TWO AUTHORS REPLY

We thank Dr. Chiolero (1) for his interest in our article (2) on the meta-analysis that compared the strength of associa-tions of the waist-to-height ratio and 3 other anthropometric indicators (body mass index, waist circumference, and waist-to-hip ratio) with future type 2 diabetes risk. As Dr. Chiolero pointed out (1), we agree that interobserver mea-surement differences are larger for waist and hip circumfer-ences compared with those for height and body weight, with the result that reliability is lower for waist circumference, waist-to-height ratio, and waist-to-hip ratio than for body mass index (3). We also acknowledge that we should have addressed the issue of measurement errors in the Discussion in our article. Nevertheless, it is unlikely that we must change the conclusion that waist circumference and waist-to-height ratio are stronger indicators for predicting future diabetes risk than body mass index, although the predictive value of the waist-to-hip ratio might have been substantially underestimated. The observed standard deviation (SD) combines the real SD and the SD derived from measurement error (4). Accord-ingly, the diabetes risk per 1 SD also would be over-estimated in proportion to the difference between the observed and real SD. However, the overestimation is counterbal-anced, possibly to excess, by underestimation due to the dilution effect of measurement bias on the regression coeffi-cient (i.e., diabetes risk per 1 SD in this case). A simple simulation for explaining the effect of the mea-surement error on the coefficient (β) is presented as follows. The relationship between the observed and real coefficient is indicated by the formula: bobserved brea