Associations between obesity and health conditions may be overestimated if self‐reported body mass index is used

In numerous studies, body mass index (BMI) is estimated using self-reported height and weight (1). In a recent systematic review, Connor Gorber et al. (2) showed that selfreported BMI is most often lower than measured BMI, i.e. that some obese individuals are misclassified as being non-obese. Prevalence of obesity is therefore underestimated (2). Moreover, it appears intuitive that any associations between obesity defined using self-reported BMI and health conditions will be underestimated because of this misclassification. However, misclassification due to self-report is not random. Moderately obese individuals are more likely to side with non-obese than severely obese, the latter having a BMI which is further away from the obese/non-obese cut-off value. Thus, a counterintuitive consequence of this nonrandom misclassification of self-reported BMI is that any association between obesity and health conditions is likely to be overestimated. Consider the following hypothetical example. Selfreported height and weight were gathered in a cohort of 1000 women. After 10 years of follow-up, self-reported BMI was shown to be strongly associated with the risk of diabetes (upper part of Table 1). The relative risk (RR) of diabetes for obese women (self-reported BMI ≥ 30 kg m − 2 ) was 20.0/10.3 = 1.95. Suppose now that height and weight were also measured and that measured BMI was, on average, 1 unit higher than obesity reviews (2007) 8 , 373–374