Reply to López-Campos et al. and to Rogliani and Calzetta

From the Authors:

We thank Dr. López-Campos and colleagues and Drs. Rogliani and Calzetta for their interest in our work published recently in the Journal (1). They raise several methodological queries regarding ETHOS (Efficacy and Safety of Triple Therapy in Obstructive Lung Disease) that we wish to clarify.

First, the mortality analyses in ETHOS followed the intent-to-treat principle, using a treatment policy estimand that analyzed data from all randomized and treated patients, regardless of whether they discontinued treatment. The “on- and off-treatment” terminology used in our manuscript is a simpler way to describe this approach. In ETHOS, the secondary endpoint of all-cause mortality was prespecified to include both on- and off-treatment data. This analysis was reported in both the primary and secondary publications (1, 2). We published the “on-treatment only” analysis in our secondary publication to provide a comparison with other studies that used this approach.

We acknowledge that ETHOS was not designed primarily as a mortality study, and the sample size was therefore not selected on the basis of considerations of power for this secondary endpoint. However, we disagree with the conclusions of the post hoc power analysis conducted by Rogliani and Calzetta. Such an analysis is of limited usefulness when a significant effect was shown, and the fact that a treatment effect of budesonide/glycopyrrolate/formoterol fumarate (BGF) 320 versus glycopyrrolate/formoterol fumarate (GFF) was detected means that this comparison was not affected by type II error (failing to show a difference when one truly exists). Although we cannot rule out a type II error in the comparisons of BGF 160 with dual therapies, which did not reach P < 0.05, we did not make any conclusive claims of a benefit, or lack thereof, with this dose. It should also be noted that the original analysis for this endpoint, published in our primary publication, was in fact included in the testing hierarchy for type I error control (2). However, the P values for this endpoint are described as unadjusted because of the results for severe exacerbations higher in the hierarchy, which did not reach significance. Although there was no adjustment for multiplicity for the supplemental analyses published in our secondary manuscript (1), a type I error (detecting a difference that does not truly exist) is unlikely given the size of the P value, the prespecification of the endpoint as secondary, the consistency with the original analysis, and the corroborating evidence from other studies.

We also wish to clarify that the original mortality analyses, published in the primary publication (2), were not missing 384 patients. These patients were included, but their vital status information did not extend to 52 weeks. The extensive follow-up conducted to retrieve 52-week vital status data for these patients did not apply to the other endpoints presented in the primary publication, which were analyzed without postdiscontinuation data. Although posttreatment data were collected for other endpoints, post-study information other than vital status could not have been collected because informed consent was no longer in place. Therefore, except for mortality, the results represent the final analyses with all available data.

Regarding the choice of covariates in the model, we disagree with the assertion that including more covariates in the model would have affected the results. As ETHOS was a randomized study, potential confounders were balanced across groups (2). Therefore, there is no reason to expect a large impact of additional covariates on the estimates. We have performed sensitivity analyses including eosinophil counts or prior inhaled corticosteroid use as additional covariates in the model; in both analyses, the hazard ratio for BGF 320 versus GFF remained 0.51. Furthermore, the subgroup analyses reported in our paper show that the incidence of death was lower with BGF 320 than with GFF across all subgroups examined for exacerbation history, FEV₁% predicted, and prior medication (except for patients not previously on inhaled corticosteroids, for whom there was a small number of events [n = 5–8 across groups], preventing meaningful conclusions), and across a broad range of eosinophil counts (1), suggesting that it is highly unlikely that this was a chance finding influenced by confounding.

We disagree with the suggestion that there may have been a different level of effort in retrieving deaths between groups. Our supplemental follow-up retrieved 52-week vital status for 99.65% of patients in the intent-to-treat population. For the 30 patients whose 52-week vital status remained unknown in the final retrieved dataset (including n = 10 on BGF 320 and n = 5 on GFF), we reported tipping-point analyses to examine the possible impact of these patients. If all 5 missing patients on GFF were alive and up to 8 of 10 patients on BGF 320 died the day after we last knew they were alive, the treatment comparison would remain significant (1). It is correct that not all retrieved deaths were included in the analysis and that the percentage of excluded deaths varied across groups. This is because the analysis only included deaths up to 52 weeks, and a greater percentage of deaths in the dual-therapy groups occurred within 52 weeks. We performed a sensitivity analysis including all retrieved deaths, regardless of how late they occurred (i.e., 37 deaths for BGF 320 and 64 deaths for GFF); this produced a hazard ratio of 0.46 (95% confidence interval, 0.30–0.71; P = 0.0004).

Overall, we agree that further trials assessing the benefits of triple therapy on mortality would be welcome. However, there would be substantial practical difficulties in conducting such trials, including the fact that two large studies (IMPACT [Informing the Pathway of Chronic Obstructive Pulmonary Disease Treatment] and ETHOS) have now shown a benefit of triple therapy on mortality, raising ethical questions on the appropriateness of randomizing this patient population to long-term dual therapy. As mentioned by Rogliani and Calzetta, clinical trial populations are only partially representative of real-life populations, and any future trials would have this same limitation. Nevertheless, many studies have now provided evidence that triple therapy reduces exacerbations and improves lung function and patient-reported outcomes compared with dual therapies (3). The findings from IMPACT and ETHOS on mortality add support for the benefit of these therapies in improving the lives of patients with chronic obstructive pulmonary disease.