Article type
Year
Abstract
Background: Network meta-analysis is a novel method for comparing multiple interventions. Over the past decades, the number of studies has increased rapidly, and the evolution of methodology is still ongoing. Recently, the estimation method of prediction intervals in network meta-analysis has been proposed. The prediction interval, which estimate the probable range of future trial, makes the interpretation of results easier and also guidance for the future trial. However, a standard evaluation approach of the prediction ability is still unclear.
Objectives: To validate empirically the prediction ability of network meta-analysis and to evaluate their performance against randomized controlled trials (RCTs) that become available after network meta-analyses are conducted.
Methods: We conducted a literature search within PubMed, Embase, and Cochrane for the studies of network meta-analyses in kidney diseases. We reanalyzed the prediction interval of published network meta-analysis without the latest study among network meta-analyses and then compared that to the confidence interval of the latest RCT. We used the latest RCT as the standard and then calculated the coverage probability of the prediction interval of NMA.
Results: Our search identified a total of eight network meta-analysis studies including 173 trials. None of these studies reported the prediction interval of the effect size. Compared to the latest RCT in the network meta-analysis, the average coverage probability of the prediction interval was 65.89% (standard deviation = 0.40). Two studies had low coverage probability (< 25%), one study had median coverage probability, and the other five studies had high coverage probability (> 75%).
Conclusions: Reporting network meta-analysis with the prediction interval could apply to the guidance of clinical trial. Also, a performance measure of prediction should be conducted in the results.
Objectives: To validate empirically the prediction ability of network meta-analysis and to evaluate their performance against randomized controlled trials (RCTs) that become available after network meta-analyses are conducted.
Methods: We conducted a literature search within PubMed, Embase, and Cochrane for the studies of network meta-analyses in kidney diseases. We reanalyzed the prediction interval of published network meta-analysis without the latest study among network meta-analyses and then compared that to the confidence interval of the latest RCT. We used the latest RCT as the standard and then calculated the coverage probability of the prediction interval of NMA.
Results: Our search identified a total of eight network meta-analysis studies including 173 trials. None of these studies reported the prediction interval of the effect size. Compared to the latest RCT in the network meta-analysis, the average coverage probability of the prediction interval was 65.89% (standard deviation = 0.40). Two studies had low coverage probability (< 25%), one study had median coverage probability, and the other five studies had high coverage probability (> 75%).
Conclusions: Reporting network meta-analysis with the prediction interval could apply to the guidance of clinical trial. Also, a performance measure of prediction should be conducted in the results.