Prediction of measured weight from self-reported weight was not improved after stratification by body mass index.

Objective: Self-reported weight may underestimate measured weight. Researchers have tried to reduce the error using statistical models to predict weight from self-reported weight. We investigate whether deriving equations within separate BMI categories improves the prediction of weight compared with an equation derived regardless of an individual's BMI.

Design And Methods: The analysis included self-reported and measured data from 20,536 individuals participating in the EPIC-Norfolk study. In a derivation set (n = 15,381) two approaches were used to predict weight from self-reported weight: (1) using a linear regression model with measured weight as outcome and self-reported weight and age as predictors, and (2) using the same model fit separately within 3 strata defined by BMI (< 25, 25-30, ≥ 30 kg m(-2) ). The performance of these approaches was assessed in a validation set (n = 5,155). Measured weight was compared to self-reported weight and predicted weight.

Results: Self-reported weight underestimated measured weight (P < 0.0001): mean difference -1.2 ± 3.1 kg (men), -1.3 ± 2.5 kg (women). Underestimation was greater in obese participants (P < 0.0001). Predicted weight using approach 1 was not significantly different from measured weight (P < 0.05). However, in individuals with BMI < 25 kg m(-2) , weight was overestimated in men (0.90 ± 3.87 kg) and women (0.57 ± 2.06 kg), but underestimated in overweight (-0.29 ± 3.58, -0.20 ± 2.62 kg) and obese (-1.46 ± 5.05 kg, -0.73 ± 3.54 kg) men and women.

Conclusions: Using separate prediction equations in strata of BMI did not further improve prediction of weight. In conclusion, predicted weight was closer to measured weight compared with self-reported weight, but using equations derived in strata of BMI did not further improve the prediction and are not recommended for prediction of weight.