Atrial fibrillation (AF) is the most common serious heart rhythm disorder and is frequently detected for the first time after surgery.1,2,3 Published articles report a wide range in the incidence of new-onset perioperative AF,2,4 which may also vary according to the type of surgery, the baseline risk in the population under study, the approach to electrocardiography (ECG) monitoring, and the definition used for perioperative AF (e.g., ≥ 30 seconds AF duration, AF detected on a monitor, or AF that results in symptoms or requires treatment).2,4 New-onset perioperative AF has been associated with higher rates of adverse outcomes and mortality as well as longer length of hospital stay after non-cardiac surgery.5,6,7 In the perioperative setting, AF often resolves before hospital discharge, and the long-term prognosis of patients with AF occurring transiently with stress (AFOTS) is not well-defined.2,8,9,10 The occurrence of AFOTS could be secondary to physiologic stressors associated with surgery with no long-term consequences; but it may also be newly detected paroxysmal AF.2 Understanding whether AFOTS is a short-term problem or the first presentation of a chronic disease (i.e., paroxysmal AF) is critical because the risk of ischemic stroke associated with paroxysmal AF can be reduced with oral anticoagulants (OAC).2 Clinical practice guidelines and position statements have reinforced this position.11,12,13,14,15,16 Variation in the incidence estimates and the risk of recurrence of perioperative AF may contribute to the uncertainty in how to manage these patients.

The objective of this study was to systematically review the published literature to estimate: 1) the proportion of patients in whom new-onset AF is detected during hospitalization for non-cardiac surgery, and 2) the proportion of such patients in whom a recurrence of AF is detected over long-term follow-up. We also hypothesized that prospective studies with continuous ECG monitoring would have higher rates of AF detection than those that did not.

Methods

Protocol

The original review protocol was pre-registered (PROSPERO CRD42017068055; 1 September 2017). The differences between the original and final protocol are as follows. We initially planned to include observational and interventional studies including more than ten participants. Nevertheless, to make the number of eligible studies manageable, we changed this threshold to those with more than one hundred participants. As for the secondary outcomes, we initially planned to collect data on risk factors for the development of AF during hospitalization for non-cardiac surgery. To make the review manageable, and because these data are not readily combinable, we elected to forego collection of these data.

Eligibility criteria

This systematic review included prospective and retrospective observational studies of patients hospitalized for non-cardiac surgery. Included studies reported the incidence of new-onset AF during hospital admission for non-cardiac surgery, and/or the rate of AF recurrence following discharge from hospital in patients who had new-onset perioperative AF. We only included studies with > 100 participants. We excluded data from patients with a history of AF before hospital admission and from patients who underwent cardiac surgery.

Search strategy

We created the search strategy (available in the Electronic Supplementary Material [ESM] as eAppendix 1) with input from a medical librarian. We searched Ovid EMBASE (1974–November 2019), Ovid MEDLINE (1946–November 2019), and The Cochrane CENTRAL Library (Wiley) (November 2019). We did not impose language restrictions. The references of eligible papers were screened and experts were consulted to identify additional studies. We did not systematically hand-search conference abstracts or grey literature, but we did include conference abstracts that were captured by our database searches. Search results were imported into Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia; available at www.covidence.org). Pairs of reviewers independently screened study titles and abstracts for eligibility using a pre-piloted template. If either reviewer thought the citation could be eligible, the full-text article was evaluated for eligibility. Full texts of the potentially eligible studies were retrieved and screened by pairs of reviewers. Reviewers recorded the principal reason for exclusion. Disagreements were resolved through discussion. Reviewers recorded the principal reason for exclusion.

Data extraction and outcomes

Independently, pairs of reviewers extracted data pertaining to study and participant characteristics, AF incidence, and AF recurrence into a pre-piloted database. Disagreements were resolved through discussion. Reviewers recorded type of surgery, methods for AF diagnosis, approach used to rule out a prior history of AF, and whether studies used continuous ECG monitoring. Standard error (SE) for each study was calculated using the following formula: SE = \(\surd\)((new-onset AF/total population) * (1- (new-onset AF/ total population))/total population). We grouped studies according to the major organ system under operation. We included studies that combined multiple types of surgeries under the heading “non-cardiac”. Reviewers extracted the number of patients with AF and at risk, excluding those with a prior history of AF when this was reported. Reviewers contacted study authors to clarify ambiguities.

Methodological quality of included studies

Pairs of reviewers appraised the methodological quality of included studies using a pre-piloted tool based on the Newcastle-Ottawa Scale (Table 1).4,17 The three domains evaluated in each study were: exclusion of patients with a history of AF, detection of AF in hospital, and detection of AF recurrence following the initial hospitalization. A score was allotted for each domain in each study. We considered studies to be of high quality if they searched medical records to exclude patients with a history of AF, used prospective, continuous, and systematic ECG review to detect AF in hospital, and employed a systematic monitoring protocol to detect AF recurrence.

TABLE 1 Quality assessment framework
Full size table

Meta-analysis and subgroup tests

We used CMA software (Biostat Inc, Englewood, NJ, USA) to perform analyses. For each study, we inputed the incidence of perioperative AF as binomial proportions and calculated the accompanying standard errors.18 For the overall group of studies and for each type of surgery, we performed meta-analysis using a DerSimonian–Laird random effects model.19 We assessed between-study heterogeneity quantitatively with the I2 test and we planned to present a pooled estimate only if I2 < 50%.19 We hypothesized that prospective studies with continuous ECG monitoring (i.e., quality score of 1 as defined in Table 1) would detect higher AF rates than those that did not, both in the overall population and in each different type of surgery. We tested this by creating a DerSimonian–Laird random effects model that pool studies based on quality score and compared subgroup effects with an interaction test. To estimate of the proportion of variance explained by continuous monitoring, we created a binary logistic meta-regression model and calculated a goodness-of-fit statistic and accompanying R2 analogue.20

Results

Screening process

The electronic search generated 39,233 unique citations (Fig. 1). After reference and full-text screening, 346 studies met eligibility criteria, including 345 that reported AF incidence and one that exclusively reported AF recurrence after hospital discharge. Twelve reported data on both incidence and recurrence.

Fig. 1
figure 1

Study selection diagram (PRISMA Format)

Full size image

Incidence studies

Study Characteristics

Three hundred and forty-five studies published between 1973 and 2019 reported the incidence of new-onset AF associated with non-cardiac surgery (Table 2; eAppendix 2, ESM). Five studies were published in Mandarin, four studies in Spanish, two studies in French, two studies in Japanese, and one study each in Hungarian, Russian, Icelandic, Romanian and Italian (eAppendix 3, ESM). The remaining studies were published in English. The intensive care unit (ICU) was the primary setting in 55 studies. Seventy studies were prospective cohort studies and 275 were retrospective. One hundred and fifty-three studies included patients who underwent pulmonary surgery, 32 studies included patients who underwent esophagectomy, 20 studies included patients who underwent thoracic surgery not otherwise specified, 60 studies included patients who underwent abdominal or digestive surgery, ten studies included patients who underwent vascular surgery, 26 studies included patients who underwent orthopedic surgery, and 44 studies reported aggregate data from patients who had undergone different types of non-cardiac surgery. The mean age of patients in the included studies ranged from 32 to 80 yr.

TABLE 2 Median incidence of new-onset atrial fibrillation according to type of surgery
Full size table

The methodologic quality of included studies varied greatly (eAppendix 2, ESM). One hundred and sixty-two studies (47.0%) reported AF incidence after ruling out a prior history of AF based on high-quality methods; 74 (21.4%) described explicit searching of medical records to rule out a prior history of AF (quality score 1), and 88 (25.5%) reported that patients in the study did not have a history of AF but did not mention searching medical records (quality score 2). Only 27 (7.8%) studies monitored for AF using high-quality methods (prospective design with continuous ECG monitoring, quality score 1); five (1.4%) of these studies used high-quality methods in both domains.

The incidence of new-onset AF during hospitalization for non-cardiac surgery reported in 345 studies (n = 5,829,758) ranged from < 0.01% to 50.3%. Table 2 reports the ranges of AF incidences according to type of surgery performed and whether continuous monitoring was used. Figure 2 summarizes the estimates of the incidence of new-onset AF in studies that used continuous, prospective monitoring. Heterogeneity was substantial (I2 > 90%) both overall and by surgical subgroup (eAppendices 4 and 5, ESM); therefore, we did not pool results to generate a summary estimate. We tested the hypothesis that prospective studies with continuous ECG monitoring would find higher incidences of AF. To do this, we created a random effects binary logistic meta-regression model that compared the incidence of AF in studies with AF ascertainment quality scores of 1 with those with AF ascertainment quality scores of 2,3, or 4 as defined in Table 1. Reported AF incidences were significantly higher in studies with continuous monitoring compared with those without (test of model P < 0.001) according to the goodness-of-fit test (P < 0.001) and R2 analogue 0.02 (eAppendix 5, ESM). Nevertheless, it suggested that only a small amount of variance (2%) was explained by differences in monitoring intensity. We found no evidence of a subgroup effect when comparing studies using continuous monitoring with those that did not across surgical subtypes (eAppendix 4, ESM).

Fig. 2
figure 2

Incidence of new-onset atrial fibrillation after non-cardiac surgery in prospective studies that used continuous electrocardiography.

Full size image

Studies of long-term recurrence of AF

Thirteen studies (n = 25,726) reported the long-term recurrence of AF following the index hospitalization (Table 3). Follow-up ranged from one month to 5.4 years after discharge. Four studies were prospective cohort studies and nine were retrospective. Follow-up duration and the methods used for AF detection varied markedly (Table 3). Only one study was assessed as having high-quality methods for AF detection—Higuchi et al.21 used a systematic 12-month protocol for the surveillance of AF recurrence in cancer surgery patients with postoperative AF. Seventy-seven patients with confirmed postoperative AF wore either a two-week event-triggered recorder or a 24-hr Holter monitor at one month and 12 months postoperatively. Atrial fibrillation recurrence of at least 30 sec, as confirmed by a cardiologist, was documented in 24 (31.1%) patients. Four other studies reported AF recurrence rates of greater than 15%. Jesel et al. followed patients with postoperative AF following lung transplantation for a mean (standard deviation) of 2.9 (2.4) years.22 Patients were assessed weekly for three months, then monthly until one year, and every three months thereafter. Atrial fibrillation was documented by 12-lead ECG or ambulatory ECG monitoring, as reviewed by two cardiologists. The rate of AF recurrence was 16.5%. Rachwan et al. followed patients who developed postoperative AF following liver transplantation.23 All were discharged from hospital in sinus rhythm. Of the 42 that had a 12-lead ECG in follow-up, ten (24%) were in AF. Gialdini et al. used administrative claims data to identify Californian patients with new-onset perioperative AF.24 Among 12,874 patients with no prior history of AF who developed new-onset perioperative AF, 4,799 (37.3%) had another healthcare encounter for AF in the ensuing 12 months. Kim et al. used a similar approach with 11,347 patients in South Korea, finding that 2289 (20.2%) had another encounter for AF over a median 47 months of follow-up.25

TABLE 3 Characteristics of included studies on AF recurrence for atrial fibrillation after non-cardiac surgery
Full size table

Discussion

In this systematic review, we found that among 345 published articles, the reported incidence of new-onset AF during hospitalization for non-cardiac surgery ranged from very rare (< 0.01%) to extremely common (~ 50%). This variability prohibited pooling data, even when reports of the same type of surgery were considered. Consistent with our hypothesis, prospective studies that used continuous ECG monitoring found higher incidences of AF than those that did not. Although some of the variability in study results was explained by detection methods and different surgical populations, much remains unexplained. There was similar variability in the 13 articles reporting on the long-term rate of AF recurrence in this population. Several studies, however, reported recurrence rates upwards of 20%. Rigorously designed studies with continuous and prospective monitoring are required to more accurately estimate the incidence and long-term recurrence of new-onset postoperative AF in patients who have undergone non-cardiac surgery.

To the best of our knowledge, this is the first systematic review to broadly assess the incidence and recurrence of new-onset perioperative AF following non-cardiac surgery. Most reviews to date have focused on cardiac surgery.26,27 Reviews in non-cardiac surgery have assessed surgical sub-populations and did not systematically assess study methodology for detection.28,29 This review has two important strengths. First, when possible, we excluded data from patients with a prior history of AF to isolate the population for whom the long-term prognosis is in question. Second, our methodological appraisal focused on methods used for AF detection and specifically tested the hypothesis that prospective studies with continuous ECG monitoring reported higher incidences of AF. As AF is often an intermittent arrhythmia, the probability of detecting AF increases with the intensity of ECG monitoring.30,31,32 As in our previous review of medical patients, studies with continuous monitoring found a significantly higher incidence of AF, although meta-regression modelling suggested significant unexplained variance.33 These findings have informed the design of the AFOTS incidence study (clinicaltrials.org: NCT03552588) which is enrolling consecutive at-risk patients admitted to an ICU and continuously monitoring them with a continuous ECG monitor for up to 14 days.34

Accurate estimates of the incidence of new-onset perioperative AF are critical to guide future research in this patient population. There is interest in preventing new-onset perioperative AF and in optimizing outcomes for affected patients.11,12,13,14,15,16 New-onset perioperative AF has been associated with adverse outcomes during the index hospital stay in at least two large, multi-centre studies.5,6 Nevertheless, the incidences of new-onset AF in these two studies (2.5 and 3.0%) are at the lower end of what we found in this review.5,6 Confounding may explain these findings; AF may be noticed more frequently in patients who are more acutely ill with closer monitoring and in patients who stay longer with longer monitoring. In addition, worse outcomes for patients with AF could be explained by a higher burden of co-morbidities.35,36,37 This review underscores the need for standardized AF detection to minimize detection bias using prospective designs with continuous ECG monitoring. Continuous monitoring will also help estimate the minimum burden of AF that is associated with adverse outcomes.

Evidence is lacking to guide the long-term management of patients with new-onset perioperative AF. The majority of patients are discharged from hospital in sinus rhythm, whether it is due to spontaneous, pharmacologic, or electrical cardioversion.9,10 Nevertheless, clinicians are faced with the decision as to whether these patients should be treated for AF, with the most pressing question being whether to start them on long-term OAC treatment. New-onset perioperative AF has been associated with an increased long-term risk of stroke in large observational studies.23,38,39 Nevertheless, the incidences of AF in these studies is again at the lower end of what was seen in our review (0.4%, 0.8%, 2.2%), raising the question of selection bias.

We assessed long-term rates of AF recurrence in this population. This interest was driven by the hypothesis that patients who have a recurrence of AF following an initial presentation of AF perioperatively could be more likely to have paroxysmal AF and might therefore benefit from OAC.2,8,40 Nevertheless, we found very wide ranges in the long-term recurrence of AF. Only one 77-patient study systematically re-assessed patients using long-term ECG monitoring and found a cumulative recurrence rate of 31.1% over one year.21,40 Long-term ambulatory ECG monitoring may identify patients with recurrent AF who might benefit from OAC treatment as we await the results of randomized clinical trials. Ongoing studies are attempting to better define the long-term prognosis and management of this population. The AFOTS follow-up study attempts to overcome some of the limitations of prior studies.8 This study identifies patients with new-onset AF during hospitalization for medical illness or non-cardiac surgery who have returned to sinus rhythm at the time of hospital discharge. Cases and matched controls wear two 14-day Holter monitors over the ensuing 12 months. At least one randomized trial is being conducted in this population: Anticoagulation for Stroke Prevention In Patients With Recent Episodes of Perioperative Atrial Fibrillation After Noncardiac Surgery ASPIRE-AF is a pilot trial (clinicaltrials.org; NCT03968393) assessing the feasibility of a full study of OAC vs no OAC in this population.

Limitations of review

This review is principally limited by the design and methodologic descriptions of the included studies. Most studies were either single centre or combined heterogeneous types of surgery. A small number of studies defined the minimum length of AF episodes they considered to be relevant. We did not collect data on the long-term risk of stroke associated with new-onset AF following non-cardiac surgery in this review. We are, however, assessing this in a separate, dedicated review (PROSPERO; CRD42017054309).

Conclusions

Rates of AF incidence and long-term recurrence of AF first detected following non-cardiac surgery vary markedly. Differences in the intensity of ECG monitoring and type of surgery may account for this variation. Atrial fibrillation recurs after hospital discharge in up to 37% of patients.