Abstract
NNT calculations during years 2 and 3 are misleading as they don–t account for differential withdrawal in year 1 http://ow.ly/UeS61
To the Editor:
Suissa [1] provides the event-based number needed to treat (NNT) for a reduction in exacerbations with fluticasone/salmeterol compared with placebo for the TORCH (Towards a Revolution in COPD Health) trial by period of follow-up (0–1 years after treatment, 1–2 years after treatment and 2–3 years after treatment) [2]. These NNT data are misleading as they fail to recognise that in chronic obstructive pulmonary disease (COPD) trials, patients who exacerbate are more likely to withdraw from the trial than patients with no exacerbations. Keene et al. [3] show that for the TRISTAN (TRial of Inhaled STeroids ANd long-acting β2 agonists) trial, the exacerbation rate is more than three per year among placebo patients withdrawing prior to 1 year compared with an exacerbation rate of one per year for patients completing a year of placebo treatment. In the TORCH trial, 25% of the placebo group withdrew compared with 15% of the fluticasone/salmeterol arm during the first year of follow-up; therefore, patients entering the second year of the trial on placebo and on fluticasone/salmeterol were no longer directly comparable. This is not accounted for in the calculation of exacerbation rates or NNT during years 2 and 3, as presented in table 2 of the study by Suissa [1].
There are two other points that require further clarification. Firstly, table 2 incorrectly labels the treatment columns. These labels should be reversed; there was a higher rate of exacerbations per patient per year on placebo than on fluticasone/salmeterol. Secondly, in the introduction, the author quotes a previous paper [4] and states that the “NNT was illuminating in weighing up the benefit of inhaled corticosteroids in preventing COPD exacerbations against their risk of inducing pneumonia”. The methodology used to reach this conclusion based on NNT ignored the repeated nature of exacerbations [5].
In summary, as the author concludes, “it is important to ensure that the measures permit a comparison of like with like and are correctly calculated” [1]. The calculations for the TORCH trial presented by Suissa [1] fall short of these aims. For these reasons, we believe that the conclusions in Suissa [1] are not adequately supported by the data analyses presented.
Footnotes
Support statement: Funding information for this article has been deposited with FundRef.
Conflict of interest: Disclosures can be found alongside the online version of this article at erj.ersjournals.com
- Received January 10, 2015.
- Accepted August 14, 2015.
- Copyright ©ERS 2016