Questionable assumptions hampered interpretation of a network meta-analysis of primary care depression treatments.

We aimed to evaluate the underlying assumptions of a network meta-analysis investigating which depression treatment works best in primary care and to highlight challenges and pitfalls of interpretation under consideration of these assumptions.We reviewed 100 randomized trials investigating pharmacologic and psychological treatments for primary care patients with depression. Network meta-analysis was carried out within a frequentist framework using response to treatment as outcome measure. Transitivity was assessed by epidemiologic judgment based on theoretical and empirical investigation of the distribution of trial characteristics across comparisons. Homogeneity and consistency were investigated by decomposing the Q statistic.There were important clinical and statistically significant differences between "pure" drug trials comparing pharmacologic substances with each other or placebo (63 trials) and trials including a psychological treatment arm (37 trials). Overall network meta-analysis produced results well comparable with separate meta-analyses of drug trials and psychological trials. Although the homogeneity and consistency assumptions were mostly met, we considered the transitivity assumption unjustifiable.An exchange of experience between reviewers and, if possible, some guidance on how reviewers addressing important clinical questions can proceed in situations where important assumptions for valid network meta-analysis are not met would be desirable.     «

We aimed to evaluate the underlying assumptions of a network meta-analysis investigating which depression treatment works best in primary care and to highlight challenges and pitfalls of interpretation under consideration of these assumptions.We reviewed 100 randomized trials investigating pharmacologic and psychological treatments for primary care patients with depression. Network meta-analysis was carried out within a frequentist framework using response to treatment as outcome measure. Transi...     »