Heterogeneity of treatment response in bronchiectasis clinical trials

Discussion

This study has reanalysed three prior bronchiectasis trials investigating drugs with different mechanisms of action to explore the concept of “treatment response” in patients with bronchiectasis. We report a number of findings which may be important for future therapeutic development. First, large placebo responses are common, particularly for end-points such as QoL, and these must be accounted for in future trial design and power calculations. Second, FEV₁, the end-point most frequently used in CF clinical trials, is poorly responsive to all interventions in bronchiectasis and is unlikely to be a useful efficacy end-point going forward. Third, across multiple different types of therapy (mucoactive, antibiotic and anti-inflammatory), we saw that patients with a history of smoking were less likely to respond to treatment compared with nonsmokers in terms of meaningful improvements in QoL, FEV₁ or exacerbations. There is an active debate about the interaction between chronic obstructive pulmonary disease (COPD) and bronchiectasis, and how smoking-related lung disease may affect the prognosis of the disease. Studies typically aim to exclude patients with investigator-diagnosed COPD, but identifying which condition is predominant can be challenging. Our data suggest caution in including patients with a history of cigarette smoking in bronchiectasis trials as they were consistently less responsive to multiple therapies.

Our study also showed that patients with bronchiectasis respond very differently to distinct treatments. The response is heterogeneous in all cases and there is very little relationship between response to different outcomes such as changes in QoL, improvement in FEV₁ or exacerbations. New treatments for bronchiectasis are urgently needed and the repeated failure of bronchiectasis clinical trials necessitates a re-evaluation.

Different primary end-points have been used in bronchiectasis clinical trials [27]. Exacerbations, either in terms of frequency or time to first event, have been used in many recent trials [9–11, 14, 15, 19–21] due to the fact they are strongly related to worse clinical outcomes such as increased short- and long-term mortality [26]. In our study, we demonstrated that a very poor relationship exists between absence of exacerbations and changes in lung function or QoL. Particularly, only 5% of the patients in the mannitol trial and 6% of the patients in the azithromycin trial showed absence of exacerbations associated with an improvement in QoL and FEV₁ at the same time. This suggests, intriguingly, that the mechanisms leading to exacerbation may be distinct from those leading to daily symptoms or lung function impairment, supporting the concept that frequent exacerbators are a phenotype potentially with a distinct underlying biology. All these findings suggest that better endo-phenotyping and a more personalised targeted therapeutic approach should be developed for bronchiectasis.

FEV₁ is a primary end-point frequently used in clinical COPD trials, where its MCID above 100 mL is well defined [25]. In CF, several trials using inhaled antibiotics have also used this end-point [18]. However, in bronchiectasis, although a decline in lung function has been correlated with worse outcomes, only few clinical trials have considered change in FEV₁ as a primary end-point [13, 19]. Our study demonstrates that only 5–14% of the patients experienced an increase of FEV₁ over the course of the trials and that FEV₁ was poorly correlated with improvement in QoL or an absence of exacerbations. Furthermore, no differences among the active treatment and placebo arms were observed in all three trials evaluated regarding changes in FEV₁, suggesting that this is not an effective measure in bronchiectasis clinical trials. Indeed, considering the vicious cycle of bronchiectasis, it has now been clearly demonstrated that an effective mucoactive treatment, an effective anti-inflammatory and a series of effective antibiotics have failed to result in clinically significant improvements in FEV₁, suggesting that this should no longer be considered a meaningful efficacy end-point for trials targeting these mechanisms of action [11, 12, 21, 28]. On the other side, exacerbations showed clear statistically superiority of active treatment over placebo for the azithromycin and mannitol trials, and they should be considered a clinically important and responsive end-point for bronchiectasis trials.

The rate of response in placebo patients in all trials was relatively high, a phenomenon that has been discussed extensively in the clinical trial literature [9, 10, 29]. Regarding QoL, the placebo response rates ranging from 26% to 56% demonstrate one of the key challenges in bronchiectasis clinical trials. Patients may receive higher quality care or may become more compliant to regular therapies such as airway clearance upon entry into a clinical trial. In addition, the placebo in some studies includes saline or low doses of mannitol, which may function as mucoactive drugs in their own right. The very high placebo responses illustrated in our present study therefore represent an important barrier to successful trials that needs to be addressed. However, despite the high rate of treatment response, both mannitol and azithromycin demonstrated statistically significant increases in the rate of response for active treatment versus placebo, suggesting that this is a valid way to evaluate response.

There are a number of important implications of the dissociation between end-points in bronchiectasis trials. First, in clinical practice patients are often simplistically divided into responders and nonresponders. Our data suggest this is not possible in bronchiectasis: a patient experiencing a symptom improvement with a treatment may experience no improvement in exacerbation frequency, for example. The common practice of giving short-term trials of treatment (e.g. 3 months) with an inhaled antibiotic or macrolide, for example, and discontinuing therapy based on a lack of symptomatic improvement would therefore be inappropriate because a lack of a symptom improvement does not exclude a potential reduction in exacerbations. From a research perspective, the failure of multiple clinical trials has led to the search for biomarkers or clinical factors associated with response. Our recent study showed a relationship between airway bacterial load and symptom improvements with inhaled antibiotic treatment [30], a finding that was recently replicated in the ORBIT trial programme [31]. The present data suggest that it cannot be assumed that a biomarker predicting symptom improvements would also predict response in terms of exacerbations. Different biomarkers may be required for different end-points. Second, the clinical characteristics of the responders are also heterogeneous in all trials evaluated, with no clinical predictors of treatment response. However, smoking history showed less QoL improvement in both the mannitol and azithromycin trials, and more exacerbations in the azithromycin trial. In COPD, current smokers also experienced a nonresponse to azithromycin in reducing exacerbations [32]. These findings may suggest that smokers were poorly responsive. Our data support the view that clinical characteristics alone will be insufficient to identify responders to most therapies and therefore overcome the intrinsic heterogeneity of bronchiectasis. We therefore suggest the development of biomarkers to support personalised medicine.

Our study has a number of limitations. First, these are post hoc analyses and therefore results need to be confirmed in further prospective studies. Second, different QoL questionnaires were used in the different studies we evaluated (SGRQ and QoL-B-RSS), although the MCIDs are established for both tools [22, 24]. Third, only two of the three trials evaluated had 12 months of follow-up, which is necessary to evaluate exacerbation response. However, we decided to include the aztreonam trial, which has only 12 weeks of follow-up, due to the importance of including an inhaled antibiotic trial, and exacerbations were excluded as treatment response criteria. In addition, the optimal time frame to assess response to QoL or lung function is not well defined. We used symptom and lung function improvements between the beginning and the end of the treatment period but acknowledge it is possible patients experienced improvements during the trial that may not be sustained. The clinical significance of transient changes in FEV₁ or symptoms is not clear and therefore was not analysed in this study. Fourth, the MCID threshold of 100 mL of FEV₁ has been used previously in CF and COPD [25, 33] but it is not validated in bronchiectasis. Since we showed such a poor relationship between FEV₁ and other clinically relevant features, and the MCID is typically determined through an anchor to another clinically meaningful end-point, we would argue FEV₁ is shown in this study to be poorly reflective of bronchiectasis treatment response and may therefore not have a valid MCID in this disease. Why the same treatment, e.g. inhaled antibiotics, produces marked improvements in FEV₁ in CF but not in bronchiectasis remains unexplained from a mechanistic point of view.

The assumption that treatment responses for different end-points are concordant is a key clinical and regulatory assumption that we have shown to be incorrect. Most trials follow a hierarchical analysis where if statistical significance is achieved on a primary end-point (e.g. exacerbations), further hypothesis testing is performed on the next end-point in the hierarchy (e.g. QoL). This approach makes assumptions about the relationship between these responses that are not supported by our present analysis. For example, if FEV₁ is placed above QoL in a hierarchy, it makes little sense to discount QoL data because an FEV₁ end-point was not statistically significant. These are measuring entirely distinct biological responses and are not expected to be concordant. Alternative methods of controlling for type 1 error should be considered.

In conclusion, we demonstrate that response to treatment is heterogeneous in three bronchiectasis randomised clinical trials and clinical baseline characteristics are not able to predict patterns of treatment response. QoL and exacerbations are good clinical markers of treatment response, although a large placebo effect was observed, especially when QoL is evaluated. FEV₁ is poorly responsive and should not be considered as an efficacy end-point for mucoactive, antibiotic or anti-inflammatory trials. Further studies focused on endo-phenotyping the disease are needed to better understand the heterogeneity of response to therapies in bronchiectasis.