Predicting the placebo response in a clinical trial based on pre-trial data

Published trial data have rarely allowed to identify factors contributing to the placebo response (PR) size, and raw data from clinical trial records have usually been unaccessible for this purpose.

Methods: We re-analysed the trial data from a recent functional dyspepsia (FD) trial with Iberogast (R) (Am J Gastroenterol 102;2007:1268–75) for factors determining predictors of PR and placebo non-response (PnR) in the placebo arm (n=157) of the study. Variables tested were the total gastrointestinal symptom score (GIS) at run-in, GIS change during run-in, type of FD symptoms, duration of illness, age, gender, body mass index, family occurence of FD-like symptoms, smoking, and alcohol consumption. In addition, distribution of PR and PnR across the 39 centers of the study were analysed.

Results: With the initial response criterion (40% improvement in GIS), 35 patients classified as placebo responders, while 122 were placebo non-responders. Responders had significantly lower GIS scores compared to non-responders at visit 1 but not at visit 2, when study medication was dispensed; hence, responders increased symptoms during run-in, while non-responders decreased their symptoms. Gender and age were not different between PR and PnR, neither were duration nor type of FD symptoms, nor the occurrence of FD symptoms in parents, siblings, and children. Consuming alcohol did not differ between PR and PnR, but smoking was significantly less frequent in responders (3%) as compared to non-responders (21%). With a more “liberal“ response criterion (<30% GIS improvement) and more placebo responders, only the change of symptoms during run-in remained predictive, with a more restrictive criterion (>60%), the BMI (higher in PR) was the only predictor of the PR. A model using “smoking“ and “symptom change during run-in“ was not predictive for being a responder or non-responder to the drug. Distribution of PR and PnR across trial centers was unequal but did not reach significance.

Conclusion: Variables predicting the PR (but not the drug response) point towards higher levels of health concern and lower levels of trust in PR as compared to PnR. However, the higher BMI may reflect symptom severity, and non-smoking may reflect an underlying biological mechanism of the PR.