Designing fecal microbiota transplant trials that account for differences in donor stool efficacy

by Scott Olesen

Microbes appear to be involved in a complicated interaction with their human hosts, protecting us from some diseases and possibly helping to cause others. Gut microbiota are suspected to be involved in inflammatory bowel disease, a condition where the human body’s immune system attacks its own digestive system. Inflammatory bowel disease, or IBD, has no known cures. Doctors treat IBD using drugs that reduce inflammation or suppress the immune system. When this is no longer enough, surgery is the only option: patients have entire parts of their digestive system removed.

This grim outcome for IBD means that doctors and patients are willing to take some risks in trying to cure the disease. It is a picture similar to Clostridium difficile, or C. diff, infections. Many C. diff infections can be cured with antibiotics. If antibiotics fail to cure the infection, or if the infection continues to recur despite antibiotics, doctors turn to fecal microbiota transplants. A fecal microbiota transplant, or FMT, involves taking the feces of a healthy person and infusing it into the digestive system of a sick person. When performed in an appropriate medical setting, FMT cures C. diff with ~90% efficacy.

The effectiveness of FMT for C. diff has generated interest in using FMT to treat IBD. Results from a recent FMT-for-IBD trial suggest an unexpected possibility: although any well-screened donor appears effective for treating C. diff, it appeared that only one of six donors used in this trial produced feces that effectively treated IBD.

If this “donor effect” is real, what are the consequences for clinical trial design? If we expect that only one in six donors produces efficacious feces, then it would be unwise to, say, treat patients using feces from a donor selected at random from the donor bank. Unfortunately, there is currently no way to predict ahead of time which donors will produce efficacious stool. Without a biomarker that predicts which donor will cure which patient, what should we do?

We reasoned that a biomarker is not essential for identifying efficacious donors: you can just use the clinical data about which patients responded to treatment from which donors to predict the quality of the donor. Although we focused on IBD, this approach has the advantage that it is not specific to any particular disease.

To get a rational, quantitative answer about which donor to use, we developed a mathematical model of differences in donor efficacy and an optimal Bayesian algorithm for identifying and using “good” donors. We compared a tractable, approximate version of this optimal algorithm against more traditional adaptive and non-adaptive approaches.

We found that, if differences in donor heterogeneity exist, it will be very important to the success of clinical trials to use approaches like the ones we developed when designing and executing the trial. More naive approaches are likely to produce false negatives, that is, to incorrectly conclude that FMT is ineffective, which would deny a useful therapy to suffering patients.

In our manuscript, we translated our scientific results into recommendations for clinicians that we hope will improve the probability of discovering if FMT is effective for treating diseases beyond C. diff.

bioRxiv. 25 July 2016. doi:10.1101/065383