Author's response to reviews Title: Prevalence of thinness in children and adolescents in the Seychelles: comparison of two international growth references Authors:

1) Results: Please rephrase lines 2-3 to correct the information “The prevalence of the thinness categories tended to change according to age.” According to the results presented by authors, using the IS reference the prevalence of the thinness categories tended to change according to age for both sexes. Using the WHO reference this trend is confirmed only for girls. We have reformulated the sentence as suggested by the Reviewer 2) Conclusion: I suggest delete the last phrase “Universal cut-offs...thinness categories”, as this statement cannot be concluded with the analysis and results of the present article. This phrase sounds like a suggestion not as a conclusion. We agree with the Reviewer. This sentence was meant to put the findings in a broader perspective. This information appears in the discussion and has been deleted from the abstract. METHODS AND POPULATION 3) The most important concern refers to the statistical methods used: to compare the prevalence of thinness using the WHO and IS references I suggest a concordance test, for example the kappa statistic. To evaluate the differences in prevalence between the categories of thinness, with the use of the different references, by chi square test or ICs, does not mean that the same children were classified in one of the categories using the two international references. So it is not possible to conclude that there were no differences between certain categories of thinness using the statistical tests described in this article. We thank the Reviewer to raise this important issue but we respectfully disagree in some details with the Reviewer. The concern of the Reviewer would be valid if thinness was defined, for example, as weight for age (and sex) in one reference and BMI for age (and sex) in another reference. Because the IS and WHO definitions are based on the very same dimension, i.e. BMI by sex and age, a same prevalence of a thinness category along both the IS reference and the WHO reference necessarily implies that the same children are classified as thin for both references along the considered thinness category. Inversely, when the prevalence along IS or WHO differs, say prevalence is 5% with IS and 8% with WHO, this necessarily means that 5% of the same children are thin along both IS and WHO and an additional 3% are thin along the WHO reference. This is well apparent in the Figures. In this respect, a kappa statistic does not add much information, unlike for dimensions that are made of several indicators. For example, if we had compared different definition of the metabolic syndrome, say ATP III and IDF, different children would be classified as having MS with each definition and a concordance statistic (such as kappa) would be useful. 4) Do the authors take into account the fact that the same child may have had an examination at different time points across 1998 and 2004? Indeed a proportion of all children have been seen two times since children can be seen every 3-4 years along the surveillance system. We have added that “Because surveys take place every year in four grades and a same child can be seen at intervals of 3-4 years, observations among 33,340 children during 9 years correspond to observations in 19’764 different children”. 5) Last paragraph: In the results section (Figures 3 and 4) the authors commented the differences in prevalence of thinness using IS and WHO references. This description is lacking in the Methods section. Please, describe the statistical analysis used to access the differences in prevalence of thinness using IS and WHO references, across genders and ages. No statistical test was used for estimating differences between prevalence along IS and WHO references. In our study, we have data on the entire population (of children) of the country. Hence, it is not a random sample of this population and estimates of variance cannot be determined in this situation. Furthermore, multiple testing (if we had random samples of the population) would be a problem because of the multitude of tests needed to assess for differences in all thinness categories in all age categories. Furthermore, testing for differences for categories with low or very low prevalence, which correspond to the lowest values of BMI, can be particularly sensitive to measurement errors in a few readings (misreporting, error in readings of height or weight) and differences in such instances should be taken with caution. Hence, we believe that the best analysis in the case of our study is to assess the overall patterns in these prevalences along sex and age. If our study and other studies in other populations consistently suggest specific differences, further studies should investigate these differences along specific hypothesis and an ad hoc pre-specified analysis plan. 6) If the authors choose to present the Figures 1 and 2 (whose figures I suggest to delete from the article), please describe why and how these figures are important to the article in the methods section. We believe that these figures are useful. Our paper has no a priori hypothesis about what difference is expected between the two references as few previous data are available. The two figures show to the readers how the IS and WHO references classify thinness categories along sex and age in the population of Seychelles. The apparent patterns (e.g. difference along age) can inform specific hypothesis for future studies. 7) Please insert indications of the software used for data entering and statistical analyses, and the adopted level of statistical significance. We have added in the MS that we used Stata 10 and reported 95% confidence intervals. See also our response to the point 5 above.