New-onset atrial fibrillation and associated outcomes and resource use among critically ill adults—a multicenter retrospective cohort study

Background New-onset atrial fibrillation (NOAF) is commonly encountered in critically ill adults. Evidence evaluating the association between NOAF and patient-important outcomes in this population is conflicting. Furthermore, little is known regarding the association between NOAF and resource use or hospital costs. Methods Retrospective analysis (2011–2016) of a prospectively collected registry from two Canadian hospitals of consecutive ICU patients aged ≥ 18 years. We excluded patients with a known history of AF prior to hospital admission. Any occurrence of atrial fibrillation (AF) was prospectively recorded by bedside nurses. The primary outcome was hospital mortality, and we used multivariable logistic regression to adjust for confounders. We used a generalized linear model to evaluate contributors to total cost. Results We included 15,014 patients, and 1541 (10.3%) had NOAF during their ICU admission. While NOAF was not associated with increased odds of hospital death among the entire cohort (adjusted odds ratio [aOR] 1.02 [95% confidence interval [CI] 0.97–1.08]), an interaction was noted between NOAF and sepsis, and the presence of both was associated with higher odds of hospital mortality (aOR 1.28 [95% CI 1.09–1.36]) than either alone. Patients with NOAF had higher total costs (cost ratio [CR] 1.09 [95% CI 1.02–1.20]). Among patients with NOAF, treatment with a rhythm-control strategy was associated with higher costs (CR 1.24 [95% CI 1.07–1.40]). Conclusions While NOAF was not associated with death or requiring discharge to long-term care among critically ill patients, it was associated with increased length of stay in ICU and increased total costs.


Introduction
Atrial fibrillation (AF) is the most common cardiac dysrhythmia, with a lifetime risk of 1 in 4 among older adults [1]. Development of AF has been associated with stroke, myocardial infarction, heart failure, and death [2]. In the intensive care unit (ICU), patients often present with pre-existing AF; however, some ICU patients may develop new-onset AF (NOAF) in the context of critical illness [3]. Unlike AF seen in non-critically ill patients, NOAF is often thought to be a consequence of critical illness pathophysiology and treatment, including inflammation, electrolyte disturbances, or proarrhythmic medications, namely vasopressors and inotropes [4]. Incidence of NOAF in the general ICU varies markedly; however, most studies suggest that 10-15% of patients will develop this complication during their ICU stay [3][4][5].
The clinical importance of critical illness-associated NOAF is a matter of ongoing uncertainty [4]. AF in and of itself may contribute to clinical decompensation through hemodynamic compromise [6]. Alternatively, NOAF may simply represent a marker of increased illness severity, and may identify patients at increased risk of death without acting causally to worsen prognosis. Some cohort studies have identified an independent association between NOAF and mortality [7][8][9], while others have not [5,10]. Whether this relationship is seen among all critically ill patients, or limited to select subgroups, is unknown. Furthermore, the factors associated with death among patients with NOAF are unknown.
Equally important is the relationship between NOAF, and subsequent ICU resource use and costs [11,12]. Since the ICU is a major source of hospital expenditure, considerable effort has been dedicated to understand the contributors to cost, in order to drive policy and optimize resource utilization [13]. With regard to NOAF, little is known regarding the degree of impact on cost expenditures. We primarily sought to evaluate the association between NOAF and outcomes, resource utilization, and costs among critically ill adult patients. Given the prognostic importance of NOAF among patients with sepsis [14,15], we secondarily aimed to evaluate the association between incidence of NOAF and associated outcomes and resource utilization among critically ill patients with suspected infection, sepsis, and septic shock.

Materials and methods
We obtained ethics approval for this study from the Ottawa Health Science Network Research Ethics Board (protocol 20160570-01H).

Study design, setting, and subjects
We studied ICU patients from two hospitals within the Ottawa Hospital network (Ottawa, ON). These hospitals have approximately 2500 combined ICU admissions per year. These are combined medical and non-cardiac surgical ICUs. We retrospectively examined prospectively collected data from the Ottawa Hospital Data Warehouse, a health administrative database used in previous studies [16][17][18][19]. From hospital admission, data is gathered daily from each patient and stored in the Data Warehouse. Data quality assessments are executed routinely, and quality-assurance initiatives are conducted regularly to ensure completeness and accuracy.
We included all consecutive patients ≥ 18 years of age, admitted to one of the two ICUs between January 2011 and December 2016. Sample size was determined pragmatically, on the basis of available patients in the Data Warehouse. We also examined pre-specified subgroups of patients, including those with suspected infection, sepsis, and septic shock, as based on the Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3) [20][21][22]. "Suspected infection" was defined as concomitant administration of oral or parenteral antibiotics, and sampling of body-fluid cultures, as performed previously [16,23], and in keeping with the Sepsis-3 definitions [20]. "Sepsis" was defined as suspected infection and an increase in the Sequential Organ Failure Assessment (SOFA) score by greater than 2 points [20,21]. Finally, "septic shock" was defined by sepsis in addition to initiation of vasopressors or a serum lactate ≥ 2.0 mmol/L [20,22].

Data collection
We obtained all data from the Ottawa Hospital Data Warehouse. We abstracted demographic data, comorbidities, Elixhauser Comorbidity Score [24], and Multiple Organ Dysfunction Score (MODS) [25] at the time of ICU admission. The Elixhauser Comorbidity Score is generated from comorbidities stored in the Data Warehouse, and the association between this index and hospital mortality has been previously validated in our database [26]. The "most responsible diagnosis" was recorded at death or discharge, based upon International Classification of Diseases, Version 10 (ICD-10, July 2015). We also noted whether there was presence of a "No cardiopulmonary resuscitation (CPR)" directive at the time of ICU admission. We collected outcome data from admission until either the point of discharge from hospital or hospital death.
We determined patient costs using the case-costing system of the Ottawa Hospital Data Warehouse, as done previously [17,23,27]. Total hospital costs include both direct and indirect sources. Direct costs refer to all hospital expenses with fee codes linked to the patient identifier. Indirect costs refer to any overhead operational fees associated with provided service. The Ottawa Hospital uses a standardized case-costing methodology, developed by the Ontario Case Costing Initiative, and based upon the Canadian Institute for Health Information Management guidelines [28]. Costs were indexed to 2018 Canadian Dollars using consumer price indices [23,27].

Outcomes
The primary outcome was hospital mortality. Secondary outcomes included discharge directly from hospital to long-term care (among survivors to hospital discharge originally from home), hospital readmission within 30 days of hospital discharge among survivors, ICU length of stay (LOS), hospital LOS, resource utilization (including invasive and non-invasive mechanical ventilation, and renal replacement therapy), and total hospital costs.

Screening for atrial fibrillation
For each patient, the occurrence of any AF was prospectively recorded by bedside nurses for the purposes of quality assurance. The date and time of AF, as captured by the bedside nurse, was stored in the Data Warehouse. Patients identified through this method were then evaluated by a single investigator (SMF), to confirm the diagnosis. Since there is no consensus definition for NOAF [4], we followed pre-existing definitions from the literature [7,8]. NOAF was defined as either (1) AF ≥ 1 h in duration, as noted by bedside telemetry (routinely evaluated in charts where electrocardiograms were not completed); (2) AF < 1 h in duration, but captured on electrocardiogram; or (3) AF initiating pharmacologic therapy or electrical cardioversion. All bedside ECGs, along with final interpretation by an attending cardiologist are stored in patient records. "Sustained" AF was defined as failure to convert to sinus rhythm 24 h following the onset of any pharmacological treatment or electrical cardioversion. We excluded patients with a previously documented or known history of AF, as determined at the time of hospital admission and stored in the Data Warehouse.

Statistical analysis
We performed all statistical analyses using R (version 3.3.3) and IBM SPSS (version 24.0). We present data as mean values, with standard deviation (SD), or medians, with interquartile range (IQR), where appropriate. Student's t test (parametric values), Mann-Whitney test (non-parametric values), and χ 2 (for categorical values) were performed to determine between-group differences. In keeping with existing guidelines, we did not perform pairwise comparisons of baseline characteristics [29]. To adjust for measured confounders in the association between new-onset AF and outcomes of interest, we followed recommendations for observational studies in the critically ill [30]. As per these recommendations, confounders were determined a priori, on the basis of their likelihood of influencing both the presence of NOAF and mortality and not acting as mediators or colliders in the association between AF and mortality, as based upon existing clinical knowledge evaluating the association between NOAF and mortality in critically ill patients [3,4]. In accordance with these recommendations [30], we used multivariable logistic regression modeling to adjust for important continuous (age, MODS at ICU admission, Elixhauser comorbidity index) and categorical (sex, individual comorbidities, "No-CPR" directive on admission, location prior to ICU admission, and most responsible diagnosis) variables. As recommended, variables on the causal pathway and potentially contributing to NOAF (e.g., vasoactive medications) were not included [30]. We evaluated for possible synergy between NOAF and sepsis through the use of an interaction term in the primary model, as performed previously [23]. If a statistically significant interaction term was found between NOAF and sepsis, we then represented this with a four-level categorization. We assessed variation in total hospital costs using a multivariable generalized linear model with gamma distribution and log link [31,32]. We present adjusted odds ratios (aORs) and cost ratios (CRs) with 95% confidence intervals. A P value of ≤ 0.05 was considered statistically significant.

Results
A total of 17,173 patients were admitted to the participating ICUs from 2011 to 2016 (Fig. 1). Of these, 2105 patients (12.3%) had a known or documented history of AF prior to ICU admission, and were excluded. A further 54 patients (0.4%) were excluded because of missing outcome data.
We included 15,014 patients in the analyses. Of these patients, 1541 (10.3%) had NOAF while in the ICU. Baseline characteristics of patient with and without NOAF are shown in Table 1. NOAF patients were older (mean age 64.7 years vs. 58.5 years), had higher severity of illness (mean MODS 5.3 vs. 4.2), and higher comorbidity burden.
Comparisons of resources used between patients with and without AF are shown in Table 4. No differences were seen in the use of invasive (54.7% in those with NOAF vs. 52.8% in those without NOAF, P = 0.16) or non-invasive ventilation (16.6% in those with NOAF vs. 17.5% in those without NOAF, P = 0.38). Vasoactive medication use was higher among patients with NOAF (64.3% vs. 61.2%, P = 0.02). In terms of treatment strategy for NOAF, 747 (48.5%) patients received antiarrhythmic medical therapy (i.e., amiodarone, procainamide, or flecainide), while 644 (41.8%) received therapy with a beta-blocker, calcium channel blocker, or digoxin. A total of 128 patients (8.3%) received a combination of the above therapies.
Finally, comparisons of patient costs between patients with and without NOAF are shown in Table 5 Table S6). Among patients with NOAF (Additional file 7: Table S7), significant predictors of total hospital costs include total hospital or ICU LOS, use of invasive mechanical ventilation or renal replacement therapy, and use of antiarrhythmic medical therapy (as compared to beta-blocker, calcium channel blocker, or digoxin treatment).

Discussion
We identified NOAF in 10.3% of critically ill adults, in keeping with known prevalence rates [4]. We found no association between NOAF and increased hospital  Patients with NOAF had prolonged ICU and hospital LOS, and NOAF was a predictor of increased total costs. Mechanical ventilation, renal replacement therapy, and use of antiarrhythmic therapy were significant predictors of total cost in NOAF patients. Taken together, our study identifies important novel associations between    NOAF and outcomes among critically ill patients, and also describes the economic impact of NOAF. Hospital mortality among critically ill adults is high, and therefore, identification of prognostic factors associated with increased risk can be helpful to clinicians in escalating or tailoring therapy. Identification of these factors may also be helpful in discussions with patients and families regarding goals of care. NOAF is often thought to be a marker of illness severity [3]; however, the evidence examining the association between NOAF and hospital mortality remains uncertain [4]. In our large cohort of critically ill adults, we did not find such an association. However, evaluation of subgroups of patients with suspected infection, sepsis, and septic shock did find an independent association between NOAF and hospital mortality potentially suggesting that the consequences of NOAF in the patient with sepsis may differ from other populations.
Important physiologic changes occur during sepsis that make the atrial substrate vulnerable to NOAF [33], and patients with sepsis have a nearly sixfold risk of developing AF, as compared to other populations [14]. Our findings identifying an independent association between NOAF and mortality from sepsis are supportive of existing literature, and potentially suggest that survival in this population may be improved through the prevention and treatment of NOAF [34]. Importantly, unlike previous studies, we defined sepsis and septic shock using the most recent Sepsis-3 definitions, indicating that NOAF has potential implications in these populations, and demonstrate an interaction between the presence of NOAF and sepsis. In keeping with the Sepsis-3 focus of sepsis as infection with concomitant organ dysfunction, NOAF may indeed represent sepsis-defining cardiac dysfunction [35]. As such, NOAF during sepsis might mediate mortality and may not simply be a marker of illness severity, as is seen in other disease processes [36].
If NOAF does represent a marker of illness severity, then its presence may be considered an important indicator of deterioration among critically ill patients. Therefore, identification of factors associated with mortality among ICU patients who develop NOAF remains an important area of ongoing research [4]. Unsurprisingly, a history of heart failure was associated with hospital mortality among patients with NOAF in our cohort, suggesting that these patients may be more susceptible to the hemodynamic effects of NOAF. Our results also found that sustained AF following 24 h of treatment was also associated with higher hospital mortality. Therefore, sustained AF may represent a particular ominous marker of illness severity among critically ill adults, and clinicians should act cautiously in those patients with NOAF and persisting AF.
We additionally evaluated the association between NOAF and hospital resource use and cost. Identification of factors associated with such outcomes remains an important focus in critical care research [11]. Few differences were found with regard to resource utilization between patients with and without NOAF. However, patients with NOAF did have prolonged ICU and hospital LOS, which translated into higher costs. The presence of NOAF was also a significant predictor of costs among our cohort, and this was independent of other important factors more commonly associated with cost, including renal replacement therapy and mechanical ventilation [37]. This prolonged LOS was also manifested in increased laboratory, pharmacy, and nursing costs. Despite higher costs, patients with NOAF had higher unadjusted mortality, which translated into significant differences in cost per survivor (a proxy indicator of cost-effectiveness).
Among patients with NOAF, treatment with antiarrhythmic medication (as compared to beta-blockers, calcium channel blockers, or digoxin) was associated with higher cost, but decrease in mortality. While this study was not designed to test the efficacy of therapeutic agents for NOAF, the little evidence that exists on this topic seems to suggest no beneficial effect of any particular agent [3,4]. This is in keeping with our results. We did not find evidence of difference in mortality associated with antiarrhythmic agents, but did note increased cost. This is likely attributable to the expense of these drugs [38], and the fact that they often cannot be administered outside of a monitored setting and therefore may prolong ICU LOS. However, it must be stressed that the efficacy of these agents can only be appropriately tested in a randomized trial, and while equipoise exists, decisions related to agents for treatment of NOAF must be made on a case-by-case basis.
We used a large multicenter database of patients with prospective identification of NOAF, and provide novel data related to hospital mortality, resource utilization, and cost. We also closely followed recommendations for control of confounding in observational studies [30]. However, our study has important limitations. Most importantly, our database lacks the granularity to investigate other potential factors that may influence outcome from NOAF, such as underlying cardiac function, pulmonary artery catheterization, rate of AF, timing of AF (particularly in relation to onset of critical illness), use of anticoagulation, and incidence of other arrhythmias. Some outcome data, such as incidence of stroke, were not available. We additionally excluded patients with existing AF on the basis of known or documented AF, but it is possible that patients may have had pre-existing AF and not known, due to absence of symptoms [39]. Overall, this lack of granularity represents an important limitation in our study. Second, patients were included on the basis of prospective identification by nursing staff. While similar methods have been used at other institutions [40], it is possible that cases of NOAF were missed using this methodology, particularly if AF was brief and transiently