Bias due to withdrawal in long-term randomised trials in COPD: evidence from the TORCH study.

Randomised controlled trials (RCTs) are considered the least biased method for evaluating drug efficacy and several large long-term RCTs in chronic obstructive pulmonary disease have been published. These usually include drugs with symptomatic benefits and have significant withdrawal rates. We aimed at examining bias due to differential withdrawal in the Towards a Revolution in COPD Health (TORCH) trial. We did an observational study nested in the TORCH trial, a placebo-controlled trial of salmeterol/fluticasone propionate combination (SFC) therapy in chronic obstructive pulmonary disease. We included 3057 patients randomly allocated to placebo or SFC in the analyses. We examined rates of withdrawal from the study and analysed change in effect parameters over time and in relation to withdrawal, as well as medication uptake after withdrawal. There was differential withdrawal with a significantly higher withdrawal rate from the group allocated to placebo than to SFC, 44% compared with 34%. Regardless of treatment group, withdrawal was associated with worse baseline lung function and more frequent exacerbations, leading to selection of a study population in better health than those originally recruited. As a result, annualized exacerbation rates in the first 6 months of the study compared with the last 6 months of the study decreased from 6.8 to 0.9 in the placebo group and from 3.0 to 0.8 in the SFC group. Also, use of medications under test in the study was frequent in patients after withdrawal. Significant bias may occur in long-term RCTs of registered medications with symptomatic benefits as a result of differential withdrawal. © 2010 Blackwell Publishing Ltd.