Weight-height relationships and body mass index: Some observations from the diverse populations collaboration

Body mass index (BMI, weight (kg)/ height (m)(2)) is the most widely used weight-height index worldwide. This universal use of BMI assumes that the rationale for its use is universally applicable. We examine two possible rationales for using BMI as a universal measure. The first rationale is that BMI is strongly correlated with weight, but is independent of height. The second rationale is that BMI correctly captures the relationship between weight and height, which implies that the slope of log weight regressed on log height is 2. We examined the weight-height relationship in 25 diverse population samples of men and women from the US, Europe, and Asia. The analysis included 72 subgroups with a total of 385,232 adults aged 25 years and older. Although BMI was highly correlated with weight in all studies, a significant, negative correlation between BMI and height was found in 31 out of 40 subgroups of men (r = -0.004 to -0.133) and 32 of 32 groups of women (r = -0.016 to -0.205). When log weight was regressed on log height, the 95% confidence intervals (CI) of the slopes did not include 2 in 25 out of 40 male subgroups. The summary estimate of the slopes across studies of men was 1.92 (95% CI, 1.87-1.97). For women, slopes were lower than 2 in 28 of 32 subgroups with a summary estimate of 1.45 (95% CI, 1.39-1.51). In most of the populations, BMI is not independent of height; weight does not universally vary with the square of height; and the relationship between weight and height differs significantly between males and females. The use of a single BMI standard for both men and women cannot be justified on the basis of weight-height relationships.