OBJECTIVES: The RCT represents the ‘gold standard’ for generating estimates of relative treatment effect. However observational data are considered more direct, relevant and generalisable to the wider population. Meta-analyses are usually conducted based on RCT evidence but interest in the inclusion of observational data in the decision-making process is growing. A case study that incorporates observational data in a network meta-analysis (NMA) is presented.

METHODS: A previously conducted systematic review and NMA based on RCT data of treatments for stroke prevention in atrial fibrillation was updated. A single comparative observational study (comparing 110/150 mg dabigatran versus vitamin K antagonist [VKA]) was identified and incorporated into the NMA analysis for the outcome of mortality by i) using this study to inform prior distributions and ii) using a design-adjusted analysis; adjusting the estimates from the observational data for potential bias.

RESULTS: The inclusion of the observational data in the NMA at face value resulted in lower odds of mortality of 150 mg/110 mg dabigatran versus VKA. In comparison, the associated 95% CrIs of the same treatment comparison estimates did not include the null value when the NMA was restricted to RCT evidence only. When 30% overestimation of observational evidence is assumed in the analysis, the observational evidence requires further discounting (adjustment for overprecision) before it could be concluded that the 110 mg or 150 mg dose of dabigatran does not decrease the odds of mortality compared with adjusted dose VKA.

CONCLUSIONS: In the exploratory analyses the conclusions around the relative efficacy of the NOACs are robust to the beliefs around the bias in terms of overprecision and overestimation within the observational data. Observational data can provide additional information on relative treatment effects and it may be relevant to include such data in meta-analyses to allow a decision-making process which includes all available data.