simply, they ignored the Alcohol Intake and Health study, and gutted the DGA’s alcohol-related guidance.

Here is part of what the original 2020-2025 DGA said about alcohol. (Feel free to skim; it’s lengthy.)

“The Dietary Guidelines does not recommend that individuals who do not drink alcohol start drinking for any reason. There are also some people who should not drink at all, such as if they are pregnant or might be pregnant; under the legal age for drinking; if they have certain medical conditions or are taking certain medications that can interact with alcohol; and if they are recovering from an alcohol use disorder or if they are unable to control the amount they drink. If adults age 21 years and older choose to drink alcoholic beverages, drinking less is better for health than drinking more.

Evidence indicates that, among those who drink, higher average alcohol consumption is associated with an increased risk of death from all causes compared with lower average alcohol consumption. Alcohol misuse or consuming alcohol in excess of recommendations increases risk of several other conditions such as liver disease, cardiovascular disease, injuries, and alcohol use disorders.”

“Adults of legal drinking age can choose not to drink or to drink in moderation by limiting

intake to 2 drinks or less in a day for men and 1 drink or less in a day for women, when alcohol is consumed. Drinking less is better for health than drinking more...

“For those who choose to drink, intakes should be limited to 1 drink or less in a day for women and 2 drinks or less in a day for men, on days when alcohol is consumed.”

The old DGA went on to define “drink” specifically enough that people could, if they choose, measure how much alcohol they consume and drink within safe limits. The excess calories provided by alcoholic drinks was also discussed.

Right or wrong, this constitutes actual guidance. Americans were provided with detailed advice on exactly what we should and should not do, and a rationale for the advice was articulated.

In contrast, here’s all that the revised 2025-2030 DGA has to say about alcohol:

“Consume less alcohol for better overall health.

+People who should completely avoid alcohol include pregnant women, people who are recovering from alcohol use disorder or are unable to control the amount they drink, and people taking medications or with medical conditions that can interact with alcohol.

+For those with a family history of alcoholism, be mindful of alcohol consumption and associated addictive behaviors.”

That’s it. Not another word about alcohol in the entire document.

“Consume less alcohol for better overall health” is a reasonable statement, but it hardly constitutes guidance.

The loss of information is staggering. The HHS and USDA, government organizations that exist to protect American health, have responded to pressure from the alcohol industry by deleting multiple paragraphs of guidance with respect to drinking.

To be fair, I appreciate the removal of the old DGA’s drink figures, since they’re out of date (e.g., for men, 2 drinks per day no longer seems to be devoid of risk) and misleadingly specific without acknowledgment of scientific uncertainty and debate.

Still, on the whole, the alcohol-related revisions in the current DGA are a travesty. The public needs more information. The DGA should acknowledge expert consensus that 2 or more drinks per day create health risks. It should emphasize that some scientists believe there’s no safe amount of alcohol consumption, while others claim that minimal intake, such as one drink per day, doesn’t increase risk. After all, we’re grownups. We can understand that experts don’t always agree.

The politicization of health data sometimes yields strong assertions (e.g., vaccines cause autism). In the case of alcohol, the current administration, following the lead of financially-incentivized experts, has asserted next to nothing. This is a threat to public health, because alcohol is a dangerous substance. Would you tell people to “eat less saturated fat for better overall health”? Sure, but you’d provide details. (The new DGA recommends that saturated fat consumption not exceed 10% of caloric intake.)

Final thoughts

It’s not a bad thing when evidence clashes with your biases. If you’re receptive, you might learn something. I take a dim view of alcohol, but I no longer believe, as I did as a teenager, than light drinking is harmful. Science changed my mind, and hopefully I’ll continue revising my beliefs as new data trickles in.

As for uncertainty, most people dislike it when it concerns their own health. I know I do. But I believe it’s better to accept uncertainty than to pretend it doesn’t exist. If you have one drink per evening, you find that it reduces stress (a health benefit that’s overlooked in the literature), and you wake up refreshed, then it’s probably a healthy choice. “Probably” is the best I can do here.

Cheers, and thanks for reading!

---

Appendix: More on methodology.

This is a brief, somewhat impressionistic sketch of the statistical approach used in each of the studies I reviewed. None of this is critical to following what I wrote in the main text.

The burden of proof study

Informally speaking, when a bunch of studies estimate the association between alcohol intake and some disease outcome, one study will differ from the next in the estimated strength of the association. That’s one kind of uncertainty. In addition, each study’s estimate of risk will include its own uncertainty interval. That’s another kind of uncertainty. Dai and colleagues combined the two to create a dose-response curve for each disease outcome as well as an uncertainty band around it. To quantify risk of harm, they focused on the 5th percentile of that distribution – in other words, the weakest expression of risk still reasonably consistent with the data. (An analogous approach was used for protective effects.)

The Alcohol Intake and Health Study

After experts chose the most credible meta-analysis for each health outcome, George and colleagues combined risk estimates from these meta-analyses ­­with data on drinking patterns in the U.S. (adjusted upward for underreporting) as well as national disease, injury, and mortality statistics. Using these inputs, the researchers estimated the lifetime risks of each health outcome among American drinkers (as opposed to lifelong abstainers), and combined them to estimate the overall health risks of different levels of alcohol consumption. A key point to emphasize here – and the reason the findings can’t be compared in a granular way to what Dai and colleagues reported – is that this isn’t a cohort study, in which each participant’s drinking behavior and health outcomes were recorded, but rather a statistically-based model of population-level risk.