Where Did the J-Curve Go?

In large population groups, in general, the risk of premature death is higher the more people drink. Observations over longer periods found that the people with the lowest risk—who therefore live the longest—are not those who drink no alcohol at all, but those who have a few drinks per week. The curve on a coordinate system of mortality and drinking is therefore not an upward-slanting straight line, but looks more like a J.

How our understanding of this curve has developed, I describe in my video see.nf/jcurve, but in short, we are probably dealing here with an error due to the sick-quitter effect, which results from the systematic misclassification of former drinkers as lifelong abstainers. For the same reason, some studies find higher mortality among former smokers than among those who keep smoking. It is not that health worsens due to abstinence, but that poor health led to abstinence.

When researchers looked at the results again and corrected the attribution error—treating former drinkers like lifelong abstainers—the J-curve disappeared. In other words, the relationship between mortality and drinking was more even, meaning: more alcohol, higher mortality, without low alcohol intake having been protective against it.

Mendelian Randomization

The fact that abstainers were confused with people who had stopped drinking because of illness draws our attention to the problem of reverse causality. We have also found it in studies that supposedly show that people who tend to sit around more and watch TV are less healthy. Does TV make you sick, or do you watch more TV because you are sick? This is one of the reasons why intervention studies with control groups provide more reliable results in the hierarchy of evidence than observational studies, into which both reverse causality and confounding factors can creep. One example: You may be more likely to drink your glass of wine with a salad than with a cheeseburger; perhaps that is why people believed wine was healthier. Moderate alcohol consumption also points to a higher socioeconomic status, which in itself is a predictor of a longer life. But sometimes randomized controlled trials are difficult to carry out. It would be impractical—not to mention the ethical dimension—to randomize participants so that, for example, they smoke a pack of cigarettes every day for a few decades, so in health care some decisions have to be based on observational data. However, we now have an additional tool at our disposal: “nature’s clinical trial,” Mendelian randomization.

Where randomized controlled trials are not possible or appropriate, Mendelian randomization provides reliable results about cause and effect. As already mentioned, debunking HDL cholesterol as a protective factor is due in part to Mendelian randomization studies in which subjects whose genetic makeup had randomly given them higher HDL levels for life did not suffer less from heart disease over the years. The scientists didn’t have to assign anyone at random; a particular sperm and egg had already done that when they met. Is there a way to study people who have been randomized since conception to not drink much? Astonishingly, yes.

As already mentioned, two enzymes in the liver convert alcohol into carbon dioxide and water, but in this process acetaldehyde is produced, the toxic intermediate product of metabolism that can cause nausea and unpleasant hot flashes. So if you are born either with a slower variant of the enzyme that breaks down acetaldehyde, or with a superfast variant of the enzyme that forms acetaldehyde, acetaldehyde can accumulate in the blood, and that makes alcohol consumption a relatively unpleasant experience for a lifetime. This means people are born who are less likely to drink. Do they have an increased risk of heart disease, as the original J-curve from the observational studies suggests? No, their risk of heart disease is lower. From this it can be inferred that even people who drink lightly to moderately are doing something for their health if they further reduce their alcohol consumption.

Actual Randomization

In some observational studies, even after checking for confounding factors and reverse causality, a J-curve still appears. Possibly the genetic variants associated with reduced alcohol intake have a protective function independent of that. That would weaken the evidentiary value of the Mendelian randomization data. As a result, a furious, bitter controversy arose in the medical literature, in which some scientists vehemently hold on to the J-curve (especially those funded by industry), while others reject any alleged positive effects as outdated wishful thinking or as the work of spin doctors for the alcohol industry. What we need, the National Institutes of Health concluded, is a randomized controlled trial to settle this question once and for all. Cue the Moderate Alcohol and Cardiovascular Health Trial.

For this study, they needed thousands of volunteers over age 50 at increased risk of cardiovascular disease. Half of them were supposed to abstain from alcohol for six years, and the other half were to have a small drink every day. In which group would there be more heart attacks, strokes, and cases of diabetes, or deaths? However, there was a problem: The NIH research team violated basic rules of US policy by letting major corporations like Anheuser-Busch and Heineken cover most of the 100 million dollars in costs for the study. The study director and other NIH officials insisted that the industry funders would play no role in the design of the study, but the public learned from the New York Times that, as could be foreseen, things had gone differently. The newspaper had obtained access to emails with the help of the Freedom of Information Act. Critics complained, for example, that the study outcomes did not take cancer and heart failure into account, even though alcohol is known to be linked to both. The trial was abruptly halted after an internal investigation showed that, as the NIH director at the time put it, “so many lines were crossed that we frankly were shocked.” The Moderate Alcohol and Cardiovascular Health Trial was over.

Even if impartial funders could be found, it would currently not be ethically justifiable to urge people, by randomization, to drink alcohol. Soon after the original study was designed, the Global Burden of Disease Study published the most comprehensive assessment of the overall effect of alcohol consumption based on results from nearly 700 data sources. The conclusion, confirmed by the WHO and the World Heart Federation, was clear and unambiguous: “The safest level of alcohol is none.”