Abstract
Two recent studies have re-analyzed published data and found that when datasets are analyzed independently there was limited support for the widely accepted hypothesis that changes in the microbiome are associated with obesity. This hypothesis was reconsidered by increasing the number of datasets and pooling the results across the individual datasets. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines were applied to identify 10 studies for an updated and more synthetic analysis. Alpha diversity metrics and the relative risk of obesity based on those metrics were used to identify a limited number of significant associations with obesity; however, when the results of the studies were pooled using a random effects model significant associations were observed between Shannon diversity, number of observed OTUs, and Shannon evenness and obesity status. They were not observed for the ratio of Bacteroidetes and Firmicutesor their individual relative abundances. Although these tests yielded small P-values, the difference between the Shannon diversity index of non-obese and obese individuals was 2.07%. A power analysis demonstrated that only one of the studies had sufficient power to detect a 5% difference in diversity. When Random Forest machine learning models were trained on one dataset and then tested using the other 9 datasets, the median accuracy varied between 33.01 and 64.77% (median=56.68%). Although there was support for a relationship between the microbial communities found in human feces and obesity status, this association was relatively weak and its detection is confounded by large interpersonal variation and insufficient sample sizes.
Importance As interest in the human microbiome grows there is an increasing number of studies that can be used to test numerous hypotheses across human populations. The hypothesis that variation in the gut microbiota can explain or be used to predict obesity status has received considerable attention and is frequently mentioned as an example for the role of the microbiome in human health. Here we assess this hypothesis using ten independent studies and find that although there is an association, it is smaller than can be detected by most microbiome studies. Furthermore, we directly tested the ability to predict obesity status based on the composition of an individual’s microbiome and find that the median classification accuracy is between 33.01 and 64.77%. This type of analysis can be used to design future studies and expanded to explore other hypotheses.
