Comparative safety of inhaled corticosteroids and macrolides in Medicare enrolees with bronchiectasis

Discussion

Overall, adverse safety outcomes occurred at rates ranging from 2.5/100 person-years (myocardial infarction and hip fracture) to just over 25/100 person-years (arrhythmia) in older bronchiectasis patients treated with macrolides and ICS. In our study, chronic macrolide use among Medicare patients was associated with a decreased risk of arrhythmia, and not associated with a statistically significant increased risk of myocardial infarction when compared to ICS. However, chronic macrolide monotherapy was associated with a statistically significant, moderately increased risk of hearing loss compared to ICS. Other safety outcomes were similar in the PS-adjusted analysis.

Our study evaluated two cardiac outcomes, arrhythmia and myocardial infarction, which are potential risks of azithromycin. Macrolides can also induce QT prolongation, particularly when used with drugs that cause QT prolongation or in patients with conditions associated with QT prolongation [8]. QT prolongation is associated with torsade de pointes and sudden cardiac death [22]. In 2012, near the end of our observation period, Ray et al. [7] linked azithromycin use to sudden cardiac death during the first 1–5 days of use in a population-based study. However, multiple subsequent studies of varying cardiac outcomes 1–90 days after azithromycin initiation have produced conflicting results, with risks confirmed for some high-risk populations [23–26]. Two meta-analyses found an increased risk of sudden cardiovascular disease and ventricular tachyarrhythmias [9] and a small but statistically significant increase in myocardial infarction, though less likely for azithromycin than erythromycin [10]. However, another meta-analysis found no long-term cardiovascular risk assessed >30 days to >3 years [12]. A fourth meta-analysis identified a possible increased risk of cardiac death in adults aged >48 years, but not overall, in those using macrolides compared to non-macrolide [11]. Most relevant to our analysis, there are some prior data on safety of long-term use of azithromycin in patients with other chronic lung disease diagnoses. A prospective, randomised, placebo-controlled trial of 250 mg azithromycin daily for 12 months evaluated efficacy and safety in 1142 COPD patients [27]. There was no observed increase in risk of QTc prolongation in the azithromycin group, although patients with risk factors for QTc prolongation were excluded, as is common in randomised clinical trials.

In our study, macrolides were associated with a lower risk of arrhythmia than ICS, even after excluding those with a history of arrhythmia. One possible explanation is that the LABAs the majority of ICS patients were taking could increase the risk of arrhythmia, although a placebo-controlled, randomised trial in high-risk COPD patients found similar rates comparing placebo, LABA alone, ICS alone and ICS/LABA combination [28]. Similarly we did not see a difference in aHR stratifying by ICS/LABA and ICS alone. Findings of a non-elevated risk for macrolides are consistent with a population-based retrospective cohort study in Canada, where macrolide and non-macrolide antibiotics had a similar 30-day risk of ventricular arrhythmia (PS-matched RR 1.06, 95% CI 0.83–1.36) in adults ≥65 years old [29]. Age, smoking, male sex and pulmonary diseases are risk factors for atrial fibrillation, the most common arrhythmia diagnosis [30]. Further, atrial fibrillation has been associated with high-dose oral or parenteral dosing (≥7.5 mg of prednisone equivalent), although not low dose or ICS in patients with and without asthma/COPD [31]. As previously published, our Medicare population taking ICS was more likely to be male, to have COPD or asthma, and have a higher Charlson morbidity index [15]. It is possible, despite adjustment for PS and oral steroids, that unmeasured risk factors were not balanced between the two exposure groups or that there was residual confounding. Regardless, Albert et al. [32] suggest that clinical history and ECG prior to starting chronic therapy for lung disease would be cost-effective and reassuring to the patient and physician.

Bone loss that increases the risk of hip fracture has long been associated with oral corticosteroid use [18]. As recently described in patients with chronic airway diseases, inhaled corticosteroid use is associated with fracture risk with longer duration and higher doses [33–35]. Our findings in a Medicare population with a median 60 days of ICS use and 61% with low dose are consistent with observed minimal or no risk with shorter duration and lower doses [33]. However, the relatively short duration of exposure may have meant it was not possible to detect an increased risk if it occurs after longer use. One longitudinal study observed this risk only with very long duration (>4 years) and high doses (>1000 µg in fluticasone equivalents) (RR 1.10 95% CI 1.02–1.19) [35]. However, a PS-matched study in Taiwan reported a hip fracture aHR of 1.54 (95% CI 1.31–1.81) in adults age ≥65 years comparing ICS users to non-users, and a clear dose response over 250 µg fluticasone equivalents [34].

Our included bronchiectasis population had a relatively high incidence of opportunistic infections, 10-fold higher than non-viral rates of 1.8–3/1000 reported in younger adults on treatment for rheumatoid arthritis [36]; even accounting for the fact that half in our study were due to herpes zoster diagnoses. This may be influenced by low herpes zoster vaccine uptake (estimated 14% of adults over 65 between 2007 and 2013 in a Marketscan database study) in this high-risk population [37]. In our study, the crude risk of opportunistic infections was higher in those taking macrolides, but not statistically different comparing those using ICS and macrolides after adjusting for PS, oral corticosteroid use and NTM history. This finding is not readily understood. Some of these infections likely reflect the underlying lung disease which could put them at higher risk for pulmonary infections (e.g. endemic coccidioidomycosis and aspergillosis). To mitigate against misclassifying fungal colonisation (e.g. Candida or Aspergillus) as disease, we required a code plus evidence of >30 days of anti-fungal treatment. Our definitions for herpes zoster also included anti-viral treatment, to increase the likelihood that we identified active cases.

A systematic review identified 41 cases of audiometrically confirmed sensorineural hearing loss associated with oral azithromycin in the literature [14]. Two of these were prospective studies of patients treated for NTM disease, where azithromycin is given in lower doses (e.g. 250 mg daily) similar to macrolide monotherapy [38, 39]. The previously described randomised controlled trial of azithromycin in COPD patients showed a small but significantly higher proportion of patients with audiogram-confirmed hearing decrement in patients receiving azithromycin (25% versus 20% in the placebo group, p=0.04) [27]. Patients had no hearing abnormalities at baseline, and the conclusion was that there was no hearing impairment because hearing loss improved upon repeat-testing regardless of whether the study drug was discontinued or not. In contrast, our study provides population-based evidence of a moderately increased risk of sensorineural hearing loss due to macrolides with an active treatment comparison, and a crude incidence of around 10/100 patient-years in the older Medicare population with bronchiectasis. Physicians should consider evaluating patients who are initiating chronic macrolide therapy for baseline hearing loss, in order to facilitate diagnosis of macrolide-associated ototoxicity, if it could impact treatment decisions.

Strengths of the study include the fact that it is a large population-based study of older bronchiectasis patients. In our clinical experience in the USA, the majority of patients with bronchiectasis are cared for outside of specialty clinics. ICS are used routinely without evidence of efficacy, and macrolides are used routinely without sufficient evidence of risks [19]. Further, there are no expert guidelines in the USA for bronchiectasis management, and our data support the need for such guidelines. Our study fills a gap in knowledge that is unlikely to be answered with any other data. The new-user comparative effectiveness design has been used extensively in other clinical settings of chronic disease (e.g. rheumatoid arthritis, COPD, cancer) to improve the validity of findings [20, 40, 41]. We compare two active treatments with PS decile adjustment to account for differences in the two treatment populations that could bias the analyses. As a result we are comparing the medications in patients with similar characteristics. We included important patient and clinical characteristics in the PS, and confirmed that key demographics and underlying diagnoses are balanced across PS deciles.

One of the main limitations of claims data is the lack of confirmation of diagnoses and misclassification of outcomes. We used validated outcome measures, with the exception of sensorineural hearing loss, which was defined by our expert study advisory committee. Second, our patient population probably included a mix of primary and secondary bronchiectasis. However, we believe the evidence supports that most are primary bronchiectasis with concomitant (or erroneous) diagnosis of COPD. First, the demographics are majority female (70%) and older (mean age 74 years), suggestive of primary bronchiectasis [42, 43]. Further, we required that codes be given by a pulmonologist, so we have more confidence in the diagnosis. Last, the bronchiectasis diagnostic claim had to be associated with a clinical visit raising the likelihood that it was clinically relevant (if not, COPD or asthma would have been used as the code associated with the visit). Bronchiectasis is not normally a diagnosis used as a general code for billing, unlike COPD. Our comparison of two drugs with very different side-effect profiles could also be considered a weakness, if there is residual unmeasured confounding from prescribing practices or patients are monitored differently. Outcomes that are recognised side-effects, such as hearing loss, could be more closely monitored in patients taking macrolides, which would inflate the observed risk due to detection bias. However, we saw the opposite for arrhythmia, which could be affected by a decrease in the use of macrolides in high-risk patients due to awareness of cardiac risks. In addition, we did not control for ototoxic concomitant medications in the hearing loss model. Last, our population was older, so the results may not be generalisable to a younger population of bronchiectasis patients who are likely to have lower rates of adverse events such as cardiac abnormalities or hearing loss.

Our results suggest that the risk of hearing loss is increased in older patients taking macrolides for >30 days and that macrolides were associated with a lower risk of arrhythmia. Older patients who are on chronic macrolide treatment should be monitored for hearing loss where appropriate, including a baseline evaluation. Further study of potential long-term risks, particularly the development of antibiotic-resistant microorganisms, and benefits of macrolide monotherapy in this older, high-risk population is recommended.