What is it about?
In this commentary, we highlight important issues discussed within a recent systematic review on the association of breast milk fatty acids with allergic diseases. We also add some ideas regarding exposure measurement, timing of exposure, and participant selection all of which may impact on interpretation of previous findings. Most importantly, we point to shortcomings in the statistical methodology employed in previous work which we have overcome ourselves in a recent article (see https://goo.gl/t6AwtK and/or the resource linked to this article).
Featured Image
Why is it important?
A lot of effort has gone into the investigation of effects of breast milk fatty acids on subsequent allergic disease development. These range from animal experimental work via human observational studies to randomized controlled supplementation trials among humans. We clearly show that neglecting the compositional aspect of the data on breast milk fatty acids in the statistical analyses (mostly of the observational studies which prompted the randomized trials) can lead to spurious results. There are simple methods to overcome this error which should be used to re-examine the existing data. A further suggestion is to start to sample formula in addition to breast milk in observational studies to overcome selection bias by restriction to breastfed children.
Perspectives
Compositional data does not only occur in studies of breast milk fatty acids expressed as percent of total fat. It is a feature of many data, including sequencing data and studies of the microbiome. Transforming these data using e.g. the centered log ratio transformation (clr), additive log ratio transformation (alr), or isometric log ratio transformation (ilr), of which clr is most often used, is an easy way to make compositional data suitable for analyses with standard methodology such as multiple pairwise correlations, principal component analysis (PCA), multivariate analysis of variance (MANOVA), and regression modelling. There are further statistical tools like Random Forests that may be suitable to analyse compositional data even without transformation. Most of this statistical methodology is readily available and some of it can be easily applied even by researchers less acquainted with statistical issues of compositional data. I feel it is time to spread the word and improve the statistical analyses of compositional data to prevent more potentially spurious results from being published.
Prof. Dr. med. Jon Genuneit
Universitat Leipzig
Read the Original
This page is a summary of: Commentary: Association of Breast Milk Fatty Acids With Allergic Disease Outcomes—A Systematic Review, Frontiers in Pediatrics, April 2018, Frontiers,
DOI: 10.3389/fped.2018.00094.
You can read the full text:
Resources
Contributors
The following have contributed to this page
