A novel framework for assessing causal effect of microbiome on health: long-term antibiotic usage as an instrument

Abstract

Gut microbiome (MB) has been widely shown to affect human health. Since MB in turn can be altered by various exposures, such as diet and medications, it holds immense potential for future treatments and healthy ageing. On the one hand, faecal microbiota transplantation and Mendelian Randomization have proven a causal link between treatment, MB and diseases. On the other hand, assessing the causality of the MB effects on health has remained challenging, since randomised trials in human subjects are often unethical or difficult to pursue, and Mendelian Randomization lacks valid instruments. Thus, novel analytical approaches are needed for inferring causal associations. To overcome these barriers, we propose a novel framework of antibiotic instrumental variable regression (AB-IVR) for estimating the causal relationships between MB and various diseases. Our inspiration originates from the popular Mendelian Randomization method that uses genetic mutations as instruments in the instrumental variable analysis (IVR). Further, we rely on the recently shown results that antibiotic (AB) treatment has a cumulative long-term effect on MB, consequently pseudo-randomizing individuals with higher AB usage to have more perturbed MB. Thus, we developed a new AB-IVR framework to utilise the long-term AB usage as an instrument in the IVR for assessing the causal effect of MB on health. We pursued a plethora of sensitivity analyses to explore the properties of our method: varying the sample's age group and maximum number of AB used; using a buffer-time for incident disease outcomes to account for feedback-mechanism; using subgroups of AB as instrument; and simulating data for disease outcomes. We detected several interesting causal effects of MB on health outcomes; some causal effects - such as MB effects on migraine, depression, irritable bowel syndrome, and several more - remain significant irrespective of the sensitivity analysis used. We believe that our AB-IVR framework has promising potential to be the new widely used method for assessing MB effect on health.