Avoiding selection bias in metabolomics studies: a tutorial.

We demonstrate that standard analytical methods, such as stratification or adjustment in regression analyses, are not suited to deal with selection bias and may even induce the bias when analysing metabolite-phenotype relationships in selected groups. Finally, we show that inverse probability weighting, also known as survey weighting, can be used in some situations to make unbiased estimates of the outcomes.