Adjusting the obesity thresholds for self-reported BMI in Ireland: a cross-sectional analysis

Type Article

Journal Article


D. Madden

Year of publication



BMJ Open







OBJECTIVE: To investigate the optimal adjustment to be made to obesity thresholds when using self-reported body mass index (BMI). DESIGN: A cross-sectional study. SETTING: Data from the Survey of Lifestyle, Attitudes and Nutrition in Ireland, a nationally representative dataset using the Geodirectory (a listing of all residential addresses in Ireland compiled by the postal service) as the sampling frame. PARTICIPANTS: A nationally representative sample of 10 364 adults aged 18+, carried out by face-to-face interview with clinical measurement applied to a number of outcomes to a representative subsample of 2174. After discarding the observations with missing values and errors, the eventual sample was 1874. PRIMARY AND SECONDARY OUTCOME MEASURES: BMI based on measured and self-reported weight and height. BACKGROUND: It is generally found that self-reported BMI understates true or measured BMI and accordingly revised obesity thresholds have been suggested. METHODS: Data from the 2007 Survey of Lifestyles, Attitudes and Nutrition in Ireland were used to analyse self-reported and measured BMI. The self-reported BMI threshold was adjusted to obtain the optimal signal for measured BMI using different criteria viz. efficiency (maximum number of correct classifications), maximisation of Youden's J, maximisation of OR, minimisation of cost of misclassification and constrained optimisation. RESULTS: The optimal threshold differed substantially depending on the criterion adopted for choosing it, with thresholds of 29.1 (efficiency criterion), 27.5 (Youden's J) and 26.0 (FN rate of 5%). Standard criteria such as Youden's J index were shown to implicitly impose relative costs of false-negatives and false-positives which may not always correspond to the values of the analyst. CONCLUSIONS: When adjusting self-reported BMI thresholds in order to obtain the optimal signal for 'true' obesity, analysts should explicitly choose the relative costs of false-positives and false-negatives.