Treatment effects in pharmacological clinical randomized controlled trials are mainly due to placebo.

OBJECTIVES

The placebo response in clinical trials has four components: regression to the mean (RTM), measurement artifacts, natural tendency (NT) of the disease, and the genuine placebo effect. Our objective is to determine what contributes to the size of the placebo-effect in clinical trials by meta-regressions of randomized placebo-controlled clinical trials.

STUDY DESIGN AND

SETTING

We identified five diseases where data on the rates of NT were available to search for a sample of n = 150 (5x30) randomized controlled trials. We extracted various study descriptors and performed meta-regressions to predict improvement in treatment and placebo groups.

RESULTS

We sampled 30 trials each from the following diagnoses: osteoarthritis of the knee, irritable bowel syndrome, depression, sleep disorders, migraine, and extracted relevant information. We estimated the effects due to RTM and NT and analyzed the improvement in placebo and treatment groups by fitting two regression models. Both models were highly significant, explaining 72% of the variance. Improvement in the placebo group can be significantly predicted by improvement in the treatment group (beta = .84), whether a study was analyzed according to intention to treat (beta = -.10) or was a multicenter study (beta = .12). Improvement in the treatment group can be explained by the improvement in the placebo group (beta = .83), whether a study was a multicenter trial (beta = -.16), and by RTM (beta = -.18). The treatment effect is smaller in sleep studies (beta = -.17).

CONCLUSION

The high correlation of r = .73 between placebo improvement and treatment improvement rates is genuine and not explainable by study or disease characteristics. We conclude from our data that the placebo-effect is the major driver of treatment effects in clinical trials that alone explains 69% of the variance. This leaves only limited space for effects due to pharmacological substances. Context effects are more important than pharmacological ones in the conditions studied by us.