Alcohol and Cancer
I want to start by thanking each of you for reading my newsletters and, in some cases, sharing them with others. Thanks to your support, the subscriber list is over 250 and still growing.
I'm grateful too for the comments and requests you've e-mailed me. This week's newsletter is a response to one of those requests.
Back in June, I wrote about a new study suggesting that small amounts of alcohol consumption can damage your brain. (I disagreed with the conclusion.) Later, one of you asked about links between light drinking and other health problems. This is a great question, since many of us drink (and studies show that we've been drinking more, on average, during the pandemic).
The problem I'll discuss this time is cancer. There are tons of studies on alcohol-cancer connections, including a recent review that focuses specifically on low levels of drinking. This review illustrates the broader theme that statistical factoids can be quite intimidating, until they're examined closely.
The fact that stats have this kind of power is no secret to people who deal in intimidation. Here's an example from a March 2016 interview:
Donald Trump: "It's 27 percent, could be 35 percent of Muslims would go to war – the hatred is tremendous, Chris."
Chris Wallace: "...you're saying 250-300 million Muslims would go to war against us?"
Donald Trump: "Why don't you take a look at the Pew poll that came out recently or fairly recently, where I think the number – I mean, I could be corrected, it's whatever it is – but it's something like 27 percent are, you know, really very militant about going after things. And you'll have to look at it. They did a very strong study. And let's see what it says. But it's a very significant number..."
Actually, it wasn't a significant number. It wasn't insignificant either. It just plain didn't exist. Neither the Pew Research Center nor anyone else had conducted such a poll, or even posed a question that could’ve yielded such data (story here). In short, a statistic was fabricated to drive home a point.
Stats can be intimidating even when the source doesn't intend them to be. A statistic is cited, maybe it's described as "significant", and we feel like we can't question it, because it's mathematical, objective, precise, calculated by means of some mysterious process, etc. In fact, we can always question statistical factoids, and we should. I'm going to illustrate that right now.
In 2018, a team of South Korean researchers published a meta-analysis in the journal Cancer Research Treatment that examined studies on links between low levels of alcohol consumption and various cancers. (If you're not familiar with the term, a meta-analysis is a type of review in which statistical methods are used to combine and analyze the results of individual, previously published studies on the same topic. In this case, 60 studies were reviewed.)
Some people might find stats from a meta-analysis especially intimidating, because they’re drawn from many studies rather than just one. The 60 studies in this particular meta-analysis provide data on more than 9 million people. When I saw those numbers, I felt a little intimidated myself. How could you question results obtained from so much data? Well, in this case the problems turned out to be fairly obvious.
For their meta-analysis, the authors reviewed studies on associations between drinking and the development of cancer in 23 different parts of the body. Studies were included in the review if participants were healthy and cancer-free at the beginning of data collection. In addition, each study had to include non-drinkers as well as members of at least one of the following groups: very light drinkers (less than half a drink per day, on average), light drinkers (half a drink to one drink per day, on average), and moderate drinkers (one to two drinks per day, on average). Later I'll say more about how a "drink" was defined.
Here are the main findings:
—Compared to non-drinkers, very light and light drinkers each have a slightly higher risk of developing breast cancer (women only), and a slightly lower risk of lung cancer. Light drinkers also have a slightly higher risk of colorectal cancer and malignant melanoma.
—The findings for moderate drinkers were the same as for light drinkers, except that moderate drinkers also have a slightly higher risk of three cancers (oropharyngeal, laryngeal, esophageal), and a slightly lower risk of three cancers (kidney, thyroid, and hematologic).
If you're a very light to moderate drinker, you might be worried about the higher risks that were reported. (And, you might be wondering how to balance them against the lower risks.) The next part of this newsletter, which describes three themes that emerge from the findings, should be reassuring.
1. The impact on risk was small in every case.
Let's take a worst-case scenario approach and look at the two examples involving the greatest calculated risk:
(a) Among all types of cancer that had been examined in at least two studies, the highest average risk ratio was 1.98 for esophageal cancer among moderate drinkers. (Risk ratio in this case is the percentage of moderate drinkers who developed esophageal cancer divided by the percentage of non-drinkers who developed it.) In other words, the incidence of esophageal cancer was nearly twice as high in moderate drinkers as in non-drinkers. However, the absolute rates of this type of cancer are not high. Judging from other data given in the reviewed studies, one might predict that among generally healthy non-smokers, roughly 2 in 1,000 moderate drinkers will develop esophageal cancer in their lifetimes, versus 1 in 1,000 non-drinkers. I can't tell you how to feel about these numbers, but from the perspective of individual risk, an increased risk observed among others of 1 in 1,000 seems small.
(b) Regarding the type of cancer with the highest base-rate (breast cancer, which occurs among roughly 12 in 100 women during their lifetimes), the highest risk ratio was 1.13 among women who drink moderately. That is, compared to women who don't drink, 13% more of the women who drink moderately developed breast cancer. Translating this 13% figure into individual risk estimates is difficult, because it's drawn from 15 studies that varied in their approach to measurement and control of health-related variables, but from the data given here's my best estimate: Among generally healthy women who don't smoke, roughly 44 in 1,000 moderate drinkers will develop breast cancer in their lifetimes, versus 39 in 1,000 non-drinkers. In other words, an increased risk of 5 in 1,000 (or 1 in 200) for moderate drinkers. Whether you consider that figure small or not, here's some additional context:
2. The estimated impacts on risk were almost certainly overestimated for all cancers.
This is probably the most encouraging news I can share if your alcohol consumption ranges from very light to moderate.
One reason for overestimated risks stems from the way drinking was measured. Self-report surveys was used (since researchers can't follow people around), and, depending on the study, participants were asked every six months, or two years, or five years, or whatever, to describe their own alcohol consumption (sometimes for a typical day, sometimes for a typical week, etc). As you can imagine, this approach to measurement turns out to be a hot mess.
For one thing, even the most careful instructions to survey respondents couldn't have yielded precise measurements (e.g., "How many drinks do you have per day on a typical week? One drink would be a 12 ounce beer, 5 ounces of wine, or 1.5 ounces of distilled spirits.") I'm sure you can see some of the causes of imprecision, including the fact that people don't typically measure how much they drink, and the fact that within each category (wine, beer, distilled spirits), different types and brands of alcohol vary widely in the amount of ethanol they contain. There were also scaling problems: In some studies, participants chose among options like 1 drink a day, 2 drinks a day, etc., while in others they chose among ranges (1-2 drinks a day, 2-3 drinks a day, etc.).
Arguably the most serious problem with the standard approach to measurement is that people aren't very accurate about recalling how much they drink, on average, when looking back over periods of months or years. You can probably tell me exactly how much you drank yesterday (unless you drank way too much), but in these studies people are asked about periods of time ranging from six months to over six years.
Why bring this up? Because the measurement of alcohol consumption isn't just imprecise; it leads to errors that tend to overestimate long-term risks among people who don’t drink much. Here’s the cause of this well-known problem: When people in these studies mistakenly estimate the amount of alcohol they consume, the overestimates will outweigh the underestimates, on average, because there’s not much room for underestimation. If you claim that you averaged half a drink per day during the past two years, any underestimation will be small, because half a drink is very close to zero. There's much more room for overestimation. Thus, the average alcohol consumption for people in groups such as "light drinker" tends to be greater than the researchers assume, and long-term risks will be overestimated for those groups. (In other words, when you see stats indicating the long-term risks of "light" drinking, they’re actually risk estimates for some greater extent of alcohol consumption.)
A second cause for overestimation, which the authors noted, is that not all variables of relevance were controlled for in the studies they reviewed. For example, I was startled to discover that some studies didn't control for tobacco use. (There's a lot of evidence that drinking and smoking are positively correlated – people who do more of one tend to do more of the other.) When a study shows that moderate drinkers have a greater risk of cancer in areas like the esophagus, but their smoking behavior isn't recorded, we have no way of knowing whether the slight increase in risk is due to the fact that some of them smoke. (The math is tricky, but smoking is more prevalent among moderate drinkers than among non-drinkers, and the difference may be enough to account for differences in the rates of certain cancers.)
A third cause for overestimation is publication bias – i.e., the tendency for journals to favor publication of significant findings (as opposed to nonsignificant ones), which in turn diminishes researchers' willingness to even try to publish nonsignificant findings. The problem with publication bias is that distorts our understanding of a topic. Scientific knowledge gradually accumulates from studies that report significant findings rather than the ones that obtain nonsignificant ones, even though both kinds of findings are part of the "truth". Publication bias can exacerbated in meta-analyses unless it's addressed, because the researchers pool the effects of published studies and may thus overestimate the true extent of the effects. This is particularly worrisome to researchers who study risk, because publication bias naturally leads to overestimates of risk, regardless of topic. The authors failed to deploy any of the statistical and non-statistical ways of addressing this problem that are often used.
In sum, the 2018 meta-analysis revealed significant links between low levels of alcohol consumption and cancer, but the effects were small, based on imprecise approaches to measurement, and almost surely overestimated risk. (Sadly, they almost surely overestimated the benefits too, for the same reasons.) The meta-analysis had other limitations that I haven’t discussed. The main point is that we have 60 studies here, with data from over 9 million people, but still no clear reason to be worried – or intimidated – by the stats.
(Keep in mind too that studies on cancer have identified other behaviors that introduce a much greater degree of risk than anything reported in this meta-analysis. If you drink heavily, chain smoke, eat excessive amounts of fried food, and spend lots of time in the sun without sunscreen, well, change your behavior!)
Thanks for reading! I'm going to set this aside now and have a small glass of wine. (3 ounces…maybe 4. I don't plan to measure.).