Assessment of bias in outcomes reported in trials on pneumonia: a systematic review.

Subjective outcomes may exaggerate intervention effects compared to objectively measured outcomes. We compared effect estimates for clinical failure and all-cause mortality clinical trials of antibiotic treatment for pneumonia. A systematic review of randomized controlled trials assessing adults with pneumonia, comparing different antibiotics, published between 2005 and 2012 was undertaken. We compared the intervention to the control arm. The all-cause mortality in the intention-to-treat population and clinical failure as defined by the study investigators for the primary analyzed population were the primary outcomes examined. Risk ratios (RRs) with 95 % confidence intervals (CIs) were pooled, using a fixed effect model. Meta-regression was used to examine the impact of clinical failure on the mortality effect size. Thirty-six trials were included, of which 30 were industry-sponsored and 30 were non-inferiority trials. There was no difference between the effect on mortality for intervention versus control (RR 1.02, 95 % CI 0.91-1.16) and clinical failure (RR 1.01, 95 % CI 0.93-1.10), without significant heterogeneity in both analyses. In double-blind trials with adequate sequence generation and concealment, there was a significant advantage to the intervention for clinical failure (RR 0.86, 95 % CI 0.76-0.98), but not for mortality (RR 0.96, 95 % CI 0.76-1.21). RRs for clinical failure did not explain the variability in the RRs for mortality significantly, with a meta-regression coefficient of 0.32 (95 % CI -0.21-0.85). In non-inferiority trials of antibiotic treatment for pneumonia, we did not find evidence for bias induced by the use of a subjective outcome overall. The small number of trials without sponsorship precludes an adequate assessment of sponsorship effects.