Statistics from Altmetric.com
I read with interest the article by Aaron et al.1 In this paper, data from the Optimal Trial2 were reanalysed for the purpose of examining the effect of differences in counting and analysing exacerbation rates on estimated treatment effects in chronic obstructive pulmonary disease (COPD). The authors compare exacerbation rates in two of the three treatment arms in the trial (ie, those randomly allocated to tiotropium + placebo or to tiotropium + fluticasone–salmeterol). They compare an “intention to treat” strategy with a strategy using “time in study only” and state that the often used method of excluding patients after they stop study medication exaggerates the estimated benefits of treatment.
Data from large controlled trial are often reanalysed and they usually provide a good and solid database for assessment of methodology. However, when trials are reused, the original study is often only described briefly in secondary publications and often crucial information is missing. This seems to be the case for the reanalysis of the Optimal Study. In the original report,2 the numbers of patients withdrawing from the tiotropium + placebo group and the tiotropium + fluticasone–salmeterol group within the 52 week treatment period were 74 (47%) and 37 (26%), respectively, indicating less treatment efficacy in the former group. In the original paper—but not in the reanalysis—it is also shown that of the patients who discontinued use of study medications, 74% in the tiotropium + placebo group and 54% in the tiotropium + fluticasone–salmeterol group received an open label inhaled steroid and long acting beta2 agonist combination inhaler for the remainder of the study. That approximately half of the patients randomised to tiotropium + fluticasone–salmeterol were given the same type of treatment does not substantially affect the analyses. However, when those stopping tiotropium + placebo are given an open label inhaled steroid and long acting beta2 agonist combination inhaler for the remainder of the study it is not surprising that the “intention to treat strategy” dilutes the effect of the triple combination treatment. Although we all want conservative treatment estimates from controlled trials, there is a real risk of getting not just conservative but in fact insignificant findings if the actual study comparison differs significantly from the setting described in the study protocol.
Aaron et al advocate that any intention to treat analysis is superior to other strategies. However, when withdrawal rates are substantial, as in the Optimal Trial, and patients withdrawing from study medication are given medication being tested in the trial, any conclusion on analysis methodology should be made with caution.
Competing interests: JV has been involved in clinical trials of inhaled corticosteroid alone or in combination with long acting beta agonists; his wife is an employee of AstraZeneca.