Conventional compared to network meta-analysis to evaluate antibiotic prophylaxis in patients with cancer and haematopoietic stem cell transplantation recipients

Our purpose was to compare conventional meta-analysis and network meta-analysis to evaluate the efficacy of different prophylactic systemic antibiotic classes in patients undergoing chemotherapy or haematopoietic stem cell transplant (HSCT). We included randomised trials if patients had cancer or were HSCT recipients and the intervention was systemic antibacterial prophylaxis. Three types of control groups were used: (1) placebo, no antibiotic and non-absorbable antibiotic separately; (2) placebo and no antibiotic combined; and (3) all three combined. These gave different network geometries. Strategies synthesised were fluoroquinolone, trimethoprim-sulfamethoxazole, cephalosporin and parenteral glycopeptide versus control groups. In total 113 trials met the eligibility criteria. Where treatment effects could be estimated with both conventional and network meta-analysis, values were generally similar. However, where events were sparse, network meta-analysis could be more precise. For example, trimethoprim-sulfamethoxazole versus placebo for infection-related mortality showed a relative risk ratio (RR) of 0.55, 95% CI (0.21 to 1.44) with conventional, and RR 0.43, 95% credible region (0.20 to 0.82) with network meta-analysis. Cephalosporin versus fluoroquinolone was comparable only indirectly using the network approach and yielded RR 0.59, 95% credible region (0.28 to 1.20) to reduce bacteraemia. Incoherence (difference between direct and indirect estimates raising concerns about network meta-analysis validity) was observed with network geometry where control groups were separated, but not where control groups were combined. In this situation, conventional and network meta-analysis yielded similar results in general. Network meta-analysis results could be more precise when events were rare. Some analysis could only be performed with the network approach. These results identify scenarios in which network meta-analysis may be advantageous.