Diabetes and atrial fibrillation in hospitalized patients in the United States

Abstract Background Data on the burden of atrial fibrillation (AF) associated with diabetes among hospitalized patients are scarce. We assessed the AF‐related hospitalizations trends in patients with diabetes, and compared AF outcomes in patients with diabetes to those without diabetes. Hypothesis AF‐related health outcomes differ between patient with diabetes and without diabetes. Methods Using the National Inpatient Sample (NIS) 2004–2014, we studied trends in AF hospitalization rate among diabetic patients, and compared in‐hospital case fatality rate, length of stay (LOS), cost and utilization of rhythm control therapies, and 30‐day readmission rate between patients with and without diabetes. Logistic or Cox regression models were used to assess the differences in AF outcomes by diabetes status. Results Over the study period, there were 4 325 522 AF‐related hospitalizations, of which 1 075 770 (24.9%) had a diagnosis of diabetes. There was a temporal increase in AF hospitalization rate among diabetic patients (10.4 to 14.4 per 1000 hospitalizations among patients with diabetes; +4.4% yearly change, p‐trend < .0001). Among AF patients, those with diabetes had a lower in‐hospital mortality (adjusted odds ratio [aOR]: 0.68; 95% CI: 0.65–0.72) and LOS (aOR: 0.95; 95% CI: 0.94–0.96), but no difference in costs (aOR: 0.95; 95% CI: 0.94–0.96) and a higher 30‐day rate of readmissions compared with no diabetes (aHR 1.05; 95% CI: 1.01–1.08), compared to individuals without diabetes. Conclusion AF and diabetes coexist among hospitalized patients, with rising trends over the last decade. Diabetes is associated with lower rates in‐hospital adverse AF outcomes, but a higher 30‐day readmission risk.


Atrial fibrillation (AF) is the most common arrhythmia in the United
States, [1][2][3] with AF cases projected to increase to 15.9 million by the year 2050. 1,2 The US burden of diabetes has also been growing, 4 with a projected 30% or more increase by 2050. 5 Diabetes increases the risk of AF by 40%. 6 Diabetes and AF often coexist, posing unique clinical challenge. 7 There is a paucity of US national data on the patterns and time trends of coexisting AF and diabetes. Insights into the burden of coexisting diabetes and AF may help in improving outcomes, as diabetes may adversely influence AF outcomes. 8 The link between diabetes and AF also has potential implications for diabetes screening among those with AF patients and vice versa, and a potential therapeutic implication. Emerging trial data suggests that sodium glucose cotransporter (SGLT)-2 inhibitors may have beneficial effects on AF outcomes among individuals with diabetes. 9 Using the Nationwide Inpatient Sample (NIS) registry data, we investigated: (1) the trends in AF hospitalization rate, in-hospital case fatality rate, average length of stay (LOS), cost and utilization of rhythm control therapies among patients with diabetes; and (2) the potential differences in in-hospital mortality, LOS, cost, utilization of rhythm control therapies, and 30-day readmission rates between hospitalized AF patients with and without diabetes.

| Data sources and study population
Our study was based on data from the National Inpatient Sample Briefly, the NIS is a 20% stratified sample of inpatient hospitalization from participating states. It is representative of >95% of all inpatient hospitalizations in the United States. 10 Prior to 2012, the NIS comprised all hospitalizations from a 20% stratified sample of community hospitals (defined as all non-federal hospitals with the exception of rehabilitation facilities). In 2012, the NIS underwent a redesign and is now comprised of a 20% stratified systemic sample of discharges from all community hospitals from participating states. 11 This design has been shown to decrease SE estimates while calculating variances at the national level and make these estimates more generalizable to the target universe. To facilitate multi-year trend analysis and ensure comparability of estimates, the AHRQ has published trend weights using the 2012 methodology on previous years. 12 The details of the NIS sampling methodology are published annually by the AHRQ. Due to the de-identified nature of this data, the study was exempt from Institutional Review Board approval from the primary institution.
To assess readmissions among patients with a primary diagnosis of AF discharged between January and November 2014, we used the 2014 Nationwide Readmissions Database (NRD), developed by the AHRQ as part of the HCUP. The NRD is the largest all-payer database of hospital readmissions in the United States. In order to support readmission analyses, the NRD contains a linkage variable (NRD_visitlink) by means of which patients can be tracked across hospitalizations. The target universe for the NRD includes inpatient discharges from community hospitals that were not rehabilitation or long-term acute care facilities. The sampling frame for the NRD includes inpatient discharges from participating states, excluding rehabilitation and LTAC facilities. All discharges from the sampling frames are included. Discharge weights for the NRD were developed with the target universe as the standard. We included hospitalized patients aged ≥18 years in both the trends and readmissions analyses.

| Atrial fibrillation and diabetes ascertainment
A primary diagnosis of AF was based on the International Classification of Diseases, Ninth Revision (ICD-9) code 427.31. Diabetes was ascertained using the Clinical Classification Software (CCS) codes 49 and 50. Prior studies have shown that the use of ICD-9-CM coding has a high degree of accuracy for identifying atrial fibrillation 13,14 and diabetes mellitus, 15 with a robust sensitivity and an excellent sensitivity and specificity (both >90% for each condition). The complete list of codes using in the study are summarized in the Table S1.
Since the NIS and Nationwide Readmissions Database (NRD) are both administrative databases, we are able to identify a diagnosis of diabetes only by the use of ICD-9 CM diagnosis and procedure codes. There are over 50 ICD-9 CM codes for diabetes with or without complications (250.xx). The CCS coding system is developed by the AHRQ and maps similar ICD-9 CM codes to a manageable number of CCS codes (https://www.hcup-us.ahrq.gov/ toolssoftware/ccs/ccsfactsheet.jsp). We chose to use CCS codes for defining diabetes mellitus because using 2 CCS codes rather than >50 ICD-9 CM codes makes our analysis less prone to errors. The CCS coding system is designed to capture all of the disparate ICD-9 codes that are used to identify diabetes in administrative data.

| Outcomes
The study outcomes included hospitalization rate per 1000 hospitalizations related to diabetes, in-hospital outcomes (in-hospital case fatality rate, LOS, cost, utilization or rhythm control therapies), and 30-day all-cause readmission.
The rhythm control therapies included radiofrequency catheter ablation and electrical cardioversion, identified using relevant ICD-9 and CCS procedure codes (Table S1). Since codes for ablation and cardioversion are not specific to AF and could be used for other arrhythmias, we excluded records with a diagnosis of other atrial or ventricular arrhythmias as having had AF ablation or electrical cardioversion. We also excluded records with implantation of permanent pacemaker during the same hospitalization because the ablation codes may represent a rate-controlled strategy with atrioventricular (AV) nodal ablation and pacemaker implantation.

| Statistical analysis
We assessed the yearly prevalence of AF and estimates of the length of hospital stay and costs, and the related time trends, among patients with diabetes based on the year of admission in all patients, across four age groups, (18-45 years, 45-64 years, 65-74 years and ≥75 years) and by sex. Survey analysis techniques were used to produce national estimates as recommended by the AHRQ. Continuous variables were expressed as mean (SE) while categorical variables were expressed as percentages. Differences in continuous and categorical variables between the diabetes and no-diabetes groups were tested using linear and logistic regression models.
Linear trends in continuous as well as categorical variables were examined using linear regression with year as a continuous predictor. We examined overall and subgroup specific trends in AF hospitalization rate per thousand diabetes-related hospitalizations, in-hospital case fatality rate, LOS, cost and utilization of rhythm control therapies. The costs were estimated by multiplying total charges with hospital cost-to-charge ratios supplied by the AHRQ. Costs were adjusted to 2014 US dollars using consumer price index inflation adjustment calculator from the Bureau of Labor Statistics. 16 The differences in in-hospital AF outcomes between the diabetes and no-diabetes groups of patients were assessed using logistic regression models. The latter included the following adjustment variables: age, sex, income quartile, Charlson comorbidity index, hypertension, chronic renal failure, congestive heart failure, obesity, peripheral vascular disease, deficiency anemia, chronic lung disease, pulmonary circulation disorders, coagulopathy, rhythm control procedure (catheter ablation and electrical cardioversion), hospital region, hospital location and teaching status.
The differences in all-cause 30-day readmission among AF patients with and without diabetes were examined using a Kaplan-Meier curve and the log-rank test. A multivariable adjusted Cox proportional hazards model was also used with adjustment for age, sex, income quartile, Charlson comorbidity index, hypertension, chronic renal failure, congestive heart failure, obesity, peripheral vascular disease, deficiency anemia, chronic lung disease, pulmonary circulation disorders, coagulopathy, rhythm control procedure (catheter ablation and electrical cardioversion), hospital region, hospital location and teaching status.
A p-value of ≤.05 was considered significant. All analyses were conducted using Stata 15.1 statistical package (Statacorp, College Station, TX).

| Trends in atrial fibrillation hospitalization in patients with diabetes mellitus
The weighted total of hospitalizations with a primary diagnosis of AF during the study period was 4 325 522, of which 1 075 770 (24.9%) had a diagnosis of diabetes. Mean age of patients with primary diagnosis of AF in the setting of diabetes was 70.4 years (SE 0.04) and 51.8% were women. There was a declining trend in the proportion of females (53%-50.5%, p-trend < .001). There was an overall increase in comorbidity burden ( Table 2). In multivariable analyses, the improvement in case fatality rate was not fully explained by changing trends in age, sex and comorbidities (Table S2).
We observed improvements in LOS (mean of 4.0 days in 2004 to 3.7 days in 2014, p-trend < .001). This trend was significant across all but the youngest age subgroup (Table 2). Similarly, the average cost declined from $9810 in 2004 to $9125 in 2014 (p-trend < .001); and this trend was significant across all subgroups ( Table 2).
The utilization of radiofrequency ablation and electrical cardioversion increased linearly in both the diabetes and no-diabetes groups of patients (p-trend < .001, Figures S1 and S2).

| Outcomes of atrial fibrillation in patients with and without diabetes
The differences in baseline characteristics by diabetes status for patients admitted with a primary diagnosis of AF are summarized in In time-to-event analysis ( Figure S3 and Table 3

| 30-day readmissions after an atrial fibrillation hospitalization in patients with and without diabetes
We observed significant differences in causes for readmission between those with diabetes compared to those without diabetes. Of the patients who were readmitted, a greater proportion of patients with diabetes had a primary diagnosis of congestive heart failure or sepsis compared with patients without diabetes (All p < .01), while a lower proportion of patients with diabetes had a primary diagnosis of cardiac dysrhythmias compared with patients without diabetes ( Figure S4).

| DISCUSSION
This study reports data on diabetes and AF comorbidities over a  AF. 8,[21][22][23] Our study showed that diabetes is inversely related to inhospital mortality, length of hospital stay and costs among patients with AF, but is independently associated with readmissions. Prior studies have suggested that diabetes among patients with AF is associated with an increased risk of both mortality and readmission, 24,25 Indeed, in the ORBIT-AF registry, diabetes was found to be significantly associated with all-cause hospitalizations. 24 It is important to indicate that in contrast to our investigation that examined in-hospital outcomes (including mortality) between patients with and without diabetes, the prior studies mainly examined longer term mortality. 8,[21][22][23] The inverse associations of diabetes with in-hospital mortality, length of stay and costs, as may be partially explained by an overall higher rate of use of thromboembolic events prevention among the admitted  29 and diabetes duration are determinants of the risk of stroke and thromboembolic events among patients with AF. 30,31 Fifth, given that we examined in-hospital and short-term outcomes, we may have not captured the full extent of the influence of diabetes on the outcomes of AF.
In conclusion, at least one in ten hospitalized patients with diabetes has atrial fibrillation, these proportion have been increasing with time. Approximately one quarter of AF patients have diabetes. In terms of atrial fibrillation outcomes, compared to those without diabetes, patients with diabetes tend to experience a shorter hospital stay and lower in-hospital mortality, but a higher rate of readmission.

ACKNOWLEDGMENTS
This work was supported by a specific funding source. The authors have nothing to disclose.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the Agency of Health Care Research and Quality. Restrictions apply to the availability of these data, which were used under license for this study.
Data are available at https://www.hcup-us.ahrq.gov/nisoverview.jsp with the permission of the Agency of Health Care Research and Quality.