Methodological and Statistical Comments on CASPIAN-III Study

which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Kelishadi and her colleagues have done a great job for determining students' high risk behaviors (1) and have published part of this study as percentiles of aminotrans-aminases in Iranian children (2) These data are only from Iran, a country in Middle East and not children of Middle East and North Africa (MENA). Only children between 10 and 18 years are studied and not all children like younger than 10 years. Moreover, only 27 provinces from 31 provinces have been selected and not even all provinces of Iran. Data are really valuable but not representative of MENA. These are not even completely generalizable to Iranian children because neither covers all provinces nor all ages. Despite I did not see any calculation based sample size estimation neither in this paper (2) nor in their previous paper about the methodology of this study (1); but, sample size seems relatively sufficient. However, sampling method is more important than sample size for generalizability. Authors claimed that difference in their findings with previous study in Tehran, which has revealed lower limit for upper normal limits for ALT and AST (3), can be due to sampling, which is only from one city and not whole country and this difference may be also due to lower sample size. They have mentioned that the population studied in Tehran was less than one-fourth of the current nationwide study. It should be considered this "one-forth", accounted for 975 samples from one province. It is higher than your sample from Tehran and more homogenous than your nationwide sample. You should not compare your total sample size with the number of studied people in one province. Their results seems more representative to Tehran due to your mentioned limitations specifically when Poustchi et al. (3) have excluded subjects having abnormal values for factors that correlated with ALT in their study. In different parts of the paper (2), authors have expressed that there are linear associations. Why the authors believe that this association is linear? Is there any pre-specified hypoth-esis? Most of the time, the associations are not linear specifically when there is no significant linear relationship or the linear association is weak. In this study R-squares are low and the shape of the association between BMI and age with AST and ALT (Figures 1 and 2) shows …