SEARCH

SEARCH BY CITATION

Introduction

  1. Top of page
  2. Introduction
  3. Methodological Biases in BMI and Mortality Analysis
  4. What Have Other Studies Shown?
  5. Generalizability vs. Validity
  6. Obesity Paradox
  7. Clinical and Public Health Implications
  8. Acknowledgments
  9. References

There is indisputable evidence from epidemiologic and clinical studies that being overweight and obese elevates the risk of developing debilitating and costly chronic diseases, including hypertension, hypercholesterolemia, type 2 diabetes, cardiovascular diseases (CVD), and cancer [1]. Nonetheless, the relationship between body mass index (BMI) and mortality remains the subject of much debate. A recent meta-analysis concluded that compared to those of normal weight (BMI < 25.0), overweight individuals (BMI 25.0-29.9) had a significantly lower mortality risk [2]. Even Class 1 obesity (BMI 30-34.9) was associated with marginally reduced mortality. In this Perspective, we discuss why this finding is likely to be largely an artifact of methodological limitations and what the clinical and public health implications may be.

Methodological Biases in BMI and Mortality Analysis

  1. Top of page
  2. Introduction
  3. Methodological Biases in BMI and Mortality Analysis
  4. What Have Other Studies Shown?
  5. Generalizability vs. Validity
  6. Obesity Paradox
  7. Clinical and Public Health Implications
  8. Acknowledgments
  9. References

Although total mortality is a straightforward endpoint, epidemiologic studies of body weight and mortality are particularly prone to two major sources of bias: reverse causation and confounding by smoking [3]. Reverse causation is a concern when a lower body weight is the result of an underlying illness through the disease process itself, or through a conscious effort to lose weight motivated by a clinical diagnosis. Furthermore, this potential for bias increases with older age as chronic diseases accumulate and loss of lean mass associated with frailty becomes common. While exclusion of participants with known disease at baseline addresses much of this bias, many chronic conditions such as pulmonary and neurodegenerative diseases remain undiagnosed for years. There is no perfect solution to deal with this problem; however, excluding deaths occurring early in follow-up can also help to reduce reverse causation.

Confounding by smoking is another major source of bias in BMI–mortality analysis. Differences in intensity, inhalation, frequency, and duration, coupled with smoking's very strong association with mortality risk and association with lower body weight make simply adjusting for smoking status in a statistical model an inadequate control for its confounding. To avoid this residual bias, it is now standard practice to conduct the analyses restricted to never smokers.

These methodological biases are exacerbated when a wide comparison group (BMI 18.5 to <25) is used because this group (especially the lower end of normal weight) contains not only those who are lean and active, but also heavy smokers, individuals with chronic diseases, and frail elderly individuals. In populations undergoing nutritional transitions (e.g., China and India), low BMI groups also include those affected by malnutrition and infectious diseases. Comparing overweight and obese groups to this heterogeneous stratum seriously underestimates their relative mortality risk.

As mentioned above, it is critical to conduct stratified analysis by smoking status. For example, in the Prospective Studies Collaboration (Figure 1), there was an approximately linear relationship among the never smokers, while a nonlinear J-shaped relationship persisted among smokers [4]. Unfortunately, the meta-analysis by Flegal et al. only included published summaries of data, and thus did not present individual-level or subgroup data for non-smokers or younger individuals (<65 years old).

image

Figure 1. All-cause mortality at ages 35–79 years versus BMI in the range 15–50 kg/m2, by smoking status (excluding the first 5 years of follow-up), Reproduced with permission from Prospective Studies Collaboration [4]. “Relative risks at ages 35–79 years, adjusted for age at risk, sex, and study, were multiplied by a common factor (i.e., floated) so that the mean for all participants (including ex-smokers and anyone with missing smoking data) matches the European rate at ages 35–79 years in 2000. Results for ex-smokers and those with missing smoking data not shown (but are, taken together, only slightly above those for never smokers). Note that many smokers were at only limited risk, since they had not smoked many cigarettes during early adult life, or had stopped shortly after the baseline survey. Risk is indicated on an additive rather than multiplicative scale. The estimates for 35–50 kg/m2 are based on limited data, so lines connecting to those estimates are dashed. Floated mortality rates shown above each square and numbers of deaths below. Area of square is inversely proportional to the variance of the log risk. Boundaries of BMI groups are indicated by tick marks. 95% CIs for floated rates reflect uncertainty in the log risk for each single rate.” (Whitlock G, Lewington S, Sherliker P, et al. Body-mass index and cause-specific mortality in 900,000 adults: collaborative analyses of 57 prospective studies. Lancet 2009;373:1083–1096. Figure 6. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]

Download figure to PowerPoint

What Have Other Studies Shown?

  1. Top of page
  2. Introduction
  3. Methodological Biases in BMI and Mortality Analysis
  4. What Have Other Studies Shown?
  5. Generalizability vs. Validity
  6. Obesity Paradox
  7. Clinical and Public Health Implications
  8. Acknowledgments
  9. References

Flegal et al. emphasize that a strength of their meta-analysis is their use of standard BMI categories. While the separate sensitivity analyses included extrapolated estimates from several large studies, the main analysis excluded many large cohorts or consortia (Table 1) [4-9] which had sufficient statistical power to allow for the analysis of finer BMI categories and assessment or nonlinear associations. Including only studies with broad BMI cut-points therefore resulted in an over-representation of smaller clinical populations, high-risk patients with particular illnesses or living in metabolic wards, and the elderly. In the excluded studies (>6 million individuals), the lowest mortality was frequently observed among those with BMI 22.5–25, especially among healthy nonsmokers (Table 1). These studies provide convincing evidence that optimal BMI for longevity is below a BMI of 25. Of note, the cohorts with self-reported (5, 7) and measured BMI (8, 9) produced generally similar results (Table 1). In Flegal's meta-analysis, the estimates for overweight and Class I obesity were almost identical for self-reported or measured BMI values and their 95% confidence intervals largely overlap.

Table 1. Summary of findings from publications on BMI and mortality among the total population and healthy never smokers, omitted from the meta-analysis by Flegal et al. (2)
StudyTotal subjectsTotal deathsMean ageMean follow-upRef BMIAll-cause mortality RR (95% CI) by BMI categoryAll-cause mortality RR (95% CI) by BMI category
Total populationHealthy never smokers
Overweight IOverweight IIObeseOverweight IOverweight IIObese
  1. Ref = reference group, CI = confidence interval, BMI = body mass index, W = women, M = men, E = East, S = South.

  2. a

    Median.

National Cancer Institute (NCI) cohort consortium (5)1,462,958160,08758 ya10 ya22.5-24.9BMI: 25-27.4 W: 1.05 (1.03-1.07); M: 0.97 (0.96-0.99)BMI: 27.5-29.9 W: 1.14 (1.11-1.17); M: 1.05 (1.02-1.07)BMI: 30-34.9 W: 1.31 (1.28-1.34); M: 1.18 (1.15-1.21)BMI: 25-27.4 W: 1.09 (1.05-1.14); M: 1.06 (1.01-1.12)BMI: 27.5-29.9 W: 1.19 (1.14-1.24); M: 1.21 (1.14-1.28)BMI: 30-34.9 W: 1.44 (1.38-1.50); M: 1.44 (1.35-1.53)
Asia cohort consortium (6)1,141,609120,75853.9 y9.2 y22.6-25.0BMI: 25.1-27.5 E. Asian: 0.98 (0.95-1.01); S. Asian: 0.98 (0.84-1.13)BMI: 27.6-30.0 E. Asian: 1.07 (1.02-1.12); S. Asian: 0.94 (0.77-1.16)BMI: 30.1-32.5 E. Asians: 1.20 (1.10-1.32); S. Asians: 1.03 (0.77-1.39)BMI: 25.1-27.5 E. Asians: 1.00 (0.95-1.06); S. Asians: 0.97 (0.82-1.16)BMI: 27.6-30.0 E. Asians: 1.11 (1.04-1.20); S. Asians: 0.94 (0.74-1.19)BMI: 30.1-32.5 E. Asians: 1.27 (1.12-1.43); S. Asians: 1.01 (0.73-1.41)
Prospective Studies Collaboration (4)894,57666,55246 y13 yNABMI: 15-25b W: 0.80 (0.75-0.80); M: 0.79 (0.76-0.82); Mortality lowest at BMI∼22.5-25BMI: 25-50b W: 1.26 (1.23-1.30); M: 1.32 (1.29-1.36)BMI: 15-25b W: 0.87 (0.78-0.97); M: 0.87 (0.78-0.97); Mortality lowest at BMI∼22.5-25BMI: 25-50b W: 1.27 (1.22-1.32); M: 1.44 (1.36-1.53)  
Cancer Prevention Study II (7)1,046,154201,62257 y14 y23.5-24.9------------BMI: 25-26.4 White W: 1.07 (1.01-1.13); White M: 1.04 (0.98-1.10); Black W: 0.90 (0.71-1.15); Black M: 1.20 (0.86-1.68)BMI: 26.5-27.9 White W: 1.10 (1.04-1.17); White M: 1.09 (1.02-1.16); Black W: 0.97 (0.77-1.22); Black M: 1.13 (0.81-1.59)BMI: 30.0-31.9 White W: 1.30 (1.22-1.39); White M: 1.32 (1.21-1.45); Black W: 1.17 (0.92-1.48); Black M: 1.29 (0.87-1.90)
European Prospective Investigation into Cancer and Nutrition (EPIC) (8)359,38714,72351.5 y9.7 y23.5-24.9BMI: 25-26.4 W: 1.01 (0.92-1.11); M: 0.91 (0.84-0.99)BMI: 26.5-27.9 W: 1.07 (0.97-1.18); M: 0.96 (0.88-1.04)BMI: 30-34.9 W: 1.17 (1.07-1.29); M: 1.24 (1.14-1.35)BMI: 25-26.4 W: 1.00 (0.87-1.15); M: 0.89 (0.73-1.07)BMI: 26.5-27.9 W: 1.12 (0.98-1.30); M: 1.05 (0.86-1.27)BMI: 30-34.9 W: 1.25 (1.09-1.43); M: 1.48 (1.22-1.79)
Korean Cancer Prevention Study (9)1,213,82982,372W: 49.4 y; M: 45 y12 y23.0-24.9BMI: 25-26.4 W: 0.98 (0.94-1.03); M: 0.97 (0.94-1.00)BMI: 26.5-27.9 W: 1.02 (0.97-1.08); M: 0.99 (0.95-1.03)BMI: 30-31.9 W: 1.16 (1.06-1.28); M: 1.20 (1.08-1.34)BMI: 25-26.4 W: 1.0 (0.9-1.0); M: 1.0 (0.9-1.1)BMI: 26.5-27.9 W: 1.0 (1.0-1.1); M: 1.1 (1.0-1.2)BMI: 30-31.9 W: 1.2 (1.1-1.3); M: 1.5 (1.3-1.9)

Generalizability vs. Validity

  1. Top of page
  2. Introduction
  3. Methodological Biases in BMI and Mortality Analysis
  4. What Have Other Studies Shown?
  5. Generalizability vs. Validity
  6. Obesity Paradox
  7. Clinical and Public Health Implications
  8. Acknowledgments
  9. References

It has been argued that exclusion of participants with CVD and cancer at baseline produce biased associations between BMI and mortality because the resulting sample would not reflect the US population. However, these exclusions are necessary to obtain valid estimates of mortality risk. For example, in a study of cigarette smoking and mortality, if patients with CVD and cancer at baseline were included, the effects of smoking in the general population would be seriously underestimated, as the “nonsmoking” patients would include ex-smokers who quit due to illness but remain at an elevated risk of early death. Clearly, validity is the overriding objective of epidemiologic studies, because non-valid results cannot be generalized to any populations, including its own participants. From a public health perspective, our ultimate goal is to identify the optimal BMI to reduce the risk of chronic disease and premature mortality, rather than pure statistical prediction.

Obesity Paradox

  1. Top of page
  2. Introduction
  3. Methodological Biases in BMI and Mortality Analysis
  4. What Have Other Studies Shown?
  5. Generalizability vs. Validity
  6. Obesity Paradox
  7. Clinical and Public Health Implications
  8. Acknowledgments
  9. References

Obesity has been associated with improved survival in patients with existing chronic diseases, including congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD), chronic kidney disease, and other wasting conditions—a phenomenon referred to as “reverse epidemiology” or the “obesity paradox” [1]. In these patients, their blood pressure and blood cholesterol also tend to fall and thus higher blood pressure or higher blood cholesterol has also been associated with better survival. One likely explanation for the “reverse epidemiology” is that chronically ill patients who lose weight and become frail are the ones who tend to die early. Clinically, weight gain is not a desirable recommendation for most chronically ill patients, who are often already overweight or obese to begin with.

Clinical and Public Health Implications

  1. Top of page
  2. Introduction
  3. Methodological Biases in BMI and Mortality Analysis
  4. What Have Other Studies Shown?
  5. Generalizability vs. Validity
  6. Obesity Paradox
  7. Clinical and Public Health Implications
  8. Acknowledgments
  9. References

Flegal et al. suggest that their meta-analysis may “help to inform decision making in the clinical setting.” However, their conclusion suggesting a reduced mortality among the overweight and Class I obese patients may seriously mislead clinicians and the lay public. While not all overweight and obese adults presently display signs of metabolic dysfunction or disease, this state deemed “metabolically healthy obesity” has been shown to be only transitory for most [10]. Maintaining a healthy weight through diet and physical activity should remain the cornerstone for the prevention and treatment of chronic diseases, and is critical to reduce skyrocketing health care costs. In addition to monitoring body weight, monitoring changes in waist circumference and the amount of weight gain since young adulthood is also important for enjoying a long and healthy life.

References

  1. Top of page
  2. Introduction
  3. Methodological Biases in BMI and Mortality Analysis
  4. What Have Other Studies Shown?
  5. Generalizability vs. Validity
  6. Obesity Paradox
  7. Clinical and Public Health Implications
  8. Acknowledgments
  9. References