Statement of principal findings
The evidence base available for NOAF management in critically ill patients identified by our scoping review was very limited. Many studies lacked a comparator group and many of those that had a comparator group did not control for confounding factors. Only two randomised trials were identified, both of which were small and inconclusive. We identified significant heterogeneity in the definitions of both NOAF and successful treatment, making a synthesis of results difficult.
The limited evidence from our scoping review suggested that beta-blockers might be more effective than amiodarone for the conversion to sinus rhythm and mortality outcomes; however, residual bias in previous studies may explain these assertions. Whether or not and when to use anticoagulation is unknown. No conclusive findings were reported owing to the low quality of the reviewed evidence. Only one clinician survey was found, with a very low response rate.16
Our analysis of the RISK-II database demonstrated that patients who developed NOAF in an ICU were older and had higher levels of comorbidity than those who did not. Even after controlling for these differences, patients with NOAF still had substantially higher mortality in hospital and during the first 90 days after discharge. Patients who developed NOAF in an ICU also had higher rates of subsequent hospitalisation with AF, stroke and heart failure than those who did not.
Our analysis of the detailed within-ICU MIMIC-III and PICRAM databases showed that NOAF is a common problem that occurs in 6–11% of eligible ICU patients, depending on the data source. In an ICU, NOAF is associated with a significant increase in heart rate and a significant decrease in blood pressure, despite an increase in vasoactive medication doses. Supporting our RISK-II findings, we identified a significantly increased hospital mortality rate associated with NOAF, even after adjusting for other factors that are predictive of mortality.
These findings highlight the importance of identifying optimal treatment strategies for NOAF in patients in an ICU. We found that the treatment of NOAF with digoxin or calcium channel blockers as first-line therapy is associated with poorer rate control and rhythm control, respectively. Prior to adjustment for confounding variables, we found that beta-blocker therapy was associated with improved hospital survival. We also demonstrated that patients who received beta-blockers were less unwell at admission and more stable after AF onset than those who received amiodarone. After comprehensive adjustment of these factors, there were no identifiable differences in any outcome between beta-blocker therapy and amiodarone.
Scoping review
The evidence base available for NOAF management in critically ill patients was very limited. A key problem with the studies identified in this scoping review was that many (n = 12) were single-group studies (i.e. lacking a comparator group). Of the 25 primary studies included in the review, only two were RCTs30,31 and only three of the non-randomised comparative studies27–29 attempted to control for confounding factors, which may have affected outcomes. For all of these studies, which used more robust approaches, there were still serious concerns about how bias (arising from their designs and/or analyses) might affect their results. Although the two RCTs30,31 did not find statistically significant differences in conversion rates between the treatments studied, each contained < 60 patients. Many studies27–30 concluded that more research is needed.
Heterogeneity in the treatment dose (e.g. total doses ranging from < 1 g to 8 g for amiodarone8,31,32,34,35,42,43) and administration (e.g. bolus or continuous infusion), and varying time points to assess conversion to sinus rhythm (e.g. within 2 hours,30 4 hours,31 12 hours30 and 24 hours8,32,35,36) were observed across studies. Comparing studies was, therefore, challenging. There is a need to establish optimal treatment dosing and administration regimens, as well as validated definitions of treatment success. Six studies33,34,36,37,40,48 (14%) were available only as conference abstracts. Only limited data could be extracted for these studies, making their results more difficult to interpret.
Studies varied in how they reported and defined NOAF. Different heart rate thresholds for NOAF were used.31,36,38,43,47,49 Studies also reported different time periods for which NOAF must be sustained27,30,41,43,47,49 and for which instances AF would be considered NOAF.8,28,29,38,39,50 Ten studies32–35,37,40,44–46,48 did not provide any definition for NOAF.
The evidence from this review34–36 suggests that beta-blockers might be more effective than amiodarone for the conversion back to normal sinus rhythm, with better outcomes for mortality reported in those who received beta-blockers than in those who received amiodarone.28,35 A recent UK-wide survey16 found that amiodarone is the most commonly used pharmacological treatment for NOAF in UK ICUs, which suggests that these studies are not changing current practice. Calcium channel blockers appeared to be less effective than beta-blockers and amiodarone for conversion to sinus rhythm.30,31,36 Two studies27,48 reported that hydrocortisone might be effective as a prophylactic treatment. However, a larger comparative study27 reported slightly higher (but not statistically significant) mortality associated with the use of hydrocortisone. All of the studies reporting effects of hydrocortisone had significant methodological limitations.
The evidence base for NOAF treatment strategies was reported as very limited in two systematic reviews, which agrees with our findings.53,56 Both systematic reviews were not able to report any conclusive findings, citing the low quality of the reviewed evidence53 and methodological differences between the included studies.56 In agreement with our findings, the review by Kanji et al.56 concluded that a standardised outcome measure of success is needed as varying time points used to assess conversion to sinus rhythm limits recommendations on treatment efficacy.56 Both systematic reviews emphasised the urgent need for further research studies.53,56
The current literature29,62 suggests that it is unclear if the benefits of administering anticoagulants in critically ill patients with NOAF for stroke prevention outweigh the increased risk of bleeding. Two review articles61,64 proposed a patient-centred approach to administer anticoagulants only in patients with high risk of arterial thromboembolic events. Outside the ICU, withholding anticoagulation for a short time in the perioperative period in patients with AF undergoing elective surgery was not associated with an increased risk of arterial thromboembolism and decreased the risk of major bleeding.99 It is, however, not known how these findings translate to critically ill patients in ICUs.
The risk assessment scores for subsequent thromboembolism following the development of NOAF have generally not been developed or validated in the ICU population. The CHA2DS2-VASc score100 has, however, once been shown to be predictive of thromboembolic event risk in the ICU setting, although with poor sensitivity and predictive value.101 Bleeding risk scores to guide anticoagulation decisions in NOAF have been developed using population-based cohorts of patients with AF in the community. Common risk factors used in these scores include older age, renal dysfunction, hypertension and a history of bleeding. Comparative studies of these tools report varying results depending on the patient population.102–105 Patients who develop NOAF during sepsis are at a higher risk of in-hospital and post-discharge stroke and death than those who do not, despite adjusting for confounders.14,106 Those with NOAF are at a higher risk of thromboembolic complications than patients with pre-existing AF.106 However, the individual risk of stroke and thromboembolic events in patients who develop NOAF during an ICU admission remains poorly understood.
Although community-based bleeding risk scores are chiefly composed of chronic comorbidities, the bleeding risk in patients in an ICU is likely to be more related to acute factors, such as illness severity, systemic inflammation, type and location of surgery, nutritional status, invasive devices, and acute coagulopathy and thrombocytopenia.70,107–109 Bleeding risk in critically ill patients with NOAF is higher than patients in the community, with one study demonstrating that 9% of patients who received systemic anticoagulation had significant bleeding warranting cessation of anticoagulation and at least one blood transfusion.8
A recent UK-wide survey suggested that around one-third of clinicians initiate anticoagulation in patients with NOAF during an ICU admission.16 Once in stable AF in the community, anticoagulation can be recommended for almost all patients with AF. However, the balance of risks in patients either in an ICU or in whom AF was demonstrated only during the ICU admission is likely to be more complex and dynamic. Modified risk scores that incorporate such complexities are, therefore, required for critically ill patients with NOAF.
The RISK-II database analysis
This analysis of the RISK-II database provides clear evidence of a patient group developing NOAF in critical care with substantially worse short- and long-term outcomes than patients without any record of AF during or prior to ICU admission. We adopted a restrictive approach to ensure that patients’ CMP and HES records provided sufficient confidence in discriminating NOAF in the ICU from prior and subsequent AF. In our sensitivity analysis, we adopted a slightly broader, yet still conservative, definition of NOAF. The outcomes, although somewhat less severe, remained consistent with the primary analysis. Outcomes were substantially worse for patients with AF than for patients with no AF, even after controlling for patient characteristics and past medical history. Patients developing NOAF during an ICU admission were less likely to survive and more likely to be readmitted with AF, stroke or heart failure than those who did not develop NOAF. Comparison with other data sources suggests that our methodology identified a subset of all patients who develop NOAF in critical care. Given the inconsistent nature with which diagnoses are recorded in HES,110 the group identified as developing NOAF in this analysis might best be interpreted as representing patients for whom their AF was of sufficient clinical importance to be documented in their clinical notes (which are then used to code diagnoses in HES records).
The evidence provided by multivariable regression indicates that the impact of NOAF on mortality is not constant over time, but rather focused on the period in and immediately after discharge from hospital. Patients who developed NOAF in an ICU had increased mortality up to the limit of follow-up. However, from 90 days after hospital discharge, the increased mortality appeared to be entirely explained by their older age, sex and comorbidities. During hospital admission and the first 90 days after discharge, roughly half of the increased mortality appeared attributable to either NOAF or unobserved clinical factors that are associated with it. These findings emphasise the importance of developing strategies for both the treatment and the anticoagulation management of NOAF in the period of critical illness and its immediate aftermath, for which current evidence is lacking.
We did not break down multivariable models for subsequent hospitalisation into similar time periods, given the complexity of interpreting proportional hazards regressions with both non-proportional hazards (effects that vary over time) and competing risks. However, given that hospitalisation is closely related to mortality, it may be reasonable to assume that a similar pattern of varying effects over time would be observed.
The MIMIC-III and PICRAM database analysis
To the best of our knowledge, this is the first study to quantify the haemodynamic changes associated with NOAF onset in ICU patients accounting for patients’ pre-AF parameters at scale. We found that NOAF during an ICU admission is associated with a significant increase in heart rate and a significant decrease in blood pressure, despite an increase in vasoactive medication doses. Organised atrial activity contributes to ventricular filling, cardiac output and the closure of the atrioventricular valves.6 New-onset atrial fibrillation precludes these mechanisms and the effects of the loss of atrial contraction on ventricular filling may be compounded by the diastolic dysfunction commonly seen during critical illness.111 These mechanisms may explain why NOAF is temporally associated with a reduction in cardiac index in non-ICU patients with chronic heart failure.7 Our findings are consistent with one previous study that found that haemodynamic instability developed in 37% of post-surgical ICU patients after NOAF.8
Atrial contractile dysfunction occurs after brief episodes of AF112 and can last for several weeks after achieving rhythm control.113 Any AF may, therefore, have considerable impact in any critically ill patient with minimal physiological reserve. Indeed, episodes of NOAF lasting for ≥ 30 minutes in the ICU are independently associated with increased hospital mortality.11 We found that NOAF is associated with hospital mortality after adjustment for confounding variables, regardless of whether NOAF is diagnosed based on continuous monitoring (ICU databases analysis) or from hospital diagnosis codes (RISK-II database analysis). In both cases, some, but not all, of this association is explained by confounding variables. The associated mortality risk was higher in our RISK-II analysis than in our in-ICU analysis. This may be explained by the differing definitions, with NOAF in the RISK-II data probably representing AF significant enough to result in a HES code versus any AF in the in-ICU data.
Together, our findings highlight the importance of optimal treatment and follow-up of patients who develop NOAF during an ICU stay. We found that the use of digoxin is associated with lower rates of achieving heart rate control than the use of amiodarone. Digoxin may be selected to reduce heart rates without inducing hypotension; however, digoxin may be less effective during states of increased sympathetic drive,114 including critical illness. One small study of patients in the ICU with AF with rapid ventricular response (not exclusive to NOAF) demonstrated that digoxin was less effective at rate control in patients receiving catecholaminergic medication.115
We found that the use of calcium channel blockers was associated with lower rates of achieving rhythm control than amiodarone. These findings are supported by a small, randomised study of paroxysmal AF outside the ICU,116 which reported a cardioversion proportion of 0% in patients who received verapamil compared with 77% in patients who received amiodarone. Our findings contrast with one RCT identified in our scoping review, which compared amiodarone with calcium channel blockers31 and found no difference in achieving rhythm control. However, this study included only 20 patients per treatment group, making it difficult to draw any conclusions. We did not identify any significant difference in hospital survival between any of the treatments when compared with amiodarone.
In the MIMIC-III database, the unadjusted analysis demonstrated an apparent mortality benefit in the beta-blocker treatment group. However, patients in the beta-blocker group were younger and less unwell at presentation than those in the amiodarone group. After developing NOAF, patients in the beta-blocker group had higher blood pressures, were less likely to be on vasoactive medications and had lower inflammatory markers. After our comprehensive adjustment, the difference in mortality was no longer evident. Our adjusted results conflict with one large study,28 which suggested a survival benefit with beta-blockers over amiodarone therapy in patients with sepsis. However, this study was unable to adjust for features around NOAF onset. Our study demonstrates that important differences exist between treatment groups after the onset of AF, which may influence treatment choice. Failure to adjust for these factors is likely to result in residual confounding.
Strengths and limitations
The scoping review was performed using systematic, transparent and robust methods. The bibliographic database searches were comprehensive, which allowed maximal identification of relevant studies while also minimising the possibility of publication or language biases affecting the review. We carried out the screening and data extraction processes in duplicate to reduce the risk of reviewer errors or biases affecting the review. The main limitation of the scoping review was the methodological shortcomings of the studies identified.
Our RISK-II database analysis allowed us to include > 4000 patients who developed NOAF and 27,000 matched comparators from ICUs across England, with long-term follow-up using routinely collected data. However, the analysis is limited by the sensitivity of diagnostic records. Along with the limitations in defining patients with NOAF, some outcomes also need careful interpretation. For example, ‘hospitalisation with stroke’ may miss both extremes of severity, in which mild strokes and transient ischaemic events may not result in hospital admission and catastrophic strokes may result in death without admission.
Our analysis of two within-ICU databases has several strengths. First, we carried out comprehensive adjustment that included variables around the onset of AF. We show significant differences in these peri-AF variables that have not been adjusted for in previous studies. Second, our analysis of granular health-care data allowed a detailed analysis of the haemodynamic changes associated with NOAF. Third, our analysis of routinely collected data over many years provided a sample size large enough to demonstrate differential efficacy in NOAF treatments. This analysis also has limitations. The study was retrospective in nature and the development of NOAF was not independently verified. Documentation of AF in the MIMIC-III database has, however, been shown to be accurate for determining AF onset to within 1 hour after independent review of a sample of electrocardiographic waveforms.89 Documentation of comorbidities in the MIMIC-III database relied on hospital billing codes, which may not have identified all cases. Furthermore, although good balance of numerous confounding covariates was achieved prior to assessing treatment efficacy, we are unable to exclude bias introduced by residual unmeasured confounding. Unmeasured variables, such as echocardiographic parameters, may have contributed to the association between NOAF in ICU and outcome, and would not be represented in the propensity weights. Finally, we acknowledge the difference in case mixes between UK and USA data. Overall, patients in the MIMIC-III database were younger and had lower mortality than those in the PICRAM database, which may explain part of the difference in NOAF incidence. Identifying patients who developed NOAF reduced these differences by identifying sufficiently unwell patients in both databases. Owing to the underlying differences, we primarily analysed each database in isolation, using the combined analysis to support these primary findings.
Uncertainties
Either amiodarone or beta-blockers are commonly used in critically ill patients to control AF, but there is little evidence to support whether or not one is superior. Purported beneficial effects of beta-blocker therapy may be because of residual confounding in some studies.
In patients who develop NOAF while in an ICU, it is not clear in whom anticoagulation following hospital discharge might be beneficial.
The incidence of AF and/or left ventricular dysfunction at hospital discharge and at 3 months following the development of NOAF while in an ICU is unknown. However, readmission with heart failure and thromboembolism is increased over the 5 years following an episode of NOAF while in an ICU, particularly in the first year.
It remains unclear to what extent NOAF in patients in an ICU is causally related to worse outcomes. Evidence for causality may be supported by future randomised prevention trials, in which a reduction in AF burden is associated with better outcomes, or through the application of robust causal inference methods in observational studies.