Main

In their commentary on the Obesity Paradox, Dixon et al.1 support the notion that a body mass index (BMI) in the obese range (over 30 kg m–2) often provides a survival advantage over having a BMI in the normal range (18.5–24.9 kg m–2). There are numerous meta-analyses that have shown the opposite—that mortality is increased in adults in the obese compared with normal weight range.2 Results on the association between overweight (25.0–29.9 kg m–2) and all-cause mortality are more mixed, but not really germane to arguments about the Obesity Paradox, which is about obesity. We are concerned by claims published by Dixon et al. that obesity provides mortality advantage and normal weight may have a causal negative effect on mortality. Cause is recognized by epidemiologists as well as other types of scientists as a powerful term, not to be bandied about without strong justification. It assumes that were one able to assign the cause randomly to individuals in a sample, the expected incidence of the outcome would be different according to assignment. In a situation in which an attribute causes something as monumental as a mortality advantage, it would follow that efforts would be made by clinicians, public health workers and policy makers to increase that attribute, that is obesity, in the population.

These authors do not advocate obesity as a public health goal, nevertheless, their message may interfere with public health efforts to prevent obesity. They seem to imply that if obesity can be shown to provide a survival advantage under any circumstances, then concern about avoiding obesity is misplaced. They provide a list of diseases and conditions for which there are in their words ‘consistent reports’ to indicate an advantage of (class I) obesity over normal weight. We will not review the copious, often mixed and controversial literature surrounding this claim, but we do point out that decades of literature on the topic point to the problem of complex confounding obscuring mechanistic or causal associations between obesity and all-cause mortality.

Recognized sources of bias include reverse causality, selective survival and bias caused by conditioning on a variable affected by BMI.3,4 These biases may be magnified in circumstances in which the sample under study is restricted to individuals suffering from serious disease or in individuals with diseases for which obesity is a cause. Much of the evidence for the apparent obesity paradox comes from just those types of studies.5, 6, 7, 8, 9 Reverse causation can be pervasive in samples with prevalent disease if subjects experience disease-induced weight loss prior to the assessment of body weight as a study exposure. This weight loss can make it appear that those who are normal or underweight are more likely to die than those with greater BMI if the disease caused both weight loss and increased mortality rates. Although not shown in all studies, some have shown that accounting for reverse causation can attenuate or eliminate an observed inverse association between BMI and mortality.3,10 Selective survival biases the BMI–mortality association if those who have more disease or a more advanced level of a disease and are obese were less likely to survive to participate in a study than individuals with normal weight. So the study sample is enriched with the type of obese subjects who have less susceptibility to die from the disease under study, as the susceptible are already deceased.

By its nature, selective survival is more of a problem in studies of the elderly, whereas studies limited to subjects with prevalent disease are particularly prone to a different type of selection bias that can impact the BMI–mortality association.4,11 This latter type of bias is perhaps best illustrated by Lajous et al.11 who posit a scenario in which there are only two causes of diabetes: one is obesity and the other a genetic factor unrelated to body weight. This genetic factor, independent of the diabetes effect, is deadly. In this scenario, all persons with diabetes are either normal weight with the genetic factor that results in higher mortality or obese (with or without the genetic factor). Restricting an analysis to diabetic patients creates an inverse association between obesity and the genetic factor that is not causal. This association attenuates, or potentially reverses the direction of a positive causal association between BMI and mortality. In this example, the factor causing both disease and death could be environmental as well as genetic, the distinguishing characteristic being that it is an alternative cause of a disease that can also be caused by obesity.

These three types of bias (reverse causation, selective survival bias and selection bias) cause tricky methodological problems that need to be considered carefully before concluding that obesity is protective. All are very difficult to study, so it is not surprising that the pertinent literature is mixed. Furthermore, there are situations in which survival is likely increased by obesity—being stranded for an extended period of time without access to food is an obvious, though extreme example. More relevant, having a potentially life-threatening disease that interferes with appetite and causes severe weight loss is a scenario in which obesity might be protective. This protection would be manifest during the weight loss phase of the disease, preventing or prolonging starvation. Patients with a severe disease, such as chronic kidney disease requiring hemodialysis, are in a complex nutritional state, and it is not surprising that it would be inappropriate to apply recommendations concerning body weight that originate from public health guidance meant for reasonably healthy individuals.

In these types of extreme situations we agree with Dixon et al.1 that personalization of the nutritional approach is useful, and we have no quarrel with their suggestion that more research is needed to better understand individual differences in responses to body weight and fatness. However, we disagree that general guidance for BMI in healthy populations should be abandoned. A personalized ideal weight may be useful in individualized medical care, but is not a practical public health concept. The widespread nature of the problem of obesity and the complexity of the myriad of societal and environmental factors that encourage obesity require population-level approaches. The BMI categories currently used12 and recently endorsed in the 2013 NHLBI/AHA/ACC/TOS Guideline for the Management of Overweight and Obesity in Adults2 are critical to a useful population-wide approach. These guidelines do address individual variation because they provide relevant BMI ranges rather than one optimal BMI value. Fine tuning of these ranges to specific individuals may be useful in clinical settings, but should not interfere with current efforts to prevent and treat obesity.

We are very concerned that confusion about the Obesity Paradox could result in decreased attention to population-level solutions and may result in busy physicians dropping from their care a healthy lifestyle treatment plan for their obese patients. It would be to the detriment of health promotion and disease prevention for medical professionals to abandon emphasis on encouraging their obese patients to obtain and maintain a healthy weight. This need for appropriate treatment includes obese patients who are without cardiometabolic risk factors, and we disagree with a treatment paradigm13 that does not recommend weight loss intervention for obese patients unless they have an established obesity-related chronic disease. Our research has shown that over a 9-year follow-up of obese adults are 4.5 times more likely to develop metabolic syndrome compared with normal weight adults.14 There is also abundant evidence that insulin sensitivity, lipids and blood pressure can be improved by weight loss in obese individuals.15, 16, 17 Further, obesity is associated with increased risk of problems like knee osteoarthritis18 and social issues that are not dependent on cardiometabolic risk factors.

Studies of the Obesity Paradox are susceptible to a list of complex sources of bias, and it may not currently be possible to untangle these effects in studies of humans. Bias affecting the BMI–mortality relationship in samples of elderly or diseased individuals are likely even more powerful than in studies of the general population. Although mortality observations can be puzzling, the effects of obesity on the incidence of diabetes,19 cardiovascular disease,2 certain types of cancer,20 sleep apnea,21 osteoarthritis of the knee22 and other morbidities19 show impressive consistency. For most obese individuals, weight reduction is a strong medicine providing an improved risk profile, decreased morbidity and reduced rates of mortality from several causes. The fact that weight reduction is difficult does not change that.