The trouble with observational studies of diet and health that don't include any intervention is the large number of omitted variables that are likely to be correlated with the variables that are directly studied. Still, it is worth knowing for which things one can say:Either this is bad, or there is something else correlated with it that is bad. When multivariate regression is used, one might be able to strengthen this toEither this is bad, or there is something else bad correlated with it that is not completely predictable from the other variables in the regression.This statistical point is directly relevant to the results of the Nurses’
Miles Kimball considers the following as important:
This could be interesting, too:
Eric Crampton writes Quality control at the New Zealand Medical Journal
Paul Krugman writes The Party of No Ideas
Lawrence H. Summers writes When Hindsight Is Not 20/20
Bertrand Badré writes Making the Most of Emerging Economies
The trouble with observational studies of diet and health that don't include any intervention is the large number of omitted variables that are likely to be correlated with the variables that are directly studied. Still, it is worth knowing for which things one can say:
Either this is bad, or there is something else correlated with it that is bad.
When multivariate regression is used, one might be able to strengthen this to
Either this is bad, or there is something else bad correlated with it that is not completely predictable from the other variables in the regression.
This statistical point is directly relevant to the results of the Nurses’ Health Study, Nurses’ Health Study II and the Health Professionals Follow-up Study analyzed by Dariush Mozaffarian, Tao Hao, Eric B. Rimm, Walter C. Willett, and Frank B. Hu in their 2011 New England Journal of Medicine article "Changes in Diet and Lifestyle and Long-Term Weight Gain in Women and Men."
Below are some of the results, as reported by Jane Brody in a 2011 New York Times article. All the quotations below are from her article.
First, the data hint that focusing only on calories is a mistake; there were many relationships between types of food eaten and amount of weight gain. Here is the view of the first author of the study:
Also untrue, Dr. Mozaffarian said, is the food industry’s claim that there’s no such thing as a bad food.
“There are good foods and bad foods, and the advice should be to eat the good foods more and the bad foods less,” he said. “The notion that it’s O.K. to eat everything in moderation is just an excuse to eat whatever you want.”
Second, although exercise is disappointing as a way of losing weight (see "Julia Belluz and Javier Zarracina: Why You'll Be Disappointed If You Are Exercising to Lose Weight, Explained with 60+ Studies), exercise may be quite helpful in not gaining weight:
The study showed that physical activity had the expected benefits for weight control. Those who exercised less over the course of the study tended to gain weight, while those who increased their activity didn’t. Those with the greatest increase in physical activity gained 1.76 fewer pounds than the rest of the participants within each four-year period.
One reason this may be true is that exercise helps reduce the insulin resistance of muscles. See "Obesity Is Always and Everywhere an Insulin Phenomenon" for a discussion of the important role of insulin resistance in weight gain. However,
“Both physical activity and diet are important to weight control, but if you are fairly active and ignore diet, you can still gain weight,” said Dr. Walter Willett, chairman of the nutrition department at the Harvard School of Public Health and a co-author of the study.
One of the biggest red flags pointing to omitted variables in the study is this:
Participants who were overweight at the study’s start tended to gain the most weight ...
That is, whatever made people gain weight before the period of time a statistical analyses focused on what likely to make them keep gaining weight, and controlling for the factors the study had data doesn't account for all the factors that made people gain weight in the past and are likely to keep making them gain weight in the future. There is an important lesson in this: if you are gaining weight, you need to do something different than you have been doing.
This is a lesson that can be applied to our nation as a whole (and almost all nations): average levels of obesity are rising, so we need to do something different than we have been doing! Stop for a moment and try to describe for yourself what the approach to weight-control has been for the last few decades and clearly label that in your mind as something that is not working.
What about particular foods? Here are the estimates of how well eating different foods predicted weight gain or weight loss in pounds over a four-year period:
- french fries +3.4
- potato chips +1.7
- sugar-sweetened drinks +1
- red meats +.95
- processed meats +.93
- potatoes +.57
- sweets and desserts +.41
- refined grains +.39
- other fried foods +.32
- 100% fruit juice +.31
- butter +.3
- fruits, vegetables and whole grains ≈0
- skim milk ≈0
- full-fat milk or cheese ≈0
- nuts and nut butter <0
- yogurt -.82
This is an interesting list to compare to the insulin index of various kinds of food and drink that I discuss in "Forget Calorie Counting; It's the Insulin Index, Stupid." Let's see if the results in the table above that couldn't be predicted from the insulin index sound like they point to omitted variables. Here are two categories: worse than expected based on the insulin index and better than expected. In both cases, let me frame things as what the omitted variables would have to be like to make the results no mystery.
Would have to be correlated with bad omitted variables to square the results of this observational study with the predictions of the insulin index:
- red meat
- processed meat
Would have to be correlated with good omitted variables to square the results of this observational study with the predictions of the insulin index:
- skim milk
Think now of the type of people who eat red meat, processed meats and butter compared to the types of people who eat yogurt and fruit and drink skim milk. Even after controlling for variables that were in the analysis, people eating yogurt (even sugary yogurt), fruit and drinking skim milk could easily be doing something else right. And people eating red meat, processed meat and butter could easily be doing something else wrong. This is more likely to be true because all the variables in the analysis are measured imperfectly, so each variable in the analysis only partially controls for the thing it is supposed to measure.
Jane Brody does point to an intriguing possibility for why yogurt might look as helpful for weight loss as it does:
That yogurt, among all foods, was most strongly linked to weight loss was the study’s most surprising dietary finding, the researchers said. Participants who ate more yogurt lost an average of 0.82 pound every four years.
Yogurt contains healthful bacteria that in animal studies increase production of intestinal hormones that enhance satiety and decrease hunger, Dr. Hu said. The bacteria may also raise the body’s metabolic rate, making weight control easier.
Currently, I don't eat much yogurt, but I do take a supplement made by Steven Gundry's company with probiotics in it, plus another complementary supplement that is supposed to make things nice and comfy in my gut for good bacteria:
If the effect of yogurt's probiotics on metabolic rate is a big factor in yogurt looking good, looking at effects on metabolic rate makes sugar and starches look bad—including the sugar in sweetened yogurt as opposed to plain yogurt:
But, consistent with the new study’s findings, metabolism takes a hit from refined carbohydrates — sugars and starches stripped of their fiber, like white flour. When Dr. David Ludwig of Children’s Hospital Boston compared the effects of refined carbohydrates with the effects of whole grains in both animals and people, he found that metabolism, which determines how many calories are used at rest, slowed with the consumption of refined grains but stayed the same after consumption of whole grains.
Here is one last hint of omitted variables: the predictions one can make based on what types of alcoholic beverages an individual drinks. Jane Brody writes:
Alcohol intake had an interesting relationship to weight changes. No significant effect was found among those who increased their intake to one glass of wine a day, but increases in other forms of alcohol were likely to bring added pounds.
On average, beer drinkers just aren't the same types of people as people who drink wine! Beer does have a higher insulin index than wine, but they are both quite low on the insulin index. So drinking wine should be better judging from the insulin index alone, but I suspect the size of the extra weight gain for beer drinkers has a lot to do with omitted variables.
By and large, the results of this analysis of data from the Nurses' Health Study and its sister studies jive quite well with the recommendations I give in "Forget Calorie Counting; It's the Insulin Index, Stupid." And where they don't, it is easy to see possible omitted variables.
Recommendations where there were surprises relative to the insulin index: There is no harm in keeping one's meat consumption low. Indeed, because meat is at a middling level on the insulin index rather than a low level, I recommend eating meat sparingly in "Why a Low-Insulin-Index Diet Isn't Exactly a 'Lowcarb' Diet." The analysis described here has skim milk and whole milk as equal, while the insulin index suggests skim milk is worse. That, plus the fact that whole milk tastes a lot better suggests that you should stick with whole milk if you drink milk at all. Whole fruit has many good nutrients, so I have always recommended eating more of it than the insulin index alone would suggest. Plain, full-fat, unsweetened yogurt is a great thing to eat anyway, if you are OK with dairy.
Thus, in my recommendations, the one place I depart most from what this analysis of Nurses' Health Study and its sister surveys is that treat butter as healthy. I may be wrong, but my actions are based on that view. Here is one way I might be totally right. Bread is very high on the insulin index. People who eat butter are likely to eat more bread. And that is unlikely to be fully captured by the the imperfect measure of bread-eating in the study.
Importantly, in our society, it is not socially acceptable to eat butter straight. So butter eating is highly correlated with eating other things, many of which are quite unhealthy. If half the population ate butter straight while the other half didn't eat butter at all, the estimated predictive power of butter for weight-gain in "Changes in Diet and Lifestyle and Long-Term Weight Gain in Women and Men" might have been very different.
Don't miss these other posts on diet and health and on fighting obesity:
Also see the last section of "Five Books That Have Changed My Life."