Mortality Risk of Obesity and Underweight Is Overestimated with Self-Reported Body Mass Index

Abstract

Obese persons (those with a body mass index [BMI] ≥30 kg/m2) tend to underestimate their weight, leading to an underestimation of their true (measured) BMI and obesity prevalence.1,2 In contrast, underweight people (BMI <18.5 kg/m2) tend to report themselves heavier, resulting in a higher BMI compared with measured BMI and an underestimation of underweight prevalence.1 Less is known about biases in the estimate of mortality risk associated with these body weight categories using self-reported data. It has been shown that the mortality risk of obesity based on self-report can be overestimated, ie, biased away from the null.3,4 Underweight is associated with an increased mortality risk5 and, because underweight persons tend to overestimate BMI, one might intuitively expect that the mortality risk of underweight based on self-reported BMI would be underestimated, ie, biased toward the null (the opposite of the effect in obese persons). Is that a misleading intuition? Consider a cohort of 1000 men, of whom 10% were underweight at baseline, 70% had normal weight or were overweight, and 20% were obese, based on measured BMI. After 5 years, 36 men died (Table). The relative risk (RR) of death was higher in underweight (RR = 2.0) and obese men (RR = 1.5) compared with that in normal weight and overweight men combined. We assumed that the mortality risk was higher in the lowest (<17.5 kg/m2) category among underweight persons and in the highest BMI category (≥31 kg/m2) among obese persons.TABLE: Hypothetical Cohort with 1000 Men Providing Measured and Self-Reported Weight and Height to Compute Body Mass Index (BMI)Participants also reported their weight and height at baseline to compute self-reported BMI. We assume that underweight men (measured BMI <18.5 kg/m2) overestimated their BMI by 1 unit, on average. Consequently, men with measured BMI 17.5–18.4 kg/m2 were classified as “normal weight” based on self-reported BMI, while in truth they were underweight (eAppendix, eFigure, https://links.lww.com/EDE/A735). Thus, using self-reported data, the prevalence of underweight was underestimated. Men with self-reported BMI <18.5 kg/m2 had a measured BMI <17.5 kg/m2. Therefore, these men had a higher mortality compared with men with true BMI <18.5 kg/m2 (8% vs. 6%). We assume that obese men underestimated their BMI by 1 unit, on average. Therefore, men with true BMI 30.0–31.0 kg/m2 were classified as normal weight/overweight based on self-reported BMI although they were in truth obese. Using self-reported data, the prevalence of obesity was underestimated. Men with self-reported BMI ≥30.0 kg/m2 had a measured BMI ≥31.0 kg/m2. Accordingly, they had a higher mortality compared with men with true BMI ≥30.0 kg/m2 (6% vs. 4.5%). Using self-reported BMI to define body weight categories, the risk of underweight relative to normal/overweight was (4/50)/[(2 + 21 + 3)/(50 + 700 + 100)] = 2.6, which was higher than the RR (2.0) obtained with measured BMI. The RR of obesity was (6/100)/[(2 + 21 + 3)/(50 + 700 + 100)] = 2.0, which was also higher than the RR (1.5) obtained with measured BMI. Although self-reports lead to an overestimation of BMI by underweight persons and an underestimation by obese persons,1 the mortality (or disease) risk in both obese and underweight persons is likely overestimated when self-reported data are used to categorize people. Our assumptions are simplistic. The errors in weight estimations depend on sex, age, and other characteristics,2,6 and the direction of the bias will depend on how people in the various categories estimate their weight. The possible effects of exposure misclassification on the estimation of RR are complex and barely generalizable.7 Still, as shown, the effects of such misclassification can be counterintuitive. Direction and magnitude of bias should be evaluated carefully for each situation.7 The overestimation of obesity risk based on self-reports has been demonstrated3,4 and recently corroborated in a meta-analysis.8 Research is needed to confirm the bias entailed by defining underweight with self-reported data. David Faeh Lucienne Roh Institute of Social and Preventive Medicine (ISPM), University of Zurich, Zurich, Switzerland, [email protected] Fred Paccaud Arnaud Chiolero Institute of Social and Preventive Medicine (IUMSP), University Hospital Center of Lausanne, Lausanne, Switzerland