Statistical analysis

Differences between the countries in bacterial relative abundances were tested using generalised linear models with negative binomial distribution. Associations between bacterial taxa to BMI and early growth were tested using correlations. To adjust for the effects of country, birth weight, breastfeeding duration, and antibiotic use on BMI, we calculated the deviance from the expected BMI, based on a linear model with the aforementioned variables. We then tested for a correlation between the BMI deviance and the log-transformed relative abundances of the bacterial taxa. The models were run using the whole data set and separately for each cohort to ascertain consistency in the associations. In addition, the models were run separately for children with minimal antibiotic exposure (0–1 lifetime courses), and those with several antibiotic courses, to see if the associations were present regardless of antibiotic exposure. Only associations that were significant in the total cohort (p < 0.05) and nearly significant (p < 0.15) in both countries separately were considered robust. All statistical analyses were conducted in R [37] using the package vegan [38]. R script for the analysis is available in Additional file 2.