Characteristics of opioid prescriptions in 2009.

we chose the latter; indeed network meta-analysis has been shown to potentially give more reliable results because of the integration of additional information. Although studies included in our review were from different sources, they were all randomized controlled trials and hence contrasts between treatment groups within each study should be comparable. In addition, we focused only on CP/CPPS categories IIIA and IIIB to reduce heterogeneity due to disease severity and focused on outcomes measured using National Institutes of Health Chronic Prostatitis Symptom Index scales to reduce heterogeneity due to measurement error. Nevertheless, we explored potential discrepancies in treatment effects between direct and network meta-analysis results using the standardized normal method (z). Directions of treatment effect for the 2 methods were identical for all 12 comparisons; moreover, the magnitude of the effects between the 2 methods were similar except for -blocker vs placebo, where z was large and reached statistical significance (2.9380, P=.003). We believe that this is an example of increased precision of treatment effects due to the network method “borrowing” information from indirect comparisons. Third, Jackson et al disagree that study data should be expanded using a Stata command so that it could be included in the meta-analysis, questioning how we could know the distribution of data. We only used this command for the treatment responsiveness outcome, which is a dichotomous outcome and does not need any assumption about distribution, normal or otherwise. We believe that using all available data, rather than omitting studies, is an advantage and will lead to more valid estimates.