OBJECTIVES: Network meta-analysis (NMA) is commonly used to compare between interventions simultaneously by synthesising the available evidence. Such evidence may vary in design and format. Evidence can be obtained from non-randomized studies (NRS) or randomized controlled trials (RCT). This evidence can be accessible in different formats as individual participant data (IPD) or aggregate data (AD). Our aim is to build a generic framework to utilize all available evidence while acknowledging the differences between their various sources.

METHODS: We conducted a literature review and identified all methods that can be used to combine IPD and AD from RCT or NRS in NMA. Publications were excluded if they were non-methodological or introducing methods only applicable in pairwise meta-analysis. We then compared and integrated the various methods in a generic model programmed in R. We applied the methods to synthesize evidence about pharmacological interventions for patients with relapsing remitting multiple sclerosis (RRMS). The dataset consists of 26 RCTs with AD, 3 RCTs with IPD and the Swiss Multiple Sclerosis Cohort.

RESULTS: Our literature search returned 1422 publications of which 11 papers met our inclusion criteria; six papers introduced methods to combine IPD and AD and 5 papers methods to combine RCTs and NRS in NMA. There are two main approaches to combine IPD and AD: a three-level hierarchical model or population adjustment methods. The latter can be implemented as matching-adjusted indirect comparisons, as simulated treatment comparisons or as a multilevel network meta-regression. There are four different approaches to combine RCT and NRS; the naïve approach, the design-adjusted analysis, using NRS as informative prior or fit a multilevel hierarchical model. We implement these approaches in R and analysed the RRMS data.

CONCLUSIONS: The resulting comprehensive model combining NRS with RCTs in the form of AD or IPD allows sensible integration of evidence and increased the power.