Is the transitivity assumption valid?




Poster session 2


Monday 17 September 2018 - 12:30 to 14:00

All authors in correct order:

Breilmann J1, Koesters M1
1 Department of Psychiatry II, Ulm University, BKH Guenzburg, Germany
Presenting author and contact person

Presenting author:

Johanna Breilmann

Contact person:

Abstract text
Treatment decisions often build on indirect comparisons of placebo-controlled trials. Network meta-analyses (NMA) have been developed as formal approach to include indirect comparisons. The approach assumes transitivity of treatment conditions across trials. However, there is some evidence suggesting differences in placebo treatments, which therefore may violate this assumption.

As an extension to our previous findings (Breilman 2018), our analysis aimed at comparing the placebo effect sizes (PES) from randomised-controlled trials (RCTs) of antidepressants.

We derived all placebo-controlled, monodrug RCTs from a recently published NMA (Cipriani 2018), which examined the efficacy of antidepressants in the acute treatment of major depressive disorder (both sexes; aged ≥ 18 years). We used the standardised pre-post PES of depressive symptom scales, weighted by the inverse variance and aggregated in random-effects models, to compare the PES of different antidepressant trials. Meta-regression was used to include baseline depression severity and publication status as moderators.

We examined the PES of 18 antidepressants. Heterogeneity of PES was high (I² = 75%). The subgroup analysis revealed statistically significant differences of the PES between the different antidepressant trials (highest PES -1.26 for vortioxetine, lowest PES -0.88 for fluoxetine; test for subgroup differences P < 0.001; k = 178). The meta-regression including additional predictors indicated that statistical differences between PES from different trials remain after controlling for severity and publication status.

Our analysis revealed significant differences in the placebo effect sizes between RCTs of different antidepressants and suggests that the placebo effect size is related to the investigated drug, independently from baseline severity or publication status. As a consequence, inferences of drug efficacy based upon indirect comparisons of results from placebo trials or NMAs may present threats to validity by undermining the transitivity assumption and should be conducted with the necessary care.

Patient or healthcare consumer involvement:
No involvement.

Breilmann J, et al. (2018). Differences in the placebo response in duloxetine and venlafaxine trials. Submitted for publication.
Cipriani A, et al. (2018). Lancet. doi: 10.1016/S0140-6736(17)32802-7.

Relevance to patients and consumers: 

Network meta-analyses are increasingly influencing clinical decision making. The validity of the evidence from network meta-analyses is therefore of great importance for patients. The findings of the present analysis indicate that one of the basic assumptions for the validity of a recent network meta-analysis may be violated and thus suggestion caution in interpreting these results.