Seth's Blog

A new BMJ paper looks at Cuban health before and after the economic crisis of 1991-1995, when the Cuban economy nose-dived. There wasn’t enough gasoline for cars. so bike riding greatly increased. In addition, people ate less. What effect did these changes (more exercise, less eating) have on health?

You know what is supposed to happen: Better health. Walter Willett, the Harvard epidemiologist, wrote a commentary about the study that concluded “The current findings add powerful evidence that a reduction in overweight and obesity would have major population-wide [health] benefits.” In other words, Willett said that what happened supports conventional beliefs.

But it didn’t. In several ways, what happened contradicts conventional beliefs.

1. A popular belief is that exercise causes weight loss. However, the percentage of “physically active individuals” doubled from 1985 to 2010 (from about 30% to 60%). In spite of this, the prevalence of obesity considerably increased (from about 13% to 18%) at the same time. Apparently exercise is considerably less important than something else. I have never heard a public health advocate say this.

2. A graph showing rates of heart disease, cancer, and stroke (the three main killers) over the period showed no change in rates of cancer and stroke. In spite of big changes in both exercise and obesity. The rate of heart disease stayed constant during the period when obesity went down. It steadily dropped during the period of time when obesity went up. Apparently the factors that control obesity and the factors that control heart disease are quite different (contradicting the usual view that exercise reduces both).

3. There is no simple connection between diabetes and obesity. During the economic crisis, when the prevalence of obesity went down by half (from 15% to 7%) and exercise greatly increased, the prevalence of diabetes slightly increased. Only after the crisis did the usual correlation (more obesity, more diabetes) emerge.

4. The only lifestyle factor to have its conventional effect: smoking. When you stop smoking, you gain weight is the usual belief (which I also believe). The data definitely support this connection. A huge reduction in the fraction of people who smoke (from 30% to 10%) did not reduce cancer but did coincide with a great increase in obesity.

5. Cubans are doing something right, as shown by the considerable decrease in heart disease and diabetes deaths. Apparently they are also more health-conscious, as shown by much higher rates of exercise and much lower rates of smoking. (Assuming that cigarettes did not become too expensive.) They are getting fatter, too, but apparently that is less damaging than we are told.

Willett and the authors of the study look at subsets of the data and use theories about “time-lag” to draw reassuring conclusions. In fact, large portions of the data are not easily explained by conventional ideas, as I’ve shown. You can look at the data many ways, but to me the study makes two main points. 1. During a period when everyone was forced to do what doctors recommend (exercise more, eat less), health did not improve. 2. During a period (post-crisis) when obesity got steadily worse, health improved (heart disease rates went down, cancer stayed the same, diabetes mortality went down). Cuba is too poor for the improvement to be due to better high-tech modern medicine. Taken together, these findings suggest we should be more skeptical of what we are told by doctors and health experts such as Willett.

A recent paper from the Cleveland Clinic reports more than a dozen studies that add up, say the authors, to the conclusion that red meat and other meats cause heart disease at least partly by increasing trimethylamine-N-oxide (TMAO), which is made from carnitine by intestinal bacteria. Meat, especially red meat, is high in carnitine.

The results were reported all over the world, including the New York Times. There are several reasons to question the conclusion:

1. The association between meat and heart disease is weak. An epidemiological paper from the Harvard Nurses Study found estimated reductions in heart disease on the order of 10-20% when a “healthy” food was substituted for meat. Conclusions about causality (eating Food X causes Disease Y) based on the Harvard Nurses Study have predicted wrongly over and over when tested in experiments, so even this weak association is questionable. A 2010 meta-analysis found no association between red meat consumption and heart disease. The absence of any correlation is surprising because red meat is widely believed to be unhealthy. People who eat more red meat would presumably do more other “unhealthy” things. (Perhaps the error rate of the underlying epidemiology is high. Errors push associations toward zero.)

2. Within the Cleveland paper, the associations between carnitine and TMAO and heart disease are weak. For example, people with the greatest sign of heart disease (“triple” angiographic evidence of heart disease) had only slightly more carnitine in their blood (about 15% more) than people with the least sign of heart disease. (Maybe it is peak levels of carnitine rather than average levels that matter.)

3. A 1996 epidemiological study (via Chris Kresser) that looked at the correlates of various “healthy” habits among people especially interested in health (e.g., they shop at health food stores) found no detectable effect of being a vegetarian. For example, vegetarians had the same all-cause mortality as non-vegetarians. Other factors were associated with reduced mortality, including eating wholemeal bread daily and eating fruit daily. This study looked at a large number of people (about 11,000) for a long time (17 years), so I consider the lack of difference (vegetarians versus non-vegetarians) strong evidence against the idea that modest amounts of meat are harmful.  (And I am going to start eating wholemeal bread in small amounts.)

I don’t dismiss the paper. Among people who eat more than modest amounts of meat, there may be something to it. Now and then epidemiology turns up a powerful risk factor — something associated with a risk increase by a factor of 4 or more (people at a high level of the risk factor get the disease at least four times more often than people at a low level of the factor). History shows that such correlations are likely to tell us something about causality. With weaker correlations (such as the correlation between red meat and heart disease), it is much more a guessing game.

To me, the important clue about heart disease is that it is very low in both Japan and France, much lower than in countries with high rates of heart disease. The two countries that have little in common besides the fact that in both people eat a lot more fermented food than in most places. In France, they drink wine, eat stinky cheese and yogurt. In Japan, they eat miso, pickles, and natto. Maybe fermented food protects against heart disease.

In a review of Salt Sugar Fat by Michael Moss, a new book about the food industry, David Kamp writes:

The term “bliss point” . . . is used in the soft-drink business to denote the optimal level of sugar at which the beverage is most pleasing to the consumer. . . .

The “Fat” section of “Salt Sugar Fat” is the most disquieting, for, as Moss learns from Adam Drewnowski, an epidemiologist who runs the Center for Obesity Research at the University of Washington, there is no known bliss point for fat — his test subjects, plied with a drinkable concoction of milk, cream and sugar, kept on chugging ever fattier samples without crying uncle. This realization has had huge implications in the food industry. For example, Moss reports, the big companies have come to understand that “cheese could be added to other food products without any worries that people would walk away.”

By “fat” Moss means animal fat (the fat in cheese, for example). I haven’t seen the book but I’m sure Moss doesn’t consider the possibility that “there is no known bliss point for fat” because people should be eating much more animal fat. In other words, it is hard to detect the bliss point when people are suffering from severe fat deprivation.

My view of how much animal fat I should eat changed abruptly when I found that large amounts of pork fat made me sleep better. One day I ate a lot of pork belly (very high fat) to avoid throwing it away. That night I slept much better than usual. I confirmed the effect experimentally. Later, I found that butter (instead of pork fat) made me faster at a mental test. This strengthened my belief that I should eat much more animal fat than countless nutrition experts have said. (Supporting data.)

My sleep and mental test evidence was clear and strong (in the sense of large t value). The evidence that animal fat is bad (based on epidemiology) is neither. That is one reason I trust what I found rather than what I have been told.

Another reason I trust what I found the fact that people like the taste of fat. That evolution has shaped us to like the taste of something we shouldn’t eat makes no sense. (Surely I don’t have to explain why this doesn’t mean that sugar — not available to prehistoric man — is good for us.) In contrast, it is entirely possible that nutrition experts have gotten things backwards. Epidemiology is a fledgling science and epidemiologists often make mistakes. Their conclusions point in the wrong direction. Here is an example, about the effect of beta-carotene on heart disease:

Epidemiology repeatedly found that people who consumed more beta-carotene had less heart disease. When the idea that beta-carotene reduces heart disease was tested in experiments, the results suggested the opposite: beta-carotene increases heart disease.

“Fat will become the new diet food” (via Hyperlipid).

For a long time, nutrition experts have told us to replace saturated fats (solid at room temperature) with polyunsaturated fats (liquid at room temperature). One polyunsaturated fat is omega-6. Omega-6 is found in large amounts in corn oil, soybean oil, and most other vegetable oils (flaxseed oil is the big exception). According to Eat Drink and Be Healthy (2001) by Walter Willett (and “co-developed with the Harvard School of Public Health”), “replacing saturated fats with unsaturated fats is a safe, proven, and delicious way to cut the rates of heart disease” (p. 71). “Plenty of proof for the benefits of unsaturated fats” says a paragraph heading (p. 71). Willett failed to distinguish between omega-3 and omega-6.

A recent study in the BMJ shows how wrong Willett (and thousands like him) were. This study began with the assumption that omega-3 and omega-6 might have different effects, so it was a good idea to try to measure the effect of omega-6 separately.

They reanalyzed data from a study done in Sydney Australia from 1966 to 1973.The study had two groups: (a) a group of men not told to change their diet and (b) a group of men told to eat more omega-6 by eating more safflower oil (and reducing saturated fat intake, keeping overall fat intake roughly constant). The hope was that the change would reduce heart disease, as everyone said.

As these studies go, it was relatively small, only about 500 subjects. The main results:

Compared with the control group, the intervention group had an increased risk of all cause mortality (17.6% v 11.8% [emphasis added]; hazard ratio 1.62 (95% confidence interval 1.00 to 2.64); P=0.051), cardiovascular mortality (17.2% v 11.0%; 1.70 (1.03 to 2.80); P=0.037), and mortality from coronary heart disease (16.3% v 10.1%; 1.74 (1.04 to 2.92); P=0.036).

A 50% increase in death rate! The safflower oil was so damaging that even this small study yielded significant differences.

The authors go on to show that this result (omega-6 is bad for you) is supported by other studies. Walter Willett and countless other experts were quite wrong on the biggest health issue of our time (how to reduce heart disease, the #1 cause of death).

The first evidence linking exercise and health was a study of London bus workers in the 1950s. The drivers, who sat all day, had more heart attacks than the ticket takers on the same buses, who were on their feet all day. It was a huge advance — evidence, as opposed to speculation. The results were taken countless times to imply that exercise reduces heart attacks but epidemiologists understood there were dozens of differences between the two jobs. For example, driving is more stressful than ticket taking. Maybe stress causes heart attacks.

The first time I learned about this study, I focussed on two differences. The ticket takers were more exposed to morning sunlight (on the top deck of double-decker buses) and they were on their feet much more. Maybe both of those things — morning sunlight exposure and standing a lot — improve sleep. Maybe better sleep reduces heart attacks. The London data were not consistent with the claims of aerobic exercise advocates because the ticket takers did nothing resembling aerobic exercise.

Later I discovered that walking an hour/day normalized my fasting blood sugar levels — another effect of “exercise” (but not aerobic exercise). I had data from only one person (myself), but it was experimental data. The treatment difference between the two sets of data being compared (no walking versus walking) was much sharper, in contrast to most epidemiology. I am sure the correlation reflects cause and effect: Walking roughly an hour/day normalized my blood sugar. This wasn’t obvious. The first thing I tried to lower my fasting blood sugar levels was a low-carb diet, which didn’t work. I discovered the effect of long walks by accident.

A recent meta-analysis combined several surveys that measured the correlation of how much you sit with other health measures. The clearest correlation was with diabetes: People who sit more are more likely to get diabetes. Comparing the two extremes (most sitting with most standing), there was a doubling of risk. Because people who stand more walk more, this supports my self-experimental findings.

I found pure standing (no walking), or leisurely (on-off) walking, did not lower fasting blood sugar (which I measured in the morning). After I noticed that walking an hour lowered blood sugar, I tried slacking off: wandering through a store or a mall for an hour. This did not lower fasting blood sugar. I concluded it had to be close-to-nonstop walking. Someday epidemiologists will measure activity more precisely — with Fitbits, for example. I predict the potent part of standing will turn out to be continuous walking. Long before that, you can see for yourself.

Tyler Cowen links to a paper by Frank Lichtenberg, an economist at Columbia University, that tries to estimate the benefits of drug company innovation by estimating how much new drugs prolong life compared to older drugs. The paper compares people equated in a variety of ways except the “vintage” (date of approval) of the drugs they take. Does taking newer drugs increase life-span? is the question Lichtenberg wants to answer. He concludes they do. He says his findings “suggest that two-thirds of the 0.6-year increase in the life expectancy of elderly Americans during 1996-2003 was due to the increase in drug vintage” — that is, to newer drugs.

An obvious problem is that Lichtenberg has not controlled for health-consciousness. This is a standard epidemiological point. People who adopt Conventional Healthy Behavior X (e.g., eat less fat) are more likely to adopt Conventional Healthy Behavior Y (e.g., find a better doctor) than those who don’t. For example, a study found that people who drink a proper amount of wine eat more vegetables. Another reason for a correlation between conventionally-healthy practices is mild depression. People who are mildly depressed are less likely to do twenty different helpful things (including “eat healthy” and “find a better doctor”) than people who are not mildly depressed. (And mild depression seems to be common.) Perhaps doctors differ. (Lichtenberg concludes there are big differences.) Perhaps better doctors (a) prescribe more recent drugs and (b) do other things that benefit their patients. Lichtenberg does not discuss these possibilities.

A subtle problem with Lichtenberg’s conclusion that we benefit from drug company innovation is that drug-company-like thinking — the notion that health problems should be “solved” with drugs — interferes with a better way of thinking: the notion that to solve a health problem, we should find out what aspects of the environment cause it. I suppose this is why we have Schools of Public Health — because this way of thinking, advocated at schools of public health, is so incompatible with what is said and done at medical schools. Public health thinking has a clear and impressive track record — for example, the disappearance of infectious disease as a major source of death. There are plenty of other examples: the drop in lung cancer after it was discovered that smoking causes lung cancer, the drop in birth defects after it was discovered that folate deficiency causes birth defects. Thinking centered on drugs has done nothing so helpful. Spending enormous amounts of money to develop new drugs shifts resources away from more cost-effective research: about environmental causes and prevention. Someone should ask the directors of the Susan K. Komen Foundation: Why “race for the cure”? Wouldn’t spending the money on prevention research save more lives?

• more black tea, less diabetes (cross-national comparison).Earlier research supports the idea that black tea reduces diabetes.
• Nassim Taleb praises an obscure French novel (translated into English) called The Opposing Shore. Says it is about “the antechamber of anticipation”.
• Natto consumption in different parts of Japan and correlation with Vitamin K levels and hip fracture risk. The correlation between natto consumption and hip fracture risk is weak but significant and cannot be explained by differences in wealth or overall soybean consumption. This is excellent epidemiology.
• Five reasons butter is a superfood. I was unaware that conjugated linoleic acid (of which butter is a good source) inhibits tumors.
• Fish oil and brain injury. Read this before you have a brain injury.

Thanks to Anne Weiss and Dave Lull.

A new epidemiological study followed about 16,000 people in Europe for about 12 years and focused on their dairy intake. Did the ones who came down with diabetes eat differently from those who didn’t?

The paper begins:

Current dietary guidelines for prevention of diabetes aim at substituting SFAs [saturated fatty acids] with unsaturated fatty acids. However, conventionally held notions that all SFAs, including those from dairy products, are detrimental to health have recently been challenged.

The shift of evidence (dairy less bad than previously believed) supports my view that what’s good for the brain (I found butter was good for my brain) is likely to be good for the rest of the body. The paper’s main conclusion is the possible protective value of cheese and yogurt:

This large prospective study found no association between total dairy product intake and diabetes risk. An inverse association of cheese intake and combined fermented dairy product [= cheese, yogurt, and "thick fermented milk"] intake with diabetes is suggested.

The combined fermented dairy association was not large in size (a risk reduction of 12%) but was significant (barely). When your main finding is barely significant you have no hope of using your data to explain it so the new information essentially stops there. The results support my view that fermented foods are unusually healthy.

In response to these findings, the director of research at Diabetes UK said, “This study gives us no reason to believe that people should change their dairy intake in an attempt to avoid [diabetes].” Wow. It is as if a prominent physicist said the earth was flat.

Thanks to Elizabeth Molin.

Epidemiology has lots of critics. In this article, for example, it is called “lying on a grand scale.” Every critique I have read has ignored history. Epidemiologists have been right about two major issues: 1. Heavy smoking causes lung cancer. 2. Folate deficiency causes birth defects. In both cases, the first evidence was epidemiological. Another example is John Snow’s conclusion about the value of clean water. In my experience, epidemiologists often overstate the strength of their evidence (as do most of us) but overstatement is quite different from having nothing worth saying.

Let’s look at an example. Many people think osteoporosis is due to lack of calcium. Bones are made of calcium, right? The epidemiology of hip fractures is clear. In spite of the conventional idea, the rate of hip fracture has been highest in places where people eat a lot of calcium, such as Sweden, and lowest in places where they eat little, such as Hong Kong. (For example.) In other words, the epidemiology flatly contradicted the conventional idea. This was apparently ignored by nutrition experts (everyone knows correlation does not equal causation) who advised millions of people, especially women, to take calcium supplements  to avoid osteoporosis. Millions of people followed (and follow) that advice.

Thanks to a recent meta-analysis we now know that experiments and better data firmly support the earlier epidemiology, which suggested that calcium supplements are dangerous. Here are its main conclusions:

In meta-analyses of placebo controlled trials of calcium or calcium and vitamin D, complete trial-level data were available for 28,072 participants from eight trials of calcium supplements and the WHI CaD participants not taking personal calcium supplements. . . .Calcium or calcium and vitamin D increased the risk of myocardial infarction (relative risk 1.24 (1.07 to 1.45), P = 0.004) and the composite of myocardial infarction or stroke (1.15 (1.03 to 1.27), P = 0.009). . . . A reassessment of the role of calcium supplements in osteoporosis management is warranted.

If the epidemiology had been taken more seriously, many heart attacks might have been avoided.

Is this an “anecdote” — a single example — proving nothing? Here’s how you can check. Randomly select a meta-analysis of epidemiological studies. Thousands have been done. Then ask if the results summarized in the meta-analysis appear random. Better yet, randomly pick two meta-analyses. Suppose the first summarizes 5 studies and the second summarizes 6. If the 11 results were shuffled together, how well could you assign them correctly?