Temporal trends and factors associated with increased mortality among atrial fibrillation weekend hospitalizations: an insight from National Inpatient Sample 2005–2014

Objective Atrial fibrillation (AF) weekend hospitalizations were reported to have poor outcomes compared to weekday hospitalizations. The relatively poor outcomes on the weekends are usually referred to as ‘weekend effect’. We aim to understand trends and outcomes among weekend AF hospitalizations. The primary purpose of this study is to evaluate the trends for weekend AF hospitalizations using Nationwide Inpatient Sample 2005–2014. Hospitalizations with AF as the primary diagnosis, in-hospital mortality, length of stay, co-morbidities and cardioversion procedures have been identified using the international classification of diseases 9 codes. Results Since 2005, the weekend AF hospitalizations increased by 27% (72,216 in 2005 to 92,220 in 2014), mortality decreased by 29% (1.32% in 2005 to 0.94% in 2014), increase in urban teaching hospitalizations by 72% (33.32% in 2005 to 57.64% in 2014), twofold increase in depression and a threefold increase in the prevalence of renal failure were noted over the period of 10 years. After adjusting for significant covariates, weekend hospitalizations were observed to have higher odds of in-hospital mortality OR 1.17 (95% CI 1.108–1.235, P < 0.0001). Weekend AF hospitalizations appear to be associated with higher in-hospital mortality. Opportunities to improve care in weekend AF hospitalizations need to be explored.


Introduction
Atrial fibrillation (AF), the most common sustained arrhythmia in clinical practice had an estimated worldwide prevalence of 33.5 million in 2010 [1]. AF weekend hospitalizations were previously reported to have higher mortality and lower rates of cardioversion [2]. Subsequent studies in this population have demonstrated improved mortality and rates of cardioversion [3,4]. To date, there has been no temporal trend analysis showing this effect. We sought to investigate the outcomes in the years 2005-2014 through a publicly available national inpatient sample database (NIS).

Methods
A description of NIS database has been elaborated in prior studies [5][6][7]. The NIS is one of the largest, allpayer database for the United States in-patient hospitalizations, and it is maintained by the Agency for Health Care Quality and Research (AHRQ). The NIS includes a 20% random sample of all inpatient hospitalizations from 46 states in the United States. Each observation represents a hospitalization with one primary diagnosis, up to NIS hospitalizations have 2 sampling strategies. Before 2012, all hospitalizations were from a random sample of 20% of acute care hospitals in the United States, stratified by bed size, region, and location. Starting in 2012, the NIS included a random sample of 20% of discharges from all acute care hospitals in the United States; this effort reduced the margin of error by 50%, and national estimates decreased by 4.3%. From 1998 to 2011, discharge weights are provided by the AHRQ after a validation process, and they are used to calculate national estimates. To account for changes in the sampling strategies, the variable "trend weights" have been used for 2011 and all preceding years to facilitate trend analysis from 1998 to 2014 as recommended by AHRQ [8].

BMC Research Notes
The study was exempted by the University of Iowa, Iowa City, institutional review board as it includes only de-identified, publicly available data. For our analysis, we only used NIS data from 2005 to 2014. Similar to previous studies, we used the ICD-9-CM code 427.31 to identify hospitalizations involving hospitalizations with principal diagnosis (dx1) of AF [9]. The variables for hospitalization demographics were provided in the dataset (example: age, gender, length of stay). The weekend hospitalizations (Saturday-Sunday) were identified using ' AWEEKEND' variable. Hospitalizations with anticoagulation were identified using the ICD-9-CM code 'V58.61' . ICD-9-procedure codes 9961, 9962, 9969 and 3734 were used to identify hospitalizations with cardioversion/ ablation.
We used survey analysis methods to account for the clustering and stratification of encounters for all continuous and categorical variables. SAS 9.4 (SAS Institute Inc., Cary, North Carolina) software were used to perform statistical analysis. We used sampling weights to estimate trends and national estimates to account for the change in sampling design as recommended by the AHRQ. For the demographics, co-morbid diseases, and weekend hospitalizations within each year were compared using Student's t test for continuous variables and the Chi square test for categorical variables. Multivariate logistic regression method was used in SAS (proc surveylogistic) to evaluate the association between weekend hospitalizations and in-hospital mortality after including the other variables for potential confounders. C-statistic was used for goodness of the model fit for a binary outcome. Like previous studies, trends in demographics, co-morbid diseases, weekend hospitalizations involving AF hospitalizations, length of hospitalization, in-hospital mortality were evaluated using the survey logistic models after creating dummy variables for each outcome of interest. A P-value < 0.05 was considered statistically significant. The checklist provided by NIS was used to ensure the appropriateness of data analysis as recommended by AHRQ [10].
Univariate and multivariate logistic regression analyses were performed. In multivariate analysis, the weekend hospitalizations were associated with higher odds of in-hospital mortality (OR 1.170, 95% CI 1.108-1.125, P < 0.0001) ( Table 2). Apart from the weekend admission status, acute respiratory failure, congestive heart failure, renal failure and urban hospital admission (teaching and non-teaching) were found to be strong predictors of inhospital mortality.

Discussion
The main findings and trends noted in the current study of weekend AF hospitalizations are (1) improving trends in in-hospital mortality over 10 years from 2005 to 2014.
(2) Weekend hospitalizations are associated with higher odds of in-hospital mortality. (3) Decreasing the mean length of hospital stay, and (4) increasing trends of utilization rates of cardioversion and anticoagulation.
The 'weekend effect' is a concern where the patients are thought to have worse outcomes when admitted to the hospital on a Saturday or a Sunday [11]. The first reports of the weekend hospitalizations having higher mortality appeared in the 1970s. Higher mortality and longer hospital LOS have been reported among AF hospitalizations  [3,12] Subsequent study reported no difference in weekend and weekday AF in-hospital mortality [13]. In comparison to the prior studies, ours is the first study analyzing the trends of weekend AF hospitalizations. Our results match the results of Weeda et al. [3] where there is improved mortality among weekend hospitalization with AF. Though the lower utilization of cardioversion has been demonstrated through the years, the rates of cardioversion have significantly been improving, and at the same time, the in-hospital mortality has been decreasing during the same time period. This can be attributed to improved access to life-saving procedures. However, the overall utilization rates of cardioversion continue to be low among the weekend hospitalizations when compared to the weekday hospitalization. This is likely due to staffing issues, the availability of anesthesia, or coverage for a trans-esophageal echocardiogram at some institutions.
In the nationwide US practice, the weekend AF hospitalizations appear to have improved rates of in-hospital mortality, rates of cardioversion utilization and improved utilization of anticoagulation. However, the overall rates of in-hospital mortality continue to be poor in comparison to weekday hospitalizations. Further studies are required to identify the opportunities to improve AF weekend care.

Limitations
Although our study has a large nationally representative database sample, these findings should be interpreted considering the following limitations. First, we identified our cases using ICD-9 discharge diagnosis codes, and details of the initial presentation (for example, emergency room visit) are not available, thereby, limiting the ability to confirm the diagnosis. Secondly, the NIS data does not provide information on important clinical predictors of outcomes such as the duration and the type of AF, left atrial diameter, the presence of thrombus in the left atrium and the baseline functional status, which can potentially influence the outcomes for in-hospital mortality. Third, given the description of ICD-9 codes in the database, it is not possible to differentiate pre-existing comorbidities from complications which have occurred during the hospitalization. Fourth, data regarding specific