Simpson’s paradox in proof-of-concept studies

To the Editor — We read with interest the article by Depommier et al. reporting improved insulin sensitivity, reduced cholesterol levels and decreased white blood cell counts in humans with metabolic syndrome after supplementation with Akkermansia muciniphila¹. Although the authors carried out elaborate laboratory experiments to test their concepts, we feel that this study lacks in the fundamental information for a phase 1 human trial report. The predictor—A. muciniphila treatment—and the outcomes—obesity, metabolic syndrome and diabetes—are affected by lifestyle factors, glucose levels and diet, forming a complex confounding relationship. Thus, these factors must be considered even in a phase 1 trial, in which the main purpose is safety assessment. Without this consideration (or adjustment), the metabolic improvements reported by Depommier and colleagues may be an example of Simpson’s paradox (which posits that when confounding factor(s) are not adjusted for, they can generate opposing and seemingly paradoxical results). If the observed results are a product of Simpson’s paradox, stratifying the results according to the levels of diet and exercise will cause the observed beneficial effects to disappear. Randomization alone does not balance all the confounding factors, unless the sample size is sufficiently large².

A diet rich in polyphenols, polydextrose, butyrate and inulin is reported to increase A. muciniphila abundance³. Butyrate and other short-chain fatty acids from high-fiber diets play an important role in gut mucosa health⁴, improve insulin resistance⁴ and gut permeability⁵ and affect the mucin layer, which is the substrate for A. muciniphila⁶. Thus, it is highly likely that diet will affect the abundance of A. muciniphila and reduce metabolic inflammation as well⁵. Therefore, phase 1 trials such as this one must follow the standard human phase 1 trial reporting format to show that confounding factors would not have biased their results, as was recently reported by Bajaj et al.⁷. Without elucidating the balanced confounding factors at baseline, the report may contain potentially biased results and trigger the lay media to produce flash headlines that may mislead the public.

Notably, the health benefits of the A. muciniphila membrane protein Amuc_1100 were more pronounced than those of live A. muciniphila¹. Hence, supplementation with Amuc_1100 rather than live bacteria will be more appropriate in future research. Additionally, we suggest that phase 2 and 3 trials should have a cross-over design to mitigate any bias stemming from the imbalance in confounding factors in the parallel group design. We sincerely hope that Amuc_1100 can curtail the obesity pandemic worldwide. The investigators should realize that confounding must be balanced at baseline; otherwise the results will be biased. Also, blinding will not balance the confounding.