Temporal trends in prevalence and outcomes of atrial fibrillation in patients undergoing percutaneous coronary intervention

Abstract Background Atrial fibrillation (AF) is the most common arrhythmia in patients undergoing percutaneous coronary intervention (PCI). Hypothesis Large administrative data may provide further insight into temporal trends in the prevalence and burden of AF in patients who underwent PCI. Methods Using the National Inpatient Sample database in the U.S., AF patients ≥18 years who underwent PCI between 2005 and 2014 and were identified by the International Classification of Diseases, ninth revision, Clinical Modification, were examined. In‐hospital mortality, morbidity, resource use, and medical costs were evaluated in crude and propensity‐matched analyses. Results Among an estimated 6 272 232 hospitalizations, of patients undergoing PCI, AF prevalence was 9.9% and steadily increased from 8.6% to 12.0% between 2005 and 2014 (P < .001); there was also a greater proportion of comorbidities. There was a marked increase in AF prevalence among those aged ≥65 years and those undergoing elective PCIs. AF was independently associated with higher in‐hospital mortality and higher rates of transient ischaemic attack/stroke, bleeding complications, and non‐home discharge. Excessive in‐hospital mortality, stroke rate, gastrointestinal bleeding, blood transfusion, length of stay, and costs among AF hospitalizations were consistently observed throughout the study period. Conclusion AF becomes more prevalent in patients undergoing PCI, possibly due to a higher comorbidity, particularly in elderly patients with non‐acute indications. Less favorable trends in mortality, bleeding, and stroke among AF patients who underwent PCI were consistent over time. Continuous efforts are needed to improve outcomes and manage strategies for AF patients undergoing PCI.


K E Y W O R D S
atrial fibrillation, percutaneous coronary intervention, epidemiology

| INTRODUCTION
Atrial fibrillation (AF) is the most common type of arrhythmia. The prevalence of AF in the general population of the US is 1% among all adults and 9% for those older than 80 years. 1 Approximately 5%-10% of patients who have undergone percutaneous coronary intervention (PCI) or acute coronary syndrome have concomitant AF. [2][3][4][5] These patients frequently require anticoagulants plus dual antiplatelet therapy and typically have significant in-hospital morbidity and mortality. 2,6 Until recently, considerable efforts have been made to determine the optimal antithrombotic therapeutic approach for simultaneously preventing thromboembolic events and bleeding complications. [7][8][9][10][11][12][13] The implementation of PCIs has changed in recent years, with second-generation drug-eluting stents being widely used and patients with greater comorbidity burdens undergoing procedures. 14,15 As the prevalence of AF in the US is expected to double from 2010 to 2050 due to an aging society, 1 the management of AF patients undergoing PCI will impose a large healthcare burden. However, there are minimal data based on nationally representative cohorts in the US that focus on contemporary trends in the prevalence, in-hospital outcomes, and medical costs of AF patients who have undergone PCI.
The aims of this study were to describe (a) changing trends in the prevalence of AF among patients who underwent PCI, (b) patient/hospital characteristics and clinical risk profiles of these patients, and (c) temporal trends and outcomes of in-hospital mortality, in-hospital morbidity, medical costs, and resource use among propensity-matched cohorts of AF and non-AF patients who underwent PCI.

| Study data
We used the National Inpatient Sample (NIS) database derived from administrative data between 2005 and 2014 for this retrospective, observational study. The NIS is the largest publicly available, all-payer administrative database in the US and contains information regarding patient discharge from around 4500 hospitals in~45 states. 16 This database contains clinical and resource use information on roughly 7 million unweighted discharges annually, which represents around 35 million weighted discharges for national estimates. We calculated the national estimates using the discharge weights provided in the NIS. The NIS represents~20% of stratified US in-patient hospitalizations and excludes rehabilitation and long-term care hospitals. A variety of studies for the association of clinical settings and procedures can be conducted using the NIS. 17,18 Because ICD code 9CM was changed to ICD10 in NIS after the middle of 2015, we used data from the 10 years leading up to 2014, prior to the ICD coding change. Institutional review board approval and informed consent were waived because the NIS is a publicly available database that contains completely de-identified patient information.

| Study population
We used international classification of diseases, ninth edition, clinical modification (ICD-9-CM) procedure codes (0066, 3601, 3602, 3605, 3606, 3607, 1755) to identify all hospitalizations of ages ≥18 years who underwent PCI (weighted n = 6 515 521; unweighted n = 1 322 243). The validation and use of the ICD-9-CM codes for identifying PCI numbers in the US has been described in previous studies. 19,20 After excluding hospitalizations with missing values for in-hospital death, length of stay, discharge location, primary insurance, and in-hospital costs, the final cohort included 6 272 232 hospitalizations (unweighted n = 1 272 853). As the variable for race was missing in 18% of hospitalizations, we performed multiple imputation using the R package mi (version 1.0), 21 which imputes missing values in an approximate Bayesian framework. Hospital bed size, location, region, and median household income were entered into the model. The study cohort was divided into two groups according to the presence or absence of AF using ICD-9-CM code 42731, which was used in similar studies. 22,23 Elixhauser Comorbidity Software (version 3.7) was used to identify congestive heart failure, peripheral artery disease, diabetes, chronic pulmonary disease, chronic renal failure, obesity, anemia, and depression. 24 Clinical Classifications Software for the ICD-9-CM was used to identify dyslipidemia and blood transfusion. 25 Other ICD-9-CM codes for identifying patient/hospital characteristics are summarized in Table S1.

| Propensity score matching
To reduce any bias associated with patient/hospital characteristics among AF and non-AF hospitalizations, we performed 1:1 the nearestneighbor matching with a caliper of 0.15 using the MatchIt R package (v3.0.2). 26 Figure S1). Absolute T A B L E 1 Baseline patient and hospital characteristics for hospitalizations with and without atrial fibrillation undergoing percutaneous coronary intervention

| Statistical analysis
In-hospital outcome and trend analyses were performed in unweighted and weighted data, respectively, which are a standard methodology used in other studies using NIS data. 23,28 The total charges provided in the NIS of each hospitalization were converted to cost estimates using the group average, all-payer, in-patient, cost-tocharge ratios. All in-hospital costs were converted to projected estimates for the year 2014 using annual inflation rates on the basis of consumer price index data available from the Bureau of Labor Statistics. 29 For trend analysis, the Mann-Kendall test was performed for pro-  Figure 1). Figure S2 shows that hospitalizations who underwent elective PCI had significant change toward a further increase in AF prevalence around 2008 by Davies test (P = .025). Figure S3 shows that the overall number  Table 1. These characteristics became well-balanced between the two matched groups (with and without AF, Figure S1), with the standardized difference for all variables being <3.0%. The trend stratified by AF for mean age and patient comorbidities in the weighted cohorts are summarized in Figure S4. There was a significant increase in the rates of congestive heart failure, diabetes, peripheral artery disease, chronic pulmonary disease, chronic kidney disease, prior myocardial infarction, and anemia; however, there was no significant difference in mean age among AF hospitalizations during the study period ( Figure S4).
The in-hospital procedures and outcomes for AF and non-AF hospitalizations among the crude and propensity matched cohorts are presented in Table 2. AF hospitalizations undergoing PCI were more  Figure S4).
The results of multivariate logistic regression analysis for inhospital outcomes are shown in Table 3. In-hospital mortality was significantly higher in AF hospitalizations who underwent PCI than in non-AF hospitalizations (3.2% vs 1.5% and 3.2% vs 2.8%, respectively; adjusted odds ratio: 1.16; 95% confidence interval: 1.10-1.21). There was no significant difference in the annual inhospital mortality in AF hospitalizations during the study period (P = .15; Figure 2). AF hospitalizations were more likely to have prolonged length of stay, non-home discharge, higher hospitalization costs, TIA/stroke, gastrointestinal bleeding, vascular complications, blood transfusion, and cardiogenic shock (P < .001 for all). During the study period, there was a significant downward trend in vascular complications and an upward trend in non-home discharge rates in AF hospitalizations (P < .001 and P = .002, respectively). However, there were no significant changes in the trends of length of stay, medical costs, TIA/stroke, gastrointestinal bleeding, and blood transfusion in AF hospitalizations during the study period. Increased rates of in-hospital mortality, in-hospital morbidity, costs, and prolonged length of stay in AF hospitalizations were consistently observed throughout the study period ( Figure 2).
T A B L E 2 In-hospital procedures and outcomes in hospitalizations with and without atrial fibrillation undergoing percutaneous coronary intervention In this nationwide study, almost 1 in 10 patients had AF during their hospitalization, and AF prevalence steadily increased by 14.0% over the study period. The reason for the increase in AF among patients who underwent PCI might be that these individuals were more likely to have clinically predisposed AF comorbidities 32 such as heart failure, hypertension, diabetes, chronic pulmonary disease, and a history of myocardial infarction. In addition, the age-stratified analysis showed that a higher AF prevalence was observed in patients ≥65 years old, despite a consistent mean age of AF patients. This suggests that a greater accumulation of comorbidities in the elderly largely contributes to the overall increase in AF prevalence in this population.
In addition, prevalence of AF has increased among the general population of other developed countries, possibly due to similar reasons, 33,34 indicating that AF in patients undergoing PCI will also increase.
When stratifying by PCI indications, AF prevalence in STEMI patients remained nearly the same (only increasing from 9.1% to 9.8% between 2005 and 2014). This finding is consistent with a previous multinational study of acute coronary syndrome, which reported a slightly decreased prevalence from 12.9% in 2000 to 11.1% in 2007, 3 as well as with a US community-based study of acute myocardial infarction that reported an almost invariable AF prevalence from 1975 to 2005. 35 However, we observed a significant increase in AF prevalence in non-ST-segment elevation myocardial infarction (NSTEMI) and elective PCI patients from 8.6% to 11.7% and 8.3% to 17.2%, respectively. Furthermore, a significant accelerated increase in AF prevalence around 2008 was detected among patients undergoing elective PCI. These findings might be secondary to the introduction of appropriateness criteria to optimize the benefit-risk balance for PCI, 36 the shift toward more out-patient care for PCI, 28  Our data identified racial disparities in the prevalence of AF in patients undergoing PCI; a higher prevalence of AF was observed in white than in black and Hispanic patients. Previous studies also reported that AF is less prevalent and left atrial appendage occlusions are performed less frequently in black and Hispanic than in white patients. [37][38][39] Predisposing genetic factors might explain these racial disparities in prevalence of AF. 40 However, further studies are needed to clarify the issues.
Our study reported a 1.5% in-hospital mortality among non-AF patients, compared to a 3.2% risk among AF patients in the crude cohort. These rates were comparable with a previous study that analyzed a prospective multicenter registry in Michigan, which assessed AF history among patients who underwent PCI and reported in- In this study, we showed that AF patients were at a higher risk for thrombotic and bleeding complications, such as TIA/stroke, gastrointestinal bleeding, blood transfusion, and vascular complications in the crude cohort, which is consistent with results of prior studies. 2,4,31 The association between AF and these in-hospital complications remained significant after vigorous propensity matching between AF and non-AF patients; these results were similar to those of previous studies. 5 higher CHADS2 scores were more likely to paradoxically discontinue the anticoagulants at discharge, possibly due to concerns for bleeding risks. 5 In addition, our propensity-matched analysis showed that AF patients are less likely to receive drug-eluting stents, which is congruent with the results of a previous study, 2 probably due to similar concerns with an increased risk of bleeding.
Consistent with the rapid evolution of the antithrombotic strategy for AF and PCI, guidelines and consensus documents incorporating new evidences are frequently updated. 37 More recent guidelines recommend considering a short duration of triple therapy or even avoiding the therapy altogether based on patients' thromboembolic and bleeding risk. 33,42 Several key randomized trials support the elevated safety of direct oral anticoagulant intake with single antiplatelet therapy (double antithrombotic therapy) compared with a regimen of triple therapy with vitamin K agonists. 7

| Study limitations
Our study has a number of limitations, which are similar to those of other studies using large administrative databases. First, this study has the potential for any bias inherent to retrospective observational studies. Second, there may be substantial coding errors and coding bias derived from the ICD-9-CM codes. However, similar approaches for capturing data on the prevalence AF and PCI were used in previous studies. 19,20,22,23 In addition, several outcome measures examined in our study, including in-hospital mortality, length of stay, cost, and non-home discharge, are rarely miscoded. Third, the NIS does not record data regarding diagnostic findings of coronary angiography, procedural characteristics of PCI, peri-procedural medications, electrocardiograms, left ventricular ejection fraction by echocardiography, or laboratory variables. In addition, data are not available in the NIS to calculate the CHADS2 scores for assessing the risk of thromboembolism. Fourth, the propensity-matching and ad-hoc multivariate analysis did not account for antithrombotic regimen and procedural characteristics. However, previous study used a similar approach. 2 Finally, we could not distinguish between preexisting AF and new-onset AF. However, it has been reported that both preexisting AF and newonset AF are associated with worse clinical outcomes 2,3,5 and that previously undetected preexisting AF may not preclude a case of paroxysmal AF. Despite these limitations, we believe that our study was strengthened by using the NIS to analyze real-world clinical practice with a nation-wide estimate that represents in-patient data in the U.S.

| CONCLUSIONS
Using the largest nationally representative cohort of trends in demographics and outcomes for AF patients who underwent PCI, we demonstrated that the overall AF prevalence among patients who underwent PCI continuously increased, particularly in elderly patients and those who underwent elective PCIs. AF patients continued to show high in-hospital mortality, morbidity, and medical costs over the recent decade. This study highlights the continued need to identify preventive and management strategies to reduce risks and costs associated with AF patients undergoing PCI.