If you’re overweight and looking for an excuse not to go to the trouble of losing weight, you can’t do better than this: An article in the journal PLoS Medicine finds that overweight people who intentionally lose weight have a higher mortality rate than overweight people who just stay overweight. In fact, those who lose weight even have a higher mortality rate than those who gain weight! (Although those with stable weight have the lowest mortality rate.)
The study surveys Finnish twins in 1975, and asked overweight people if they “intended” to lose weight. Their weight was re-checked in 1981, and their mortality was followed until 1999. To summarize the results:
|Mortality Hazard Ratios|
|Lost Weight||Stable Weight||Gained Weight|
|Intended to Lose Weight||1.87||0.84||0.93|
|Did not intend to Lose Weight||1.17||1.00||1.58|
In other words, an overweight person who intends to lose weight, and succeeds in doing so, is 87% more likely to die than an overweight person who makes no effort to lose weight, and in fact maintains a stable weight. However, an overweight person who intends to lose weight and fails to do so is slightly less likely to die than someone who never intends to lose weight in the first place. Those who have no intention of losing weight but do so anyway are have a slightly higher risk of mortality, but not nearly as much as those who lost weight on purpose. Those who have no intention of losing weight and in fact gain weight have a substantially higher risk — but still less than those who lost weight on purpose. In short, if this study is right, the worst thing for an overweight person to do is to try to lose weight and succeed; the best thing to do is to try to lose weight and fail. Not trying at all is somewhere in between.
Not only is this the best possible result for those who like excuses (“Doctor, I’m trying to lose weight, but I’m not succeeding and that’s good!”), it also flies in the face of most previous work in the field, which of course shows that losing weight (or at least, not being overweight) is good. The article is by actual scientists and appears in a peer-reviewed journal, so it cannot be dismissed out of hand. However, given the weight of the evidence (sorry for the pun!) on the other side of the question, it can’t be accepted uncritically, either.
I can think of several possible reasons why the authors might have found the results they did that would still leave the conventional wisdom mostly intact:
- Perhaps those who are losing weight on purpose are doing dangerous things to accomplish the loss. For example, not all diets are safe, even if they are successful. (Three doctors told me that if the only way I could lose weight was to use the Adkins diet, I was better off staying fat.) Also, some people may exercise beyond their body’s capabilities in an effort to lose weight.
- The study counted as overweight anyone with a body mass index (BMI) greater than 25. This is a fairly loose standard; while many experts say 25 is the “ideal” BMI, few would label as “overweight” someone with a BMI of 25.1. Yet, this study appears to do so. Furthermore, there is some evidence that the longevity-maximizing BMI is actually more like 26.5, rather than 25. In this study, less than 10% had a BMI greater than 30 (i.e., were “obese”), and in fact those included had a median BMI of 26.7 — and “losing weight” was defined as reducing BMI by at least 1. So for about half the people in the sample, losing weight meant moving away from the longevity-maximizing BMI — so we would actually expect an increase in mortality. So perhaps it’s not that losing weight is bad for people who are actually overweight — rather, the people in the sample weren’t overweight enough to benefit from weight loss, and many of those who lost weighted actually dropped below their ideal weight.
- The statistical model they used controlled for sex, age, current smoking in 1981, hypertension, physical activity, life satisfaction, work status, and income. Patients with previous heart attacks, diagnosed angina, or diabetes as of 1981 were excluded. However, the model did not control for family history of these conditions, nor did it exclude people with these conditions. It is quite possible that family history of heart attacks and diabetes — both of which are correlated with obesity — could both increase the probability of death by 1999, and cause people to want to lose weight. In other words, patients in the study know their family history, and if it contains heart attacks or diabetes, they know they are at higher risk of death. They also know — or at least, believe — that the risk of these diseases can be reduced by losing weight. So they report an “intention” to lose weight. Among those who report an intention to lose weight, those with a family history of obesity-related diseases are more motivated to succeed, and thus over-represented among those who both intend to lose weight and actually do so. However, weight is only one factor in those diseases; even after losing weight they still have a higher risk of those diseases — and thus early death — than people without that family history (even if their risk is less than what it was when they were overweight). The study then finds that people who intend to lose weight and succeed have a higher risk of mortality. But it’s not that the weight loss is causing the mortality, it’s precisely the opposite: The higher risk of mortality is known to the patients in advance, and thus causes the weight loss. (This is what econometricians call “reverse causation;” in fact, a brief Google search confirms that this has occurred previously in research on weight loss.) The weight loss may reduce the risk of mortality, but if so it does not reduce it enough to bring it down to the level of the general population.
- Participants were asked once — in 1975 — whether they “intended” to lose weight (and if so, how). The determination as to whether they had done so was made in 1981, when they were not asked whether they had carried out their intentions. Lots of things can change in six years. As an extreme example, someone who got cancer in 1978 (say) may have lost a lot of weight from the disease by 1981, and be more likely to die by 1999. Perhaps such a person, knowing about cancer risk, reported an intention to lose weight in 1975, but failed to do so — until getting cancer three years later. This is of course a purely hypothetical example — and purely speculative. The point I’m trying to make is that the data may be insufficient to draw conclusions about the link between weight loss and mortality.
- Another possibility is that they just got a weird draw from a random distribution. If you look at enough studies, this is bound to happen. Analogy: If you flip 10 coins, it’s very unlikely that you’ll get 10 heads, or even 9 — but it is certainly possible. If you flip 10 coins once, there’s only a 1.07% chance that you get 9 or more heads. But if 500 people flip 10 coins each, there’s a 99.54% change that somebody gets 9 or more heads. It’s like that with these studies, too. Assume that the conventional wisdom is right — that weight loss actually does reduce mortality. Even if that’s true, it’s still random in individual cases — even if weight loss reduces mortality on average, some people are going to lose weight and die early anyway, and some people are going to stay fat and live a long time anyway. Not as many, but some. Now, if you do a study on weight loss and mortality, it’s like flipping coins — it’s very likely you are going to find data consistent with the theory. But if 500 people do studies like that, somebody is going to find data inconsistent with the theory. But if it’s just one or a few studies, that doesn’t mean the theory is wrong — any more than one person flipping 9 out of 10 heads means that the “theory” of coin-flipping produces a 50% chance of heads.
The complete article is here. There is also a synopsis, a “patient summary,” and a note on “perspectives” which raises some other issues with the study, none of which I think are as serious as the issues raised above. Personally, I think the most likely culprit is reverse causation.