Claims for improved survival from systemic corticosteroids in diverse conditions: an umbrella review.

BACKGROUND: Systemic corticosteroids have been proposed for numerous indications and there are many claims that corticosteroids can reduce mortality in diverse conditions. METHODS: We performed an umbrella, agenda-wide review of the evidence on systemic corticosteroids and mortality, focusing primarily on large trials (defined as those with > 100 deaths) and meta-analyses. Searches were performed in PubMed and Cochrane Central Register of Controlled Trials (last update February 2011). We also examined whether spurious subset analyses may be responsible for claims of survival benefits in indications where only small trials had been available. RESULTS: Among 257 identified randomized trials with mortality data in their abstract, we found 14 large trials pertaining to 10 different indications. Although 10 of these 14 trials have reported statistically significant survival differences in subset analyses, none shows a nominally statistically significant (P < 0·05) decrease in death risk for any of the tested conditions when all deaths on all randomized patients are analysed. Meta-analyses for these conditions show statistically significant reductions in mortality only with antenatal corticosteroids for preterm labour (relative risk 0·77, 95% CI, 0·67-0·89) and in tuberculous meningitis (relative risk 0·78, 95% CI, 0·67-0·91). For conditions without any large trials, statistically significant reductions in mortality in meta-analyses were noted for Pneumocystis pneumonia (relative risk 0·54, 95% CI, 0·38-0·79) and alcoholic hepatitis (relative risk 0·63, 95% CI, 0·50-0·80). Many small trials that claim significant benefits, even those for classic indications such as typhoid fever and tetanus, have shown these benefits only in subset analyses. CONCLUSIONS: Corticosteroids have been documented to decrease mortality in some indications, in particular, antenatal use for preterm labour, tuberculous meningitis, Pneumocystis pneumonia, and alcoholic hepatitis. Many postulated benefits of corticosteroids on mortality may reflect 'vibration of treatment effects' leading to false-positive claims from spurious subset analyses and even for standard indications, such biases may have inflated the treatment effect estimates. More large trials are needed for serious, common conditions where use of corticosteroids is proposed.