In this retrospective study, we consecutively collected a total of 14,932 patients, who underwent isolated CABG at Beijing Anzhen Hospital between January 2018 and September 2021. The exclusion criteria were as follows: History of atrial fibrillation (AF), non-CABG surgical procedures, missing postoperative continuous electrocardiography (ECG) and medication records, or being 18 years old. This resulted in 3084 patients being excluded from this study, leaving a total of 11,848 patients. Fig. 1 depicts the flowchart of the included and excluded patients.

Fig. 1.

Study flowchart. CABG, coronary artery bypass grafting; ECG, electrocardiography; POAF, postoperative atrial fibrillation; PSM, propensity score matching; IPTW, inverse probability of treatment weighting.

POAF was diagnosed using a continuous ECG monitor, whereby it was defined after surgery lasting at least 5 minutes as either newly onset atrial fibrillation or atrial flutter, and treatment was required in accordance with the definition provided by the Society of Thoracic Surgeons (STS) National Database [8]. Patients with POAF were treated with amiodarone, and electrical cardioversion was administered to patients with POAF who developed hemodynamic instability. All patients were first transferred to the ICU postsurgery, where POAF occurrence was continuously monitored during their stay duration, as well as during the first 4 days after being transferred back to the general ward. If POAF was detected, patients were continually monitored until a normal SR returned or they were discharged. The POAF duration was defined as the interval between the first day and the last day that POAF occurred (continuous or intermittent POAF during the monitoring time) [7]. Ultimately, patients were first divided into 2 groups: non-POAF (n = 8244) and POAF (n = 3604). Then, the POAF group was further divided into 2 subgroups based on whether POAF had occurred for longer than 48 hours: POAF durations shorter than 48 hours (n = 2473) and longer than 48 hours (n = 1131). The 48-hour cutoff was based on AF, with that duration considered the nodal point for initiating long-term anticoagulation therapies [9].

After anesthesia, the patients who underwent CABG had access through a sternotomy. Off-pump patients were fully heparinized at the completion of harvesting the conduits, while the heart was stabilized when conducting the grafts using the Octopus tissue stabilizer system (cardiopulmonary bypass was established first in on-pump patients, and the Octopus tissue stabilizer system was not used). After all grafts were performed, proximal anastomosis was performed on the aorta using partial clamping. The heparinized state was reversed using protamine and the chest was closed (on-pump patients were firstly withdrawn from cardiopulmonary bypass, and after the heart resumed beating, the chest was closed).

Patient data were collected using an electronic medical records system, including demographics, comorbidities, in-hospital outcomes, medications, preoperative laboratory, echocardiography, and continuous ECG monitoring data. The primary outcomes measured were in-hospital mortality, stroke, acute respiratory failure (ARF), acute kidney injury (AKI), and significant gastrointestinal bleeding (GIB). In-hospital mortality was defined as death from any cause during hospitalization. Stroke was defined as a permanent neurological deficit with imaging evidence of cerebral artery occlusion, as diagnosed by neurologists. ARF was defined as arterial partial O${}_2$ pressure (PaO2) remaining $\le$60 mmHg after oxygen therapy, as measured via nasal cannula, or PaO2/inhaled partial pressure of oxygen remaining $\le$200 mmHg after oxygen therapy, as measured via venturi mask, with or without pH 7.35 and arterial partial CO${}_2$ pressure $>$45 mmHg [10]. AKI was defined as a 0.3 mg/dL or 50% increase in serum creatinine from baseline or urine volume 0.5 mL/kg/h for 6 hours [11]. Significant GBI was defined as nasogastric drainage, yielding obviously red or coffee-ground-appearing blood, as well as melena (black, tarry stools), hematemesis (blood vomiting), or hematochezia (passage of fresh blood through the anus), along with hemoglobin levels decreasing $>$2 g/mL/day, and transfusion of blood products being required [12]. The measured secondary outcomes were postoperative length of stay (LOS) and intensive care unit (ICU) duration. Considering that the inclusion of outcomes that partially occurred during POAF duration may blur the relationship between POAF duration and outcomes; thus, to clarify whether the differences in outcomes were caused by “prolonged duration of POAF”, outcomes that occurred during POAF duration in the short-duration group and within 48 hours of POAF duration in the long-duration group were excluded.

The normality distribution test for continuous variables was performed using the Kolmogorov‒Smirnov test. The results are presented as the mean $\pm$ standard deviation for normally distributed continuous variables and were compared between two groups by Student’s t test. When continuous variables were not normally distributed, interquartile ranges (25th–75th percentiles) were used, and groups were compared using the Mann‒Whitney U test. Categorical data are presented as n (%) and were compared using chi-squared tests. Missing data are absent completely at random with a missing degree of less than 1%. Mean imputation was used to fill in the missing baseline data values.

Analyses were performed by propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) to control the differences in baseline characteristics of patients between the non-POAF group and POAF group and between the group with POAF durations shorter than 48 hours and the group with POAF durations longer than 48 hours to account for selection bias and potential confounding factors in outcome comparisons between the groups. A propensity score for each patient was calculated using multivariable logistic regression. PSM was performed using the 1:1 nearest neighbor matching method and optimal matching with a caliper width of 0.25 standard deviations. Regarding IPTW, the inverse propensity score served as the weight for patients with POAF or POAF durations longer than 48 hours, and the inverse of 1 minus the propensity score served as the weight for patients without POAF or POAF durations shorter than 48 hours.

A restricted cubic spline (RCS) model with four knots (at the 5th, 35th, 65th, and 95th percentiles) and including major confounding variates associated with adverse perioperative outcomes after cardiac surgery, including age, male sex, body mass index (BMI), history of hypertension, diabetes, chronic kidney disease (CKD), stroke and cardiac surgery, operation time and preoperative level of left atrial diameter (LAD), left ventricular ejection fraction (LVEF), creatinine (Cr), creatine kinase MB (CK-MB), troponin I (TnI), and brain natriuretic peptide (BNP), was used to explore the nonlinear dose‒response relationship between POAF duration and each primary outcome. Univariable and multivariable logistic regression models were used to estimate the odds ratios (ORs) and 95% confidence intervals (CIs) of the independent association between POAF or prolonged POAF duration and the occurrence of primary outcomes. PSM, IPTW, and multivariable logistic regression analyses were performed by including the same confounding variables in the RCS. In addition, we used logistic regression analysis to analyze the risk factors that cause a prolonged POAF duration.

Statistical analyses were performed by R software (version 4.2.2, R Foundation for Statistical Computing, Vienna, Austria). The statistical significance level was set at two-tailed p 0.05.

Out of the 11,848 patients included in this study, 75.6% were male. POAF was detected in 3604 (30.4%) patients, which was consistent with previous studies regarding the POAF prevalence among large patient cohorts who had undergone CABG [1, 2, 13]; POAF durations longer than 48 hours were observed in 1131 (31.4%) patients.

With respect to the patient baseline parameters, significant differences were observed between the non-POAF group and POAF groups for age, male sex, the prevalence of hypertension, chronic obstructive pulmonary disease (COPD), stroke, hyperlipidemia, prior cardiac surgery, the rate of on-pump CABG, postoperative anticoagulation and the level of the operation time, LAD, LVEF, left ventricular end-diastolic dimension (LVEDD), triglyceride (TG), total cholesterol (TC), Cr, CK-MB, TnI, K${}^{+}$, Ca${}^{2+}$, Mg${}^{2+}$, and BNP (all p 0.05) (as shown in Supplementary Table 1). Additionally, significant differences were present in age, the prevalence of CKD, prior percutaneous coronary intervention (PCI), the rate of on-pump CABG, postoperative anticoagulation and the level of operation time, LAD, LVEF, LVEDD, Cr, K${}^{+}$, Ca${}^{2+}$, Mg${}^{2+}$, and BNP (all p 0.05) between the groups with POAF durations shorter than 48 hours and the group with POAF durations longer than 48 hours (Table 1). After applying PSM and IPTW, the standardized mean difference for all baseline variables was 0.1, indicating that the baseline characteristics between groups were balanced (Supplementary Figs. 1,2).

We compared the primary outcomes between the non-POAF and POAF groups and found that after applying PSM and IPTW adjustments the in-hospital mortality, ARF, AKI, and significant GIB were all significantly higher among POAF patients than among non-POAF patients (Supplementary Table 2). These outcomes were determined to be independently associated with POAF by logistic regression analyses (Supplementary Table 3). However, even though the stroke incidence was higher among POAF compared to non-POAF, no statistically significant difference was present, nor was stroke occurrence associated with POAF in the logistic regression analyses. In terms of secondary outcomes, we also found that POAF patients had longer LOS and ICU durations.

To determine whether POAF duration affected the prognosis of postoperative CABG patients, we compared the primary outcomes between the groups with POAF durations longer than 48 hours and POAF durations shorter than 48 hours. For the primary outcomes and prior to applying PSM and IPTW adjustments, we found no significant difference in the incidence of stroke (1.9% vs. 1.3%, p = 0.147) in patients with POAF durations longer than 48 hours compared to the controls. However, there were higher rates of in-hospital mortality, ARF, AKI, and significant GIB in the patients with POAF durations longer than 48 hours (all p 0.001). The lack of any significant differences between the two groups was also observed in stroke occurrence after adjusting for PSM and IPTW (PSM 2.0% vs. 1.3%, p = 0.239; IPTW 1.9% vs. 1.4%, p = 0.275), as well as for in-hospital mortality (PSM 3.1% vs. 2.5%, p = 0.436; IPTW 3.3% vs. 2.5%, p = 0.301); however, significant differences were still present for the other outcomes (Table 2).

To further elucidate the relationship between POAF duration and the primary outcomes, RCS analyses were conducted, whereby a linear relationship was present between POAF duration and in-hospital mortality, stroke, and ARF (all p for nonlinearity $>$0.05). In contrast, a nonlinear relationship was found between POAF duration and AKI (p for nonlinearity = 0.009) and significant GIB (p for nonlinearity = 0.0219), for which the inflection point was 49 hours, indicating that the incidences of AKI and significant GIB increased after POAF durations $>$49 hours (Fig. 2).

Fig. 2.

Restricted cubic spline plots of associations between POAF duration and postoperative complications. (A) In-hospital mortality, (B) stroke, (C) acute respiratory failure (ARF), (D) acute kidney injury (AKI), (E) significant gastrointestinal bleeding (GIB). The solid line and shaded area represent the log-transformed odds ratios and corresponding 95% confidence intervals. A linear relationship was present between POAF duration with in-hospital mortality, stroke, and ARF (all p for non-linearity $>$0.05). By contrast, a non-linear relationship was found between POAF duration and AKI (p for non-linearity = 0.009) and significant GIB (p for non-linearity = 0.0219). POAF, postoperative atrial fibrillation.

We also conducted logistic regression analyses, in which a longer POAF duration was not independently associated with in-hospital death (adjusted OR: 1.60, 95% CI: 0.97–2.65, p = 0.068) and stroke (adjusted OR: 1.28, 95% CI: 0.71–2.34, p = 0.414) but was independently associated with ARF (adjusted OR: 2.96, 95% CI: 1.47–6.09, p = 0.003), AKI (adjusted OR: 2.37, 95% CI: 1.42–3.99, p = 0.001), and significant GIB (adjusted OR: 2.60, 95% CI: 1.38–5.03, p = 0.004). Moreover, the results were still valid even after adjusting for PSM and IPTW (Table 3). Furthermore, the POAF group with a duration longer than 48 hours had a longer postoperative LOS and ICU duration (Table 2).

To control the bias in the effect of anticoagulation and antiplatelet therapy on bleeding/ischemic events, we further included postoperative anticoagulation and antiplatelet therapy (only records of postoperative anticoagulation and antiplatelet regimens prior to stroke or significant GIB occurrences), applying the same confounding variables in RCS into the multivariable logistic regression of stroke and significant GIB. The association between prolonged POAF duration and stroke (adjusted OR: 1.23, 95% CI: 0.67–2.25, p = 0.497) and significant GIB (adjusted OR: 2.61, 95% CI: 1.38–5.05, p = 0.003) remained consistent with the previous multivariable logistic regression analyses (stroke: adjusted OR: 1.28, 95% CI: 0.71–2.34, p = 0.414; significant GIB: adjusted OR: 2.60, 95% CI: 1.38–5.03, p = 0.004), after including postoperative anticoagulation and antiplatelet therapy as confounding variables in the logistic regression analyses.

After including POAF risk factors, such as age, male sex, BMI, hypertension, diabetes, LAD, LVEF, CK-MB, K${}^{+}$, and BNP, in the multivariable logistic regression analysis, we found that the statistically significant risk factors were age, BMI, on-pump CABG, long operation time, Mg${}^{2+}$, CK-MB, TnI, BNP, LAD, and low LVEF (all p 0.05, shown in Supplementary Table 4).