OBJECTIVES: In the last decade, network meta-analysis (mixed treatment comparison meta-analysis) has been introduced as a generalization of pair-wise meta-analysis. Many randomized controlled trials (RCTs) report treatment effect estimates for the outcomes of interest at multiple time points. In this paper we compare different methods for network meta-analysis of repeated measures. METHODS: Different network meta-analysis models for the synthesis of study level data of RCTs evaluating interventions for osteoarthritis were compared: separate network meta-analyses per time point; models assuming a linear development of treatment effects over time; network meta-analysis models with fractional polynomials; and network meta-analysis with splines. All analyses were performed in a Bayesian framework. RESULTS: The primary limitation of multiple network meta-analyses of study level data by time point was the inconsistency in the used evidence base for each time point. Of the models that estimate a relationship between outcome and time, fractional polynomials had an advantage over splines for the current dataset because the former model resulted in more stable parameter estimates and still provided sufficient fit to the data. CONCLUSIONS: To understand treatment effects over time, reported treatment effects of RCTs need to be synthesized simultaneously. Network meta-analysis of longitudinal study level data with second order fractional polynomials are a very useful approach to evaluate trends of treatment effects over time when there is not too much variation in treatment effects from one time point to the next. An additional advantage is that the methodology can handle differences in follow-up time across trials.