Atrial fibrillation (AF) is frequently observed in the medical intensive care unit (MICU) 1, with up to about 15% of MICU patients showing periods of AF 234. AF directly leads to loss of the atrial kick and, as a consequence, reduces ventricular loading. Especially if the ventricular compliance is decreased, as is the case in sepsis and many other medical conditions, this reduction results in decreased cardiac performance. By performance, we mean the capacity to meet pressure and volume requirements. The irregular and mostly rapid ventricular response also shortens the ventricular filling time, and thereby shortens the preload. AF therefore reduces cardiac performance. The reduction is more serious in patients with pre-existing cardiac dysfunction due to decreased ventricular compliance. A persistent high ventricular rate may lead to tachycardia-mediated cardiomyopathy 5. Conversion to sinus rhythm (SR) improves ventricular function in patients with heart failure 6. These findings urge most intensivists to treat AF.

Most intensivists may have adopted an AF treatment modality based on their individual experience combined with extrapolation of the treatment of other, mostly unrelated, but well-defined and well-established, patient groups. In most cases this means that, after correction of assumed or perpetuating factors, treatment directly aimed at the rhythm disorder itself will be started. To date, treatment of AF in the MICU cannot be supported by sufficient evidence from the literature. Notwithstanding the large number of patients involved, thorough research in this field is scarce 7. There are important reasons to believe that MICU patients are different from other patients with AF and therefore require a more tailored therapy. Fundamental questions that remain unanswered for MICU patients are summarised in Table 1.

To find answers for these questions we searched for direct clinical evidence and – when not available – searched for evidence from related areas. Direct evidence will be considered all results derived from randomised controlled trials or well-conducted epidemiological studies in MICU patients. The aim of the present paper is to improve insight, to explore future research goals and to define an optimal treatment mode based on current knowledge for the population admitted in MICU. We will describe the evidence found per question presented in Table 1 according to the patient group from which it is derived. Each section will start with MICU patients, followed by mixed intensive care unit (ICU) patients, surgical ICU patients and cardiothoracic surgery ICU patients, and will end with the least related patient category – outpatients.

We conducted a computer literature search in the databases of MEDLINE, EMBASE and the Cochrane Library, from 1966 to 2007, combining the following key words: 'intensive care' or 'critical care' or 'critically ill' and 'atrial fibrillation' or 'atrial tachyarrhythmia' and 'treatment' or 'aetiology' or 'risk factors'. Reference lists of all selected articles were reviewed to identify other relevant articles. For relevant articles the search was extended in PubMed with the 'related articles' search function. PubMed was checked for other publications by authors of key papers. Web of Science^® was checked for papers citing key papers. All selected articles were reviewed by two different reviewers.

AF is a supraventricular tachyarrhythmia characterised by uncoordinated atrial activation with subsequent deterioration of atrial mechanical function. On the electrocardiogram, AF is described by the replacement of consistent P waves with rapid oscillations or fibrillatory waves that vary in size, shape and timing, associated with an irregular, frequently rapid, ventricular response when atrioventricular conduction is intact 8. Recurrent means at least two episodes of AF. Paroxysmal means self-terminating, and persistent means that self-termination is absent and that electrical or pharmacological conversion is necessary to end AF 9. MICU patients are patients admitted to the ICU not for surgical or cardiological reasons.

There are no data on MICU patients specifically, nor data for surgical ICU patients. There are, however, risk factors identified in these patient categories. Risk factors due to causality can in general not be distinguished from epiphenomena. Risk factors can at least suggest a certain pathophysiology, however, and therefore they may help in the identification of a patient population. Independent risk factors for AF are age, disease severity, hypertension, hypoxia, previous AF, congestive heart failure, chronic obstructive pulmonary disease, chest trauma, shock, a pulmonary artery catheter, previous use of calcium-channel blockers, low serum magnesium, withdrawal of β-blocker or angiotensin-converting enzyme-inhibitor and withdrawal of catecholamine use 1011121314151617.

In patients after noncardiac surgery, the right atrial pressure rather than fluid overload or right heart enlargement seems to be correlated with AF 14181920. Cardiothoracic surgical (CTS) patients with AF, however, tend to have a more positive fluid balance 2122. Interestingly, systemic inflammatory response syndrome and sepsis are also independent risk factors 1014. A proinflammatory state, as measured by leucocytosis or monocyte activation, is associated with AF, although the mechanism is not clear 232425. AF is sometimes the first sign of sepsis 4. A genetic predisposition for an increased inflammatory response is associated with an increased incidence of postoperative AF 26. Catecholamines influence the susceptibility for AF 1027. Hypovolaemia is also a risk factor 28.

Most knowledge about AF is gained from studies in noncritically ill patients. AF is probably the final common pathway of structural changes in combination with a trigger leading to abnormal activation patterns in the atria 8. Structural changes can be multiple; for example, fibrosis and amyloidosis. Structural changes increase with age, which might be the explanation for the fact that age is the most important risk factor for AF. There are numerous triggers that can lead to AF when combined with a substrate and a perpetuating factor. Ischaemia, and local (pericarditis or myocarditis) and generalised inflammation can affect the atria 2930. Hypovolaemia and hypervolaemia or a sudden increase in afterload, as in pulmonary embolism, and mitral or tricuspid valve dysfunction are examples of increased atrial workload that can cause AF. Nervous (both sympathic and parasympathic) tone, hormonal changes, electrolyte disturbances and also the preload and the afterload influence excitability and conduction in the atria and atrio–ventricular junction 27. The cumulative effect of structural changes and one or more of these triggers and perpetuating factors will determine whether AF will occur and will persist 831.

AF did not influence mortality significantly in a mixed medical–cardiac ICU 2. In a general ICU population, however, patients with AF appeared to have a significantly higher mortality compared with patients without AF 3. Furthermore, surgical patients with new-onset AF have a significantly higher disease severity and higher ICU mortality 4113233. A persistent elevated increased heart rate, frequently due to AF, is associated with increased mortality 34. In a large, retrospective, cohort study in cardiac surgery patients, AF was not an independent predictor for inhospital mortality 35. Patients outside the ICU setting with AF have increased overall mortality and mortality of cardiovascular causes 3637.

AF did increase the length of stay (LOS) in a mixed medical–cardiac ICU 2. Onset of AF in a patient in the surgical ICU increases their LOS in the ICU and in the hospital 111632333940. Onset of AF reduces the systolic blood pressure 4142, and also decreases oxygen saturation and increases the pulmonary artery wedge pressure. An increased heart rate is associated with increased morbidity 34.

A number of symptoms in noncritically ill patients have been described 8. Most relevant for ICU patients is the decreased cardiac output, which is caused by the loss of coordinated atrial contraction, by irregularity of ventricular contraction 43, by inadequate filling time for the left ventricle due to tachycardia, and by tachycardiomyopathy 844. Tachycardiomyopathy can occur as soon as 24 hours after the start of AF 44.

Although advocated in the early days of intensive care, there is no evidence that digoxin or any other antiarrhythmic drug can prevent AF in critically ill patients 4145. There are no trials investigating prevention of AF in MICU patients.

In surgical ICU patients, and especially in CTS patients, there are trials and guidelines evaluating preventive measures 4647. Although prophylactic digoxin, verapamil and β-blockers all decrease the heart rate in cases of postoperative AF, only β-blockers decrease the incidence of postoperative AF as shown in a meta-analysis 48. In CTS patients, β-blockers can reduce AF by 75% 12.

In randomised controlled trials, amiodarone prevented AF in patients undergoing CTS, and also reduced the hospital LOS and the ICU LOS 49505152535455. There is no consensus, however, about the clinical relevance of this finding since data are conflicting 5657. Amiodarone, for example, was found to increase the ICU LOS and the need for vasoactive medication or other haemodynamic support in some studies 1358. More recent meta-analyses show that amiodarone prevents AF but the influence on the LOS or the mortality is not yet unequivocally established 5960.

Magnesium and atrial pacing cannot prevent AF in CTS patients, as shown in several randomised controlled trials 13526162. In a comparative trial, however, magnesium could prevent AF equally as effectively as sotalol; both drugs combined had a synergistic effect 63. Amiodarone and magnesium are also synergistic 64, but synergism could not be shown for propranolol and magnesium 65. Recent meta-analyses show that magnesium can prevent AF but without any effect on the LOS or on the mortality 6667. Cholesterol synthesis inhibitors and corticosteroids also are preventive, perhaps by interaction with inflammatory pathways 686970.

Studies on prevention have extensively been reviewed recently 15596071. Guidelines advise the prophylactic use of β-blocker or amiodarone for elective CTS patients 15465960. Generalisation of prevention studies in CTS patients to MICU patients is unproven.

There are no randomised placebo-controlled trials in MICU patients aimed at treating AF once it has occurred. There are, however, comparative trials between drugs that are supposed to be effective. Procainamide and amiodarone are equally effective; after 12 hours, 70% of the patients were in SR 72. Magnesium, when compared with amiodarone, has been found to be more effective in restoration of SR, while the two treatments are equally effective in rate control 73. Ibutilide, a relatively new class III agent, can restore SR in 70% of patients that fail rhythm control with amiodarone treatment 74. Ibutilide can restore SR – with 80% conversion to SR in haemodynamically unstable patients without unmanageable proarrhythmic side effects 75.

In the CTS population, 80% of patients with AF convert to SR within 24 hours. The use of β-blockers before the start of AF and the absence of diabetes and left ventricular hypertrophy were independent predictors of conversion to SR 76. In a retrospective study of surgical patients with new-onset supraventricular tachycardia (93% with AF), 75% had SR within 48 hours after the start of continuous infusion of amiodarone 77. In a mixed population with severe left ventricular dysfunction, amiodarone had no apparent negative effect on haemodynamics 78. When compared with amiodarone, propafenone gives earlier conversion to SR but the ultimate conversion percentage was equal after CTS 79. Ibutilide showed a dose-dependent conversion rate in a randomised controlled trial 80. Ibutilide and amiodarone have an equal conversion rate to SR and an equivalent time to conversion, but amiodarone causes more hypotension – probably due to vasodilatation 8182. Direct-current cardioversion has a low rate of conversion to SR in postsurgical new-onset AF 108384.

Treatment of AF in CTS patients has been the topic of several reviews and guidelines 8586. The studies in these patients are sufficiently powered to detect effectiveness for their primary end point, prophylaxis or treatment of AF, but are underpowered to detect differences in mortality or adverse effects due to the low incidence of these events.

There are also studies in mixed ICU populations. Diltiazem and amiodarone appeared equally effective in achieving rate control; however, discontinuation of the study drug because of hypotension occurred more often in the diltiazem group 87. Ibutilide is effective for rapid conversion, but with potentially life-threatening proarrhythmic side effects 88. Magnesium is more effective in rate control and probably in conversion than diltiazem in a mixed population with longstanding AF paroxysms 52. With digoxin treatment, no rate control or rhythm control can be reached in a mixed ICU population 2841. The success rate of electric cardioversion is also low in this population 2841.

The management of AF in noncritically ill patients has been studied and reviewed extensively 8990. New-onset AF has a high spontaneous conversion rate of 64–90% within 24 hours 91. Treatment with digoxin has been replaced by treatment with β-blockers and calcium-channel blockers because better rate control can be achieved with these latter drugs. Especially in seriously ill patients, digoxin fails to achieve an adequate reduction of the ventricular rate 92. Class I and class III antiarrhythmic drugs are effective in conversion of AF in recent-onset AF, especially when combined with verapamil 899093. Amiodarone is also an effective drug because high-dose oral or intravenous amiodarone has a higher conversion ratio to SR than placebo 9194959697. A meta-analysis showed that class IA, class IC and class III antiarrhythmic agents are equally effective in obtaining SR 98. Meta-analyses comparing amiodarone with class IC antiarrythmic drugs or placebo showed that treatment was equally effective, although conversion was earlier in class IC treatment 9699. None of the drugs was associated with an increased or a decreased mortality 98.

Depending on the AF duration, amiodarone is highly effective in conversion with no more adverse effects than other drugs 100. In patients with severe congestive heart failure, amiodarone controls the heart rate immediately 101102. Magnesium is safe, reliable and cost-effective compared with diltiazem 52. Ibutilide is a safe and effective drug in persistent AF 103. Angiotensin-converting enzyme-inhibitors might be effective in preventing structural changes (for example, fibrosis) and might therefore enhance outcome in AF patients, even in patients with worse underlying heart disease 104. Glucocorticoid therapy reduces the proinflammatory state as measured by C-reactive protein and probably, as a consequence, the incidence of AF 105.

Electrical cardioversion in noncritically ill patients is effective but has a high relapse rate 8. The timing of treatment is important because applying electric cardioversion too early leads to an increased recurrence of AF 106. Whether the findings in noncritically ill patients are relevant for MICU patients is uncertain, but this evidence gives us a direction for research in mechanisms and therapy.

Pharmacokinetics and pharmacodynamics are changed in ICU patients 107. Multiple drug use may cause drug interactions 107. These factors might render ICU patients more prone to side effects 107108. There are limited data, however, for MICU patients. Amiodarone-induced pulmonary toxicity has been described in postmortem MICU patients suffering from acute respiratory distress syndrome 109110. In surgical ICU patients, amiodarone induces hypotension after intravenous loading 8182. Severe hepatoxicity due to amiodarone has been described 111. Ventricular tachycardia occurred in CTS patients 80.

In non-ICU patients admitted for AF there is a high incidence of adverse events, mainly cardiac, from antiarrhythmic drugs 112. On the other hand, the incidence of amiodarone-induced proarrhythmic effects is low 113114115. Nevertheless amiodarone remains a drug with many side effects. Amiodarone pulmonary toxicity, especially in the previously damaged lung, is a hazardous adverse effect 108110116. The occurrence is probably cumulative, dose dependent and duration dependent, but adverse pulmonary effects can also be seen within 3 days after the start of administration 110114115. Drug interactions might be more frequent for amiodarone but have not extensively been studied 117. The implications for the ICU patient of the effect of amiodarone on thyroid gland function, which is a major problem in outpatients, are not yet clear 118119. Amiodarone has a complex pharmacokinetic and pharmacodynamic profile 120.

There are few data on the effect of treatment of AF on mortality in ICU patients. A meta-analysis in non-ICU patients showed that class IA, class IC and class III antiarrhythmic agents are equally effective in reaching SR. No impact, however, on the quality of life or the mortality could be found 98. β-Blockers improve survival in patients with heart failure and AF 121. Amiodarone treatment in patients with AF and congestive heart failure improved conversion to SR and survival 122.

There are no data on MICU patients. In a retrospective study in surgical patients with new-onset supraventricular tachycardia (93% with AF), continuous infusion of amiodarone did not lead to significant differences in haemodynamics in responders compared with nonresponders 77123. Another retrospective study in a selected population of critically ill patients showed that amiodarone improved haemodynamic parameters compared with pretreatment values 42.

In a mixed ICU patient population, conversion to SR did not increase the systolic blood pressure 73. Most patients are already haemodynamically unstable before AF, and the contribution of AF is uncertain 124.

There are no data in MICU patients. In a pilot trial in CTS patients there was no difference in the LOS or rhythm at discharge between rate control and rhythm control strategies 125126. After cardiac surgery in haemodynamically stable patients, rate control is preferred over rhythm control because almost all patients convert spontaneously within 6 weeks after surgery 1286125127.

Five randomised-controlled trials in non-ICU patients did not show a beneficial effect of rhythm control over rate control in haemodynamically stable patients 128129. These studies have been described in three meta-analyses; rate control showed less adverse events and less hospitalisations 989130. These meta-analyses, however, do not sufficiently cover specific patient groups 124.

There are no data on stroke incidence in the MICU. Short-term postoperative AF is a risk factor for stroke in CTS patients 131. Postoperative AF doubles the risk compared with patients without AF, despite the use of aspirin 2232131132.

AF is an independent risk factor for stroke in non-ICU patients 133. In patients with AF, an inflammatory response is an independent risk for stroke 134. The prothrombotic state due to inflammation is probably more important than the presence of AF 25. An increased C-reactive protein level is a risk factor for thromboembolism in patients with AF 135.

Since there are no data on stroke incidence in MICU patients, there are also no data on prevention.

In elderly patients undergoing cardiothoracic surgery receiving preventive treatment with amiodarone in addition to β-blocker, the incidence of AF and stroke was significantly reduced but the mortality was not changed 53. This effect was also shown in a meta-analysis 55.

The stroke incidence in non-ICU patients can be reduced with anticoagulation. The bleeding risk is outweighed by the advantage of a reduced stroke incidence in most patients 8. There is no difference between rate control and rhythm control in stroke incidence when the patient is on anticoagulation treatment 130. Treatment of the proinflammatory state can reduce the incidence of stroke.

Although AF is a frequent symptom associated with a high mortality in critically ill patients, there are still many lacunae in our knowledge. We evaluated the actual level of knowledge with the purpose to reach a treatment protocol based on best available evidence. There is no literature, however, presenting the criteria of evidence-based medicine. Even the questions of whether AF is the cause of mortality or just an epiphenomenon 3 and of whether treatment improves outcome are still not answered. A treatment protocol therefore has to be based on extrapolation of results from studies performed in other patient groups. But even in these patient groups, there is still a lot of debate about the optimal treatment protocol 137.

Because the beneficial effect of treatment is not certain, any protocol should at least not add serious adverse events; first, do no harm. Doing as little as possible is a defendable credo. This means optimising the fluid balance, correcting electrolyte disturbances, reducing sympaticus tonus and avoiding proarrhythmic drugs. Reduction of the systemic inflammatory state is tempting but is of course always the purpose of ICU treatment. The evidence for the use of steroids for this indication is insufficient. When the ventricular rate is arbitrarily judged acceptable and there is little haemodynamic compromise, no further action is probably required. If this condition is not met, we have to seek the balance between benefit and harm.

Direct-current cardioversion is not useful because of the high relapse rate. In some situations, however, judged to be desperate, direct-current cardioversion will be performed. It has also not been proven that electrical cardioversion does not damage a heart already involved in the multiorgan failure of critical illness. Although the effectiveness of magnesium has been questioned there are no reports on adverse effects. In nonacutely threatened patients, therefore, an attempt to achieve rate control and even rhythm control with intravenous infusion of magnesium is worthwhile. If further treatment is deemed necessary, a choice has to be made between various antiarrhythmic drugs. Class IA, class IC and class III antiarrhythmic drugs all are effective but are also proarrhythmic. Calcium-channel blockers are less effective and have the disadvantage of causing hypotension. Intensivists may have an emotional barrier to using β-blockers in patients also receiving vasopressors and inotropes, but β-blockers could be a rational choice. The choice made by most intensivists, however, is for amiodarone: this drug is effective, although not as fast acting as some other drugs. The acute adverse effects seem to be very limited, but the adverse effects in the long term might be a problem. We therefore advocate short-term use of amiodarone if treatment is deemed necessary.

A protocol concerning AF should also have a statement about stroke prevention. There are, however, no data to support such a statement. We have no data on the stroke incidence of medical ICU patients with AF. Owing to the proinflammatory state, this incidence is probably higher than in other patients with AF. On the other hand, there is also an increased, but unquantified, bleeding risk. Risks and benefits of anticoagulation can therefore not be weighed in general. This balance has to be estimated for individual patients, and an educated guess has to be made 136.

A rational treatment protocol could therefore consist of several steps. First, treatment of predisposing factors is necessary. Second, a short attempt at magnesium supplementation can be done. Third, amiodarone can be administered for a short-term period. Most patients will by then have an acceptable rate or rhythm; however, if the patient does not, ibutilide (a class III drug) can serve as rescue treatment.

We have treated 29 patients in a MICU with this protocol. Ninety per cent of the patients had SR after 24 hours and all patients had an acceptable heart rate. We did not need to use ibutilide treatment, nor direct-current cardioversion. The inhospital mortality in this patient group, however, was still 38% 138.

Having a protocol with a reasonable success rate does not release us from doing further research. The high mortality rate could be caused by the fact that AF is just an epiphenomenon in critically ill patients. The possibility that the mortality is in part caused by insufficient treatment of AF or, on the contrary, is caused by adverse effects of the treatment, however, is too realistic to be ignored. All we have stated about the treatment of AF in MICU patients is based on extrapolation and is therefore just a hypothesis. We should therefore explore the possibility of randomised controlled trials against placebo. These trials should be based on a better understanding of AF in critically ill patients.

AF = atrial fibrillation; CTS = cardiothoracic surgery; ICU = intensive care unit; LOS = length of stay; MICU = medical intensive care unit; SR = sinus rhythm.

The authors declare that they have no competing interests.