Finally, it should be noted that applied statistical methods did not protect against Type I errors associated with multiple testing. However, although significant differences in particular metabolites could be obtained by chance, the clinical improvement seems to confirm their relevance.
