We would like to thank Dr Vestbo for his comments. We agree that in the Optimal Trial more patients originally randomised to the placebo arm prematurely discontinued study medications, and that many of these patients were subsequently put on open label ICS/LABA products.1 As discussed in our paper, the relative risk reduction decreased from 21% if patients were prematurely excluded once they discontinued study drugs to 15% when an intent to treat analysis was used.2 We agree with Dr Vestbo that our intention to treat analysis was conservative, and it did slightly reduce the possibility of a difference being found between placebo and active treatment but we would argue that this analysis was necessary in order to prevent bias.

An intention to treat analysis is necessary as it is impossible to know a priori the ultimate direction of the bias when patients who stop study medications early are banished from a clinical trial. Will the bias favour the drug or favour placebo? For example, a similar analysis of a trial assessing tiotriopium for chronic obstructive pulmonary disease (COPD)3 showed that the bias can work exactly in the opposite direction and instead favour placebo over active drug. In this study, higher incidence rates of fatal events occurred following premature discontinuation of study medication, especially in those patients randomised to the placebo arm. Presumably, patients who were taking placebo in this study were doing poorly and many prematurely stopped study drugs and then, shortly thereafter, they died. In this case, early exclusion of these patients would have introduced bias because the factors which determined whether a patient might have been excluded were also related to the outcome. If these patients had been dropped from the trial after premature discontinuation of study medications, this would have meant that their deaths would not have been discovered, and this would have produced a biased mortality incidence ratio in favour of placebo over tiotropium. The authors of this study concluded that failure to consider outcomes of patients who discontinue study medications early may bias results against effective therapies.4 Only by ensuring full follow-up of all randomised patients and by using a proper intention to treat analysis was this potential bias eliminated.

There is an old saying in medicine “You can’t find a fever if you don’t take a temperature”. This applies to clinical trials as well; the investigator cannot know what really happened in a clinical trial unless he/she evaluates outcomes in all randomised patients for the full study follow-up period, regardless of patient compliance.