Bidirectional association between aortic dissection and atrial fibrillation: Findings from a huge national database

Abstract Objective To explore the link between aortic dissection (AD) and atrial fibrillation (AF). Methods Using the National Health Insurance Research Database (NHIRD), cohorts were constructed for evaluating the incidence of AF in patients with AD (study 1) and the incident AD among AF patients (study 2) based on propensity matching analysis. Cox proportion hazard regression models were used to examine the effect of AD on the risk of AF, shown as hazard ratios (HRs) with 95% confidence intervals (CIs). Similar statistical procedures were used for study 2. Results The study 1 consisted of 11 813 patients in the AD cohort and 11 813 controls in the non‐AD cohort and the study 2 consisted of 190 494 patients in the AF cohort and 190 494 controls in the non‐AF cohort. The overall incidence density of AF was 1.32‐fold higher in the AD cohort than in the non‐AD cohort (11.1 and 8.3 per 1000 person‐years), with an adjusted HR (aHR) of 1.74 (95% CI = 1.53‐1.98). The AF cohort had 1.18‐fold higher incidence of AD than the non‐AF cohort (0.55 vs 0.47 per 1000 person‐years), with an aHR of 1.24 (95% CI = 1.07‐1.44). Conclusions Bidirectional association between AD and AF was shown for the first time in this study.


| INTRODUCTION
Aortic dissection (AD) is a life threating disease once left undiagnosed or untreated. 1,2 The phenomenon of AD presenting with atrial fibrillation (AF) has indeed been discussed previously [3][4][5][6][7][8][9] ; in case of subclinical AD, AF may occur and be a sign of alert. For clinicians who care for patients with AF, it is well known that stroke, heart failure, and death are common AF complications. [10][11][12] To date, whether there is an increased risk of AD in patients with AF remained unknown.
To provide additional evidence linking AD and AF from the view point of clinical aspect, investigation on the relationship between AF and AD might be thoughtful. Hence, we sought to utilize the Taiwanese national dataset to describe the incidence of AF in patients with AD and the incidence of AD in patients with AF, using propensity score methods, multivariate controlling and combining a large number of comorbidities in our analysis to explore the link between AD and AF.

| Sampled participants
For study 1, we identified patients aged 18 or older years with AD diagnosed between 2000 and 2010 (ICD-9-CM codes 441.0) and control individuals without AD. The index date for control patients was randomly appointed a month and day with the same index year of the matched AD cases. We defined the diagnosed date of AD as the index date for each patient. We excluded patients with a diagnosis of AF (ICD-9-CM codes 427.31) at baseline and those with incomplete medical records information. Patients in the AD and non-AD cohorts were selected by 1:1 matching based on a propensity score. 13 The propensity score was calculated using a logistic regression model to estimate the probability of the AD status assignment, based on the baseline variables including year of AD diagnosis, sex, age, and comorbidities of hypertension, diabetes mellitus (DM), hyperlipidemia, coronary heart disease (CHD), heart failure (HF), chronic obstructive pulmonary disease (COPD), peripheral artery disease (PAD), chronic kidney disease (CKD),

| Outcome
Subjects in the study 1 were followed until the diagnosis of AF or until withdrawal from the NHI program or death, or December 31, 2011.
Subjects in the study 2 were followed until the diagnosis of AD or until withdrawal from the NHI program or death, or December 31, 2011.

| Statistical analysis
For study 1, the distributions of the sex, age, and comorbidities were compared between the AD cohort and the non-AD cohort, and the differences were examined using the standardized mean difference (SMD). A SMD of ≤0.10 indicates a negligible difference between the two cohorts. The overall, sex-, age-, comorbidity-specific, and followup period incidence densities rate of AF (per 1000 person-years, PY) were measured for each cohort. Univariable and multivariable Cox proportion hazard regression models were used to examine the effect of AD on the risk of AF, shown as hazard ratios (HRs) with 95% confidence intervals (CIs). The multivariable-adjusted models included covariates that were not adequately balanced in Tables 1 and 3 (standardized difference > 0.1). The cumulative incidence curve of AF was computed using the Kaplan-Meier method and the differences between both cohorts were examined using the log-rank test. Similar data analysis procedures were performed to calculate the incidence density rates of AD (per 1000 person-years, PY) and HRs (95% CIs) for the AF and non-AF cohorts in the study 2. Data analyses were conducted using statistical package SAS (Version 9.4, SAS Institute Inc., Carey, North Carolina). A two-tailed P value < .05 was considered statistically significant.

| Study 1
The study 1 consisted of 11 813 patients in the AD cohort and 11 813 controls in the non-AD cohort (Table 1). Men represented the majority of the study cohorts (71.4% vs 67.9%) and over a half of study population were more than 65 years old. The AD cohort were slightly younger than the non-AD cohort. The average followup duration was 3.71 ± 3.19 years for the AD cohort and 4.85 ± 2.99 years for the non-AD cohort. Figure 1A shows that the cumulative incidence of AF was higher in the AD cohort than in the non-AD cohort (the log-rank test P < .001) after 12 years of follow-up.

| Study 2
The study 2 consisted of 190 494 patients in the AF cohort and 190 494 controls in the non-AF cohort (Table 3). Both cohorts had more men (54.9% vs 55.3%) and more than 75% of the study population were aged ≥ 65 years. The average follow-up duration was 3.47 years for the AF cohort and 4.19 years for the non-AF cohort. Figure 1B shows that the cumulative incidence of AD was higher in the AF cohort than in the non-AF cohort (the log-rank test P = .01) after 12 years of follow-up. The AF cohort had 1.18-fold higher incidence of AD than the non-AF cohort (0.55 vs 0.47 per 1000 person-years), with an aHR of 1.24 (95% CI = 1.07-1.44) ( Table 4). The sex-specific AD risk for the AF cohort relative to the non-AF cohort was significantly higher for women (aHR = 1.37; 95% CI = 1.08-1.75 AD and AF to date, this is indeed data that could be helpful in the understanding and management of the growing population of adults with AD and AF. The topic of secondary AF has received increasing attention as prior anecdotal beliefs that AF resolved after resolution of acute illness triggers have yielded to evidence suggesting high AF recurrence, T A B L E 4 Incidence and hazard ratios of aortic dissection for atrial fibrillation (AF) cohort compared to non-AF cohort by demographic characteristics, comorbidity and follow-up year Patients with any comorbidity of hypertension, diabetes mellitus, hyperlipidemia, CHD, heart failure, COPD, PAD, CKD, hyperthyroidism, sleep disorders, gout, cerebrovascular disease, chronic liver disease, cancer, asthma, peptic ulcer disease, and VHD were defined as the comorbidity group. b Model was adjusted for age, and comorbidities of hypertension, diabetes mellitus, and heart failure.
*P < .05; **P < .01; ***P < .001. morbidity, and mortality after secondary AF. 14 Patients with AD complicated AF might have an increased risk of AF-associated adverse events resulting in premature mortality. [10][11][12] The two pathologies, AD and AF, are essentially different diseases. It is possible that before the development of AF in AD, there may be several factors that are also considered to be a cause of AF incidence. However, we adopted a propensity score-matching analysis and multivariate adjustment to minimize these biases and the results were statistically true, 13 and we observed that such association was stronger in those without comorbidity, implying that the development of AF in patients with AD might be independent of comorbidities.
Identification of AD is of importance in patients with AF because of high risk of death from AD. 1,2 The current study showed that the risk ratio was highest among women, old age, and short follow-up times; implying that more attention should be paid to these populations. The potential for higher incidence of AD in patients with AF, although has been corrected for covariates not adequately balanced in Table 3, it could be the case that AF and AD are two potential manifestation of possible variables not considered in the present study, without any pathophysiological relation.
AD, which might involve coronary injury, pericardial involvement, and other direct cardiac effects, could increase the risk of AF. [3][4][5][6][7][8][9] However, there is no argument made in support of an association between AD and subsequent AF although several molecular mechanisms involving the atrial remodeling and weakness of the aortic walls might be possible explanations. [15][16][17][18] Such findings based on this big dataset deserved further investigation.

| LIMITATIONS
First, the Taiwanese NHIRD has the power of large numbers, but it does not provide additional physiological insight. Second, diagnoses were retrieved from only inpatient files. This might introduce a bias, as AF patients with concomitant disorders might more often be hospitalized. Third, information about treatment was not collected in this study and this might have influenced the occurrence of both diseases and represents a possible bias in the interpretation of the results.
Fourth, although we have used propensity matching and then conducted a multi-variable analysis, it should be mentioned that uncontrolled potential confounders could be an issue in this type of study.
Finally, the diagnostic accuracy of the diseases using ICD codes might be potentially the major limitation of the present study. However, this nationwide database has been validated and high accuracy was guaranteed. [19][20][21][22] 6 | CONCLUSION This is a study evaluating the relationship between the presence of AD and the incidence of AF and vice versa in a large numbers of patients from Taiwan. A positive association for both was found in this study.