For instance if the body fat cut off is based on some western standards, how can we determine cut off for our study, based on the study results? ROC curve?
If your data are distributed in a Gaussian way (normal distribution), then the cut off will be the mean+2SD. However, and to be more precise, you can create a 2*2 table, from which you can calculate the performance indicators of your test- mainly the sensitivity and specificity- from which you can construct a Receiver Operator Characteristic (ROC) curve where your tests' cut off shall be the point which possess, theoretically, the maximum sensitivity and specificity values.