Flegal et al. reply.

Abstract

Listen

Translate article

We thank Dr. Stevens for her observations (1). We agree with Stevens that standardized body mass index (BMI; weight (kg)/height (m)2) cutpoints can aid comparisons among studies examining different populations. The standard BMI cutpoints of 18.5, 25, and 30, indicating underweight (<18.5), normal weight (18.5–<25), overweight (25–<30), and obesity (≥30), were developed by the National Heart, Lung, and Blood Institute (2) and the World Health Organization (WHO) (3) and were reiterated in the 2013 obesity guidelines (4). Since their inception, these standards have been widely used in studies of BMI and risk of death and appear to be well accepted. In a previous systematic search (5), we found well over 100 published studies of weight and mortality as of 2012 that had reported results using those standard categories, including approximately half of the studies published since 2000. Turner et al. (6) used those BMI standards as their example of an approach to categorization that used well-recognized, published boundaries. In our article (7), we did not suggest that a new or augmented set of standards was needed. Rather, we emphasized that use of the current standard BMI groupings in analysis would avoid issues arising from ad hoc and post hoc selection of BMI categories. Finer BMI categories are not clearly necessary and can create problems of interpretation. Dr. Stevens is correct that we did not mention the narrow BMI ranges discussed by the 2002 WHO expert consultation on BMI in Asian populations (8). The WHO expert consultation was directed towards policy and intervention. It recommended use of the standard BMI cutpoints noted above and also identified “further potential public health action points (23.0, 27.5, 32.5, and 37.5 kg/m2)” (8, p. 157). When the PubMed searches previously described (5) were repeated after being limited to studies published from 2004 through August 2014, they identified 6,627 potential studies of weight and mortality. Of these, only 26 (0.3%) cited the 2002 WHO consultation. Of those 26 studies, only 1 (9) used the array of finer cutpoints suggested by the 2002 consultation. Thus, the WHO consultation appears to have had relatively little impact on studies of weight and mortality. Dr. Friedman's editorial (10) suggested dividing the normal-weight reference category into 3 subdivisions. Stevens comments that “an additional BMI cutpoint between 18.5 and 23, as suggested by Friedman, may help to separate persons who are in the normal-weight range due to healthy diet and exercise behaviors from those who have anorexia associated with disease” (1, p. 1128). Our article did not recommend changes to the reference category of normal weight. As is shown in Table 4 of our article (7, p. 293), excluding BMI <20 from the reference category produced almost no effect on the hazard ratios for overweight and obesity. Even though the risk was slightly higher for persons with BMI <20, the prevalence of that category was quite low, so the effect was negligible. In a previous investigation (11), we found that the primary effect of the reference category on estimates of excess mortality was seen in changing from a reference category of BMI 23–<25 to a reference category of 22–<25. Further lowering of the lower bound below 22 had almost no effect. Stevens (1) and Friedman (10) note that risk may vary within the normal-weight category, but as our article shows (7), risk also varies within the overweight category, with no difference in risk between a BMI of approximately 23–25 and a BMI of approximately 25–27.5. The summary hazard ratios were 0.97 (95% confidence interval: 0.96, 0.98) for men and 1.03 (95% confidence interval: 1.00, 1.05) for women (7). This effect might be masked by comparing the standard overweight category with a nonstandard reference category. The existing standardized categories can be used as part of the analysis to facilitate comparisons between studies. Other analytical approaches, such as splines, are more suited to exploration of the shape of the relationships.