Effect of methodological choices and inclusion criteria on network meta-analysis results in psoriasis.

Abstract

Background: When conducting network meta-analysis (NMA), researchers need to make several methodological and analytical decisions, which can influence the results of NMAs. Our objective was to evaluate the impact of different methodological choices on the conclusions from the analyses of a network of 20 active treatments in patients with psoriasis.

Methods: We re-analysed the available data of a living Cochrane NMA evaluating the systemic treatments in psoriasis under various analytical scenarios defined by the combination of pre-specified methodological choices. We performed NMAs on three outcomes: Psoriasis Area Severity Index (PASI) 90, PASI 100 and serious adverse events (SAEs). Variability of the effect estimates across NMAs was summarized using ratio of relative risks (RRR) and ratio of odds ratio (ROR). We estimated the level of agreement between the treatment hierarchies using the Average Overlap (AO).

Results: Overall, 560 NMAs were conducted. The median number of included interventions was 18 (IQR 17-19), for a median number of included studies of 68 (IQR 57-93). The median RRR was 1.06 (IQR 1.06-1.08) for PASI 90, 1.07 (IQR 1.06-1.10) for early PASI 90, 1.14 (IQR 1.06-1.15) for late PASI 90, 1.04 (IQR 1.01-1.05) for PASI 100, and 1.02 (IQR 1.02-1.06) for SAEs. The criteria with the greatest impact on the effect estimates were the inclusion or exclusion of studies with biological-naïve patients, inclusion or exclusion of phase II trials, and the inclusion or exclusion of studies evaluating conventional treatments. The analysis choice with the greatest impact was the use of the Mantel-Haenszel method instead of the inverse variance method. There was a high agreement of treatment hierarchies between analyses. For the top 6 ranking treatments, the median AO across all scenarios for PASI 90 was 0.84 (IQR 0.72-0.97). For early PASI 90, late PASI 90, PASI 100, and SAE, the median AO were 0.94 (IQR 0.91-0.97), 0.75 (IQR 0.65-0.97), 0.94 (IQR 0.91-0.97), and 0.59 (IQR 0.59- 0.90), respectively.

Conclusions: We found that different methodological choices could influence NMAs' results. However, even though moderate variation in effect estimates could be observed across the analyses, treatment hierarchies remained stable for the top-ranking drugs.