Risk of interstitial lung disease in patients treated for atrial fibrillation with dronedarone versus other antiarrhythmics

Abstract Purpose To compare risks of interstitial lung disease (ILD) between patients treated with dronedarone versus other antiarrhythmics. Methods Parallel retrospective cohort studies were conducted in the United States Department of Defense Military Health System database (DoD) and the HealthCore Integrated Research Database (HIRD). Study patients were treated for atrial fibrillation (AF) with dronedarone, amiodarone, sotalol, or flecainide. Propensity score matching was employed to create analysis cohorts balanced on baseline variables considered potential confounders of treatment decisions. The study period of July 20, 2008 through September 30, 2014 included a 1‐year baseline and minimum 6 months of follow‐up, for patients with drugs dispensed between July 20, 2009 and March 31, 2014. Suspect ILD outcomes were reviewed by independent adjudicators. Cox proportional hazards regression compared risk of confirmed ILD between dronedarone and each comparator cohort. A sensitivity analysis examined the effect of broadening the outcome definition. Results A total 72 ILD cases (52 DoD; 20 HIRD) were confirmed among 27 892 patients. ILD risk was significantly higher among amiodarone than dronedarone initiators in DoD (HR = 2.5; 95% CI = 1.1–5.3, p = 0.02). No difference was detected in HIRD (HR = 1.0; 95% CI = 0.4–2.4). Corresponding risks in sotalol and flecainide exposure groups did not differ significantly from dronedarone in either database. Conclusions ILD risk among AF patients initiated on dronedarone therapy was comparable to or lower than that of amiodarone initiators, and similar to that of new sotalol or flecainide users. This finding suggests that elevated ILD risk associated with amiodarone does not necessarily extend to dronedarone or other antiarrhythmic drugs.


| INTRODUCTION
Antiarrhythmic drugs are an important component in the treatment of atrial fibrillation (AF). They are often used in conjunction with anticoagulants, which are proven effective in decreasing the risk of thromboembolism and stroke in AF patients. 1,2 Dronedarone (Multaq ® ) received United States market approval in July 2009 for the treatment of AF and atrial flutter. Dronedarone is a benzofuran derivative with an electrophysiological profile resembling that of amiodarone, but with different relative effects on individual ion channels and with structural modifications intended to eliminate the non-cardiovascular adverse effects of amiodarone. 3,4 Amiodarone is today one of the most common causes of druginduced interstitial lung disease (ILD) in registries, with a reported incidence of 1.2%-8.8%. [5][6][7][8][9] In a retrospective study of 500 consecutive patients treated with amiodarone in Japan 40 patients (8%) presented signs of drug-induced ILD during a mean follow-up of 48 months. 10 ILD has also been reported among patients using other antiarrhythmic agents, including cases identified in post-marketing surveillance among persons using dronedarone. 11 The purpose of this study was to determine if dronedarone is associated with an elevated risk of ILD, relative to other antiarrhythmic agents. Patients were required to have a 365 day period (baseline), defined by continuous eligibility in their health plan prior to and including the index prescription's first dispensing date (index date). An AF diagnosis was defined as having one or more medical services with an AF diagnosis code (ICD-9-CM diagnosis code 427.31) during the baseline period. Exclusion criteria were a diagnosis of cancer, organ transplant, HIV, ILD, pneumonia or sarcoidosis during the baseline period.
Women who were or became pregnant in the 280 days immediately before or following the index date and patients with unknown gender were also excluded. To assure the index prescription was an incident prescription, patients were excluded if they were dispensed the same antiarrhythmic drug at any time during the baseline period. Patients with multiple study drug fills on the index date were also excluded.
Use of non-index study drugs during the baseline period was allowed.
The study outcome was a diagnosis of hospitalized ILD following the index date. Each patient's follow-up period began the day after their index date and continued until the earliest of (1) first inpatient service with an ILD-related diagnosis code, (2) death, (3) termination of health plan eligibility, or (4) the end of the study period. Any record of treatment in an acute care hospital or skilled nursing facility with a discharge diagnosis for any one of the ICD-9-CM codes 515, 516.3-516.37, 516.8 or 516.9 12 was flagged for clinical review and adjudication.
Given the complexity of an ILD diagnosis, presence of these diagnostic codes was only used to screen for potential ILD. An independent panel of three established ILD expert clinicians individually reviewed the medical records of patients with flagged events in order to determine if there was a definitive clinical diagnosis of new onset (incident) ILD. 13 Each clinician manually reviewed the de-identified patient profiles, which consisted of patient demographics (i.e., age and gender) and chronological medical history (i.e., inpatient and outpatient encounters, diagnoses and procedures, and drug dispensings) for up to 1 year preceding the patient's initial study drug fill through the end of the patient's follow-up period. Reviewers were blinded to specific study drug exposures (drug name, formulation and quantity dispensed), although the dispensing date and days supply were provided.
In addition to relevant diagnoses, the panelists reviewed available information with regard to timing of the ILD event relative to a patient's ongoing medical and drug history. While laboratory and radiology test results were omitted from the profiles, as they were not available for all patients, ILD Adjudicators looked at the relative timing of orders for pulmonary function tests and pulmonary imaging as a proxy. The panelists independently assigned each event a determination of "Yes," "No" or "Indeterminate" regarding its validity as a true

Key Points
• Hospitalized ILD risk among AF patients initiated on dronedarone therapy was comparable to that of new sotalol or flecainide users.
• The known elevated ILD risk associated with amiodarone does not seem to extend to the entire AAD therapy class including dronedarone.
Propensity score matching (PSM) was used to control for unmeasured confounding in treatment decisions. Potentially confounding medical history and personal characteristics were identified from baseline in-and out-patient records. Propensity scores (PS) were calculated from a logistic regression model using baseline covariates to predict the probability of being prescribed dronedarone versus a comparator drug. Variables included in that model were determined via stepwise processing, with variables entered through the step with the lowest Akaike Information Criterion value retained for the final PS model. The same covariates were employed in both databases to develop the reduced model from which the PS were calculated, although differences in the databases led to different variables being retained in the two reduced models. The reduced logistic regression models were derived from full models run against datasets combining all cohorts in the respective databases, and then independently fitted to three subsets respectively containing all the dronedarone patients and all patients from one of the comparator cohorts (see Table S1 for the list of covariates in the full and reduced PS models). PSM was conducted within each of the three sets of PS using a nearest-neighborwithin-caliper matching algorithm. The caliper width was 0.2 times the standard deviation of the PS for each cohort matched to dronedarone, and matching was conducted without replacement. To ensure that each cohort would be compared to the same group of reference patients, a qualifying PS-matched patient from each of the three comparator cohorts was required for a dronedarone patient to be included in the analysis. Dronedarone patients lacking a match in any one of the comparator cohorts were excluded. Therefore, the analytical (matched) datasets are smaller than, and not directly comparable to, the full sets of cohorts. Standardized differences were used to assess balance of baseline covariates between the dronedarone cohort and each of the comparator cohorts in the PS matched datasets.
The risk of ILD was modeled separately for dronedarone versus each of the three comparators using Cox Proportional Hazards (CPH) regression. Separate CPH models used exposure group (dronedarone as the reference drug) as the main independent variable predicting risk of a confirmed ILD outcome. The models were adjusted for confounders that remained unbalanced between the dronedarone and respective comparator cohorts after PSM.
Recognizing the non-specific clinical presentations of ILD, which may prevent a definitive diagnosis (e.g., cough, non-specific pathology upon chest X-ray), a sensitivity analysis was conducted using a broader definition of suspected ILD (i.e., addition of ICD-9-CM Codes 495.9 and 518.82). The sensitivity analysis was conducted only in the DoD cohorts and used the same methodology (i.e., PSM and CPH) as the primary analysis. The expanded definition of ILD was also applied to the baseline eligibility criteria, which resulted in 9580 patients being reclassified as prevalent for ILD and excluded from the analysis.
Applying the same inclusion and exclusion criteria to the HIRD data resulted in a total of 26 165 eligible patients. Like the DoD database, the largest index drug cohort was amiodarone (n = 12 615, 48.2%; Table 2  There were 113 suspected ILD cases in DoD and 33 in HIRD, identified from electronic health records or diagnostic claims during the follow-up period (Table 3). Adjudication resulted in a total of 72 confirmed ILD cases, 52 (46.0% of suspected) in DoD and 20 (60.6% of suspected) in HIRD.
As noted in the methodology, dronedarone was the reference group for all comparisons in the CPH models. In the DoD database, amiodarone exposure was associated with an elevated hazard of con-  This study is distinguished from previous research by its analysis of only events confirmed as ILD by expert clinicians. Adjudication of suspected cases was a critical element as ILD is not clearly identified in EHR and claims databases. Fewer than half of suspected ILD cases in the DoD and less than two-thirds in HIRD were confirmed. While more suspected cases were flagged in a sensitivity analysis using a broader ILD definition, fewer total cases were confirmed in that exercise as some suspected incident cases from the primary analysis were screened out as prevalent at baseline under the expanded definition.
Adjudicators evaluated all suspected cases based on their considerable diagnostic experience, although precision of the assessments would have benefitted from availability of relevant test results (e.g., pulmonary function tests and/or lung scan results) and more T A B L E 1 Baseline characteristics of dronedarone patients in the ILD propensity score matched analysis datasets: DOD database-July 20, 2009 to September 30, 2014 complete information for conditions where data were limited or possibly underreported (e.g., smoking).
Comparing the risk of ILD across treatments, particularly in database studies, is hampered by the complexity of a clinical diagnosis and low specificity of diagnostic codes, which limits analysis to databases with access to medical records. This limitation was evidenced by the paradoxically decreased number of confirmed ILD cases in the sensitivity analysis, where expanding the definition of ILD to include nonspecific or otherwise unclassified pulmonary conditions resulted in reclassification of some previously presumed incident cases. These analyses intentionally focused on patients with disease severe enough to require inpatient care, utilizing diagnostic codes to screen patient records for potential new diagnoses of ILD which were then confirmed through adjudication by clinical experts.

Dronedarone was approved with a Risk Evaluation and Mitigation
Strategy (REMS) that includes periodic assessment of education efforts regarding the drug's risks. 15 Although the effect of the REMS on clinical practice could not be explored in this study, low incidence of ILD in these two large populations might reflect care taken to monitor potential disease symptoms and pulmonary function, and discontinue treatment in patients with concerning findings before their conditions worsened. Low outcome rates are also consistent with the relatively small proportions of patients with baseline risk factors, which might also suggest careful attention to prescribing information.
In addition, the similarly low prevalence of baseline risk factors across the comparator cohorts might suggest that the REMS accompanying dronedarone's approval could have affected prescribing patterns for these other antiarrhythmic drugs.
In conclusion, dronedarone was associated with comparable or significantly lower risk of hospitalized ILD than amiodarone therapy in AF patients, and with consistently comparable hospitalized ILD risk in similar patients initiated on sotalol or flecainide. The data suggest that the elevated risk of hospitalized ILD associated with amiodarone use in AF patients does not necessarily extend to dronedarone and other antiarrhythmic drugs.