Risk factors for new-onset atrial fibrillation in patients with sepsis: a systematic review and meta-analysis

Abstract

Objective:

Atrial fibrillation (AF) frequently develops in patients with sepsis and is associated with increased morbidity and mortality. Unfortunately, risk factors for new-onset AF in sepsis have not been clearly elucidated. Clarification of the risk factors for AF during sepsis may improve our understanding of the mechanisms of arrhythmia development and help guide clinical practice.

Data Sources:

Medline, Embase, Web of Science, and Cochrane CENTRAL

Study Selection:

We conducted a systematic review and meta-analysis to identify risk factors for new-onset AF during sepsis.

Data Extraction:

We extracted the adjusted odds ratio for each risk factor associated with new-onset AF during sepsis. For risk factors present in more than one study, we calculated pooled odds ratios (meta-analysis). We classified risk factors according to type and quantified the factor effect sizes. We then compared sepsis-associated AF risk factors with risk factors for community-associated AF.

Data Synthesis:

44 factors were examined as possible risk factors for new-onset AF in sepsis, 18 of which were included in meta-analyses. Risk factors for new-onset AF included demographic factors, comorbid conditions, and most strongly, sepsis-related factors. Sepsis-related factors with a greater than 50% change in odds of new-onset AF included corticosteroid use, right heart catheterization, fungal infection, vasopressor use, and a mean arterial pressure target of 8–85mmHg. Several cardiovascular conditions that are known risk factors for community-associated AF were not identified as risk factors for new-onset AF in sepsis.

Conclusions:

Our study shows that risk factors for new-onset AF during sepsis are mainly factors that are associated with the acute-sepsis event and are not synonymous with risk factors for community-associated AF. Our results provide targets for future studies focused on AF prevention and have implications for several key areas in the management of patients with sepsis such as glucocorticoid administration, vasopressor selection, and blood pressure targets.

Introduction

Atrial fibrillation (AF) frequently complicates the course of critically-ill patients(1), and is the most common arrhythmia in patients with sepsis(2). New-onset AF that occurs in the setting of sepsis is frequently associated with acute hemodynamic decompensation(3), increased intensive care unit (ICU) length of stay(4) and long-term stroke and mortality risks(5). Although new-onset AF occurs in up to 25% of patients with sepsis(4), the risk factors for new-onset AF during sepsis are not currently well defined.

Improved understanding of risk factors for new-onset AF during sepsis can provide mechanistic insights and guide clinical practice. For example, Surviving Sepsis Guidelines recognize the importance of avoiding AF during sepsis by recommending hemodynamic targets and vasopressor strategies that limit arrhythmia risk(6). However, some recommendations – such as the suggestion to use “dopamine as an alternative vasopressor agent to norepinephrine only in selected patients (e.g., patients with low risk of tachyarrhythmias)” – may be difficult to interpret without an understanding of risk factors for tachyarrhythmias during sepsis. Although one prior study found that risk factors for AF during sepsis did not appear to reflect AF risks in the community setting (7), a comprehensive synthesis of risk factors for AF during sepsis may better address knowledge gaps and inform application of sepsis guideline recommendations. Thus, we conducted a systematic review and meta-analysis to identify risk factors for new-onset AF in adult patients with sepsis.

Materials and Methods

Details of the pre-registered protocol for this systematic review and meta-analysis are available on PROSPERO(8).

Study selection

We searched Medline (1950-present), Embase (1947-present), Web of Science (1900-present) and the Cochrane CENTRAL database for studies that identified risk factors for new-onset AF or new-onset atrial flutter in adult (18 years or older) patients with sepsis. Atrial flutter often co-exists with AF during sepsis and thus was included to improve search sensitivity. We also scanned the reference list of included studies. Eligibility criteria included observational and randomized controlled studies (except for case reports and series) that were published as original report or as conference abstract presented in the English language or that could be adequately translated to the English language. We excluded studies with only unadjusted risk factors, studies that did not separate paroxysmal AF from new-onset AF, and studies that involved new-onset AF in patients only after cardiothoracic surgery.

Our search strategy employed medical subject headings (MeSH) and natural language text words related to sepsis and AF ({e.g. “Sepsis”[MeSH] OR “Shock, Septic”[MeSH] OR “Sepsis*” OR “Septic shock*” OR “Severe sepsis*”} AND {“Atrial fibrillation”[MeSH] OR “Atrial fibrillation*”}). Specific search strategies were designed by the first author and reviewed using the Peer Review of Electronic Search Strategies standard(9) by a health sciences librarian. The search strategy for each database can be found in Supplemental Digital Content 1. All titles and abstracts were screened for inclusion/exclusion based on eligibility criteria. When eligibility could not be determined by abstract alone, full texts were examined. After initial eligibility was ascertained by title/abstract, manuscripts of eligible studies were then examined in full to determine final inclusion. Kappa coefficient was calculated to determine inter-rater agreement. Endnote (Endnote^™, Clarivate Analytics, Philadelphia, PA) and Rayyan(10) web-based software were used to remove duplicates and to facilitate the screening process.

Study characteristics

We extracted data from included studies into a spreadsheet after performing training and calibration exercises. For studies missing relevant data, we contacted study authors by email. From each study we collected 1) citation information, demographics, type of study and study setting 2) definitions of sepsis and AF, and 3) the adjusted odds ratio (OR), standard error and 95% confidence interval (CI) for each candidate risk factor.

Methodologic quality

The methodological quality of observational studies was assessed using the Newcastle Ottawa Scale (NOS)(11). The risk of bias in randomized trials was assessed using the Cochrane Collaboration’s tool (12). Study selection, data extraction and study quality were all abstracted in duplicate (NAB, DMC) with third party adjudication (AJW) for disagreements.

Risk factor analysis

Quantitative analysis

Our quantitative meta-analysis examined risk factors identified in at least two studies. Our primary outcome was the pooled adjusted OR for each risk factor using a general inverse variance method with a random effects model. Statistics were performed with Review Manager Version 5.3 software(13). When not provided by the source publication, standard errors were calculated using the following formula:

$$SE=\frac{\mathrm{l}\mathrm{n}\left(UpperCI\right)-\mathrm{l}\mathrm{n}\left(LowerCI\right)}{3.92}$$

Qualitative analysis

For our qualitative analysis, we included all risk factors associated with new-onset AF reported in at least one study (pooled and un-pooled factors). Factors were grouped by factor type (demographic, comorbid condition, sepsis-related condition, ICU-related) and stratified by effect size with specific consideration of factors associated with a greater than 50% change in odds of new-onset AF. In order to assess if patients with factors increasing their risk for community-associated AF also have increased risk for sepsis-associated AF, we compared pooled and un-pooled risk factors and effect sizes for new-onset AF in sepsis with risk factors for community-associated AF as identified previously in Chamberlain et al.(14) and Schnabel et al.(15).

Results

Study selection

Search of Medline, Embase, Web of Science, and the Cochrane Central Register of Controlled Trials yielded 1136 studies; 11 studies met criteria for inclusion for qualitative assessment(4, 7, 16–24) and eight studies for quantitative meta-analysis(4, 7, 17–20, 23, 24) (Figure 1). Interrater reliability was strong (ĸ 0.80).

Figure 1.

Flow diagram for study inclusion. This flow diagram summarizes the four steps (identification, screening, eligibility, and final inclusion) used to select studies for inclusion in the review and meta-analysis.

Study characteristics

Of the 11 included studies, six were multi-center (4, 7, 16, 17, 20, 23) and five were single center (18, 19, 21, 22, 24). Five studies used retrospective cohort designs(7, 17, 19, 23, 24), four were prospective cohorts (4, 18, 20, 21), one was a case-control study(22), and one was a randomized controlled trial(16). Five studies examined new-onset AF in septic shock(16, 18–21), two in severe sepsis(23, 24), and four in sepsis(4, 7, 17, 22). A summary of included studies’ characteristics can be found in Table 1.

Table 1.

Characteristics of included studies

					Newcastle-Ottawa Scalea
Study ID	Type of Study	Sepsis Severity (sepsis definition)	Identification of atrial fibrillation	Patients with Sepsis	Selection	Comparability	Outcome
1. Asfar 2014(16)	Randomized control	Shock (SIRSb)	Undefined	776	N/Ac	N/Ac	N/Ac
2. Chui 2015(17)	Retrospective cohortd	Sepsis (not specified)	Administrative claims code	not reported	***	**	**
3. Guenancia 2015(18)	Prospective cohort	Shock (infection, vasopressor use, organ dysfunction)	Continuous EKG monitor review	66	****	**	***
4. Kane 2014(19)	Retrospective cohortc	Shock (ICDe codes for sepsis)	Nursing documentation review	109	****	**	**
5. Klouwenberg 2017(4)	Prospective cohort	Sepsis (SIRSb)	Nursing documentation review	1782	****	**	***
6. Launey 2017(20)	Prospective cohort	Shock (not specified)	12-lead EKG or continuous EKG monitor review	261	****	**	**
7. Lewis 2016(24)	Retrospective cohort	Severe (SIRSb)	EKG review	131	****	*	**
8. Seemann 2015(21)	Prospective cohort	Shock (SIRSb)	Continuous EKG automatic detection	65	****	**	***
9. Swamy 2015(22)	Case-controld	Sepsis (not specified)	Chart review	400	*	**	***
10. Walkey 2011(23)	Retrospective cohort	Severe (ICDe code 995.92)	Administrative claims code	49082	****	**	***
11. Walkey 2013(7)	Retrospective cohort	Sepsis (ICDe code 038.xx, 995.91, or 995.92)	Administrative claims code	60209	****	**	***

^aThe Newcastle-Ottawa Scale (NOS) assesses the quality of case-control and cohort studies based on categories of selection, comparability, and outcome (or exposure for case-control). NOS uses a star-based system where more stars represent higher quality within a specific category. Studies are awarded a maximum of four stars for selection, two stars for comparability, and three stars for outcome.

^bSystemic Inflammatory Response Syndrome (SIRS).

^cAsfar 2014 was evaluated using the Cochrane Bias tool for randomized trials. The risk of bias was determined to be low across multiple domains (random sequence generation, allocation concealment, blinding of outcome assessment, incomplete outcome data, and selective reporting); the risk of performance bias was determined to be high due to providers being unblinded to treatment.

^dConference abstracts.

^eInternational Classification of Diseases (ICD).

Methodologic quality

The methodologic quality of the observational studies was rated high, with five studies scoring nine of nine stars on the NOS, two scoring eight of nine, two scoring seven of nine, and one scoring six of nine. The risk for performance bias was high for Asfar 2014(16) because providers were not blinded to treatment group; other categories of bias were low.

Risk factor analysis

Quantitative analysis

Figure 2 demonstrates the effect sizes of all 44 factors (un-pooled and pooled) examined as potential risk factors for new-onset AF during sepsis. 18 factors were present in at least two studies and were included in the pooled meta-analysis. Seven pooled factors were associated with increased odds of new-onset AF: age, any acute organ failure, obesity, bacteremia/endocarditis, acute renal failure, prior congestive heart failure (CHF) and right heart catheterization. Only CHF and right heart catheterization were associated with a greater than 50% increase in odds.

Figure 2.

Risk factors for new-onset AF during sepsis stratified by risk factor type and pooled versus un-pooled analysis. Filled circles show the pooled adjusted odds ratios for risk factors from the meta-analysis. Open squares show the adjusted odds ratios for un-pooled risk factors. Risk factors with odds ratios greater than one are associated with an increased risk of AF during sepsis. Risk factors with odds ratios below one are associated with a decreased risk of AF during sepsis. Error bars denote 95% confidence intervals. Citation numbers for studies from where risk factors are reported are shown in the brackets. MICU: medical intensive care unit; SICU: surgical intensive care unit; ICU: intensive care unit; LOS: length of stay; COPD: chronic obstructive lung disease; BMI: body mass index; CHF: congestive heart failure; LBBB: left bundle branch block; EKG: electrocardiogram; UTI: urinary tract infection; SOFA: sequential organ failure assessment; MAP: mean arterial pressure. Inflammation is defined as C-reactive protein greater than 70mg/L or white blood cell count greater than 15×10⁹/L.

Five pooled factors were associated with decreased odds of new-onset AF in meta-analysis: Black race (compared to White), other race (compared to White), diabetes mellitus, urinary source of infection, and corticosteroid administration. Only corticosteroid use was associated with a greater than 50% decrease in odds. Six factors were not associated with new-onset AF after pooled analysis: hypertension, female sex, acute respiratory failure, and gastrointestinal, respiratory, and skin sources of infection.

Heterogeneity

Heterogeneity analysis showed that 10 pooled risk factors that included Walkey 2013 had I² greater than 75% (Figure 2). The direction of effect for individual study ORs from seven of the 10 high heterogeneity pooled factors that included Walkey 2013 were concordant. Individual ORs from the other three high heterogeneity pooled risk factors (CHF, gastrointestinal source of infection and skin or soft tissue source of infection) had discordant direction of effect estimates. Analysis of the acute renal failure and CHF effect estimates showed that the high heterogeneity was driven by the inclusion of Walkey 2013. When Walkey 2013 was removed from analysis, I² decreased to 0% and OR increased to 1.42 for acute renal failure and I² decreased to 68% and OR increased to 2.87 for CHF. When Walkey 2013 was removed from analysis from other pooled risk factor analyses, there was no decrease in heterogeneity.

Qualitative analysis

For our qualitative analysis comparing sepsis-associated AF risk factors with community-associated AF risk factors, we included 26 additional factors that were evaluated in only one study. Un-pooled factors associated with a greater than 50% increase in odds of new-onset AF included fungal infection, prior stroke, any ICU stay, hypotension requiring vasopressors, use of norepinephrine to obtain a mean arterial pressure of 80–85mmHg (compared to blood pressure target of 65–70mmHg), pre-hospitalization beta-blocker use, hyperlipidemia, and left bundle branch block on ICU admission. Albumin per 1 g/dl increase and admission to the medical ICU (compared to surgical) were associated with a greater than 50% decrease in odds of new-onset AF.

Of the 11 community-associated factors identified in Schnabel et al. and the 20 community-associated factors identified in Chamberlain et al., 10 factors had similar definitions with sepsis-related factors evaluated in our study and could be directly compared. A comparison of the effect sizes of risk factors for sepsis-associated versus community-associated AF can be found in Table 2. In general, demographic and comorbid conditions were stronger risk factors for community-associated AF than sepsis-associated AF. Some factors associated with increased risk for community-associated AF (e.g., diabetes mellitus, body mass index per point increase) were associated with lower risk for new-onset AF in sepsis.

Table 2.

Comparison of risk factors for sepsis-associated atrial fibrillation versus community-associated atrial fibrillation

		Effect Size
			Community-associated AF
	Factors	Sepsis-associated AFa	Schnabel 2009(15)b	Chamberlain 2011(14)b
Demographics	Female (vs male)	0.84 (0.68, 1.04)	0.53 (0.44, 0.63)	0.52 (0.43, 0.62)
	Black race (vs White)	0.65 (0.6, 0.71)	-	0.6 (0.47, 0.76)
	Age (per year)	1.05 (1.05, 1.06)	1.13 (1.11, 1.15)	-
Comorbid conditions	Treatment for hypertension or prior beta-blocker use	2.64 (1.03, 6.76)	1.8 (1.48, 2.18)	2.55 (2.13, 3.04)
	BMI (per 1 kg/m2)	0.94 (0.88, 0.99)	1.04 (1.02, 1.07)	-
	CAD/Prior MI	0.95 (0.87, 1.04)	1.44 (1.02, 2.03)	-
	Diabetes mellitus	0.87 (0.78, 0.98)	1.1 (0.87, 1.38)	1.87 (1.51, 2.32)
	Valvular heart disease	0.99 (0.92, 1.06)	2.38 (1.71, 3.32)	1.92 (1.49, 2.47)
	Hypertension	0.93 (0.84, 1.02)	-	2.16 (1.67, 2.79)
	Congestive heart failure	1.63 (1.01, 2.64)	3.2 (1.99, 5.16)	3.03 (2.32, 3.95)

Shown are effect sizes for factors evaluated in our study (sepsis-associated AF) and by Schnabel et al.¹⁵ and Chamberlain et al.¹⁴ (community-associated AF) that are similarly defined. Factors are stratified within each group (demographics, comorbid conditions) by the difference in effect size between risk factors for sepsis-associated AF and for community-associated AF. AF: atrial fibrillation; BMI: body mass index; CAD: coronary artery disease; MI: myocardial infarction.

^aOdds ratio (95% confidence interval).

^bHazard ratio (95% confidence interval).

Discussion

We evaluated risk factors for new-onset AF during sepsis using systematic review and meta-analysis. Although some risk factors for community-associated AF were also identified as risk factors for new-onset AF during sepsis (age, congestive heart failure), effect estimates of community-associated AF risk factors tended to be more modest during sepsis, and many community-associated risk factors (diabetes mellitus, hypertension, valvular heart disease, and coronary heart disease) were not associated with increased risk for new-onset AF during sepsis. Rather, the strongest and most frequently identified risk factors for new-onset AF during sepsis were factors associated with the acute sepsis event (corticosteroid use, right heart catheterization, hematologic failure, inflammation {C-reactive protein of greater than 70mg/L or white blood cell count greater than 15×10⁹/L}, fungal infection, vasopressor use, and higher blood pressure targets). Our findings have mechanistic and clinical ramifications, especially for guiding the selection or avoidance of medications (e.g., dopamine) with arrhythmogenic side effects during sepsis.

Two previous reviews have characterized AF during critical illness. A 2011 study by Yoshida et al.(25) reported un-pooled adjusted risk factors for AF during critical illness (not sepsis alone) from five studies, four of which included primarily surgical patients (two mixed ICUs and two surgical ICUs). Similar to our results, Yoshida et al. identified past use of nodal blocking agents, greater severity of illness, shock, sepsis, and catecholamines as factors associated with new-onset AF. Kuipers et al.(26) characterized un-pooled unadjusted and adjusted risk factors for new-onset AF during sepsis, but only two studies(7, 23) presented adjusted risk factors. Since 2014, an additional seven studies have published adjusted risk factors for new-onset AF during sepsis and were included in our analysis. In comparison to Kuipers et al., we identified several new factors including sepsis treatments (corticosteroid use and blood pressure targets) that may have important clinical implications for preventing AF during sepsis, and compared risk factors for AF in the community and during sepsis.

The risk factors examined in our analysis inform the understanding of AF pathogenesis during sepsis. AF is thought to occur in a two-step process of 1) development of an arrhythmogenic substrate, followed by 2) a triggering event(27). Our analysis identified chronic factors that are known to potentiate the formation of atrial fibrosis with long-term exposure (27) (e.g., CHF), but also several sepsis-related factors that may act to form an arrhythmogenic substrate over the short time frame of an acute sepsis event. For example, pneumonia due to streptococcus pneumoniae has been shown in animal models to result in myocardial microabscesses that progress to fibrotic lesions that may predispose to arrhythmia formation(28). Formation of arrhythmogenic lesions may explain our findings of an elevated risk of AF in patients with bacteremia/endocarditis and gram-positive infection.

Several potential AF triggers were also identified in our study. Catecholamines and increased adrenergic tone are thought to lead to AF initiation through complex mechanisms including increased myocyte automaticity(29). We found several risk factors for AF that increase circulating catecholamines: vasopressor use, higher blood pressure target in septic shock, and sepsis severity defined by increasing organ dysfunction. Our finding of increased risk with pre-hospitalization beta-blocker use may also suggest a catecholamine-related mechanism: removal of beta-blocker therapy during states of high adrenergic tone could further potentiate the effects of catecholamines on myocardial receptors thus predisposing to AF development. Catecholamine-related risk factors for AF suggest that, contrary to guideline recommendations(6), there may be not be a role for highly arrhythmogenic vasopressors such as dopamine in the treatment of sepsis as any patient who is in need of vasopressors likely also has an elevated risk for catecholamine-mediated AF.

Infiltration of inflammatory cells in the myocardium is implicated in AF risk(30). Our study identified two inflammation-related factors associated with AF risk during sepsis: increased inflammation: elevated C-reactive protein or white blood cell count (associated with increased AF risk), and the administration of glucocorticoids (associated with a decreased AF risk). Glucocorticoids may decrease inflammatory cell mediated damage to the myocardium or, alternatively, may reduce levels of circulating catecholamines by decreasing the time to shock resolution and vasopressor requirements(31, 32). The role of inflammation in AF development may also explain the decreased risk of AF during sepsis in patients with diabetes mellitus. Diabetes mellitus has been associated with lower risks of other sepsis-associated events including acute respiratory distress syndrome(33), potentially due to limited oxidative damage from dysfunctional neutrophils(34). The catecholamine and inflammation mediated risk factors identified in our study provide potential treatment targets for medications (e.g. beta-blockers, catecholamine sparing agents, anti-inflammatory medications) to be used in future studies examining AF.

Given the risk of acute hemodynamic compromise(3) and poor long-term outcomes(5) associated with new-onset AF during sepsis, identifying modifiable risk factors represents an important step to guide efforts to prevent and treat new-onset AF during sepsis. Our findings suggest that the risk for new-onset AF during sepsis is driven more strongly by sepsis-related events and sepsis-related treatments than by common community-associated AF risk factors such as underlying cardiac disease. Importantly, risk factors for AF that occur during sepsis may be modifiable. Targeting modifiable AF risk factors during sepsis may potentially prevent or expedite resolution of AF and improve patient outcomes, a hypothesis supported by meta-analyses showing that avoidance of AF risk factors such as dopamine(35), right heart catheterization(36), and high mean arterial pressure targets(37) during septic shock (or use of corticosteroids(38)) is associated with improved outcomes. Further identification of unmodifiable risk factors for AF, such as increasing age, CHF, and location of infection, identify patients who are at risk for new-onset AF in whom introduction of additional risk factors should be avoided. Our findings that pre-admission beta-blocker use - and likely withdrawal during sepsis - is associated with new-onset AF informs future studies of beta-blocker therapy during septic shock as a prophylactic strategy to potentially reduce AF incidence and improve outcomes(39).

Our study has several limitations. The designs of the included studies were heterogeneous. Studies variably included patients with different sepsis severity and identified AF using different strategies (e.g. electrocardiogram review, bedside nursing identification, administrative claims code). Differences in sepsis severity between trials may impact the effect size of risk factors (i.e. effect modification by sepsis severity). For example, corticosteroid use may have a stronger negative correlation with AF incidence during sepsis in patients with more severe septic shock. No sepsis-related risk factors were present in two or more studies examining different sepsis severity populations, and thus we could not determine the effect of effect modification by sepsis severity on our results. Differences in AF identification may have led to variable study inclusion of clinically relevant AF vs occult AF that may have distinct risk factors and outcomes. In our pooled analysis, we identified high heterogeneity measured by I² more frequently in analyses that included Walkey 2013, likely due to differences in defining risk factors (e.g., new dialysis vs. acute renal failure); despite heterogeneity, the direction of effect was generally consistent between studies.

Another possible limitation of our study is the inclusion of studies that identified AF and sepsis through administrative claims codes. In studies using administrative claims codes, the temporal relationship between AF and sepsis may not be known, and it may be that AF occurs well before or well after an acute sepsis event. However, the AF risk factors we identified from studies using administrative claims codes included sepsis-specific factors (e.g. right heart catheterization and bacteremia/endocarditis) that likely occurred near the timing of sepsis itself and we thus feel confident that sepsis-associated factors from studies using administrative claims codes correlate with risk of AF during a hospitalization that includes sepsis.

No studies included in our analysis identified sepsis or septic shock using the current Sepsis-3 Guidelines(6). Most studies instead defined sepsis according to the Sepsis-1 definition operationalized by the Systemic Inflammatory Response Syndrome(40). Although both the Sepsis-1 and Sepsis-3 definitions have reasonable discrimination for hospital mortality(41), further research is needed to corroborate AF risk factors identified using older sepsis definitions with the current Sepsis-3 definition.

Finally, there is not yet a well validated tool for assessing the risk of bias from observational studies alone. Although the NOS score allows reviewers to assess and compare the quality of observational studies with good inter-rater reliability, the criterion validity and intra-rater reliability have not been established.

Conclusions

Our systematic review and meta-analysis identified acute and chronic factors that were associated with the risk for new-onset AF during sepsis. Most of the factors identified were related to the sepsis event itself, and many of the strongest factors were related to the effects of catecholamines or inflammation. Recognition of sepsis-related factors points to potential differences in the mechanisms of AF development in sepsis versus in the community and provides potential targets for future treatment studies and clinicians focusing on AF prevention in sepsis.

Supplementary Material

Supplement

1. Database search strategies. Search strategies for Medline, Embase, Web of Science, and the Cochrane Central Register of Controlled Trials.