Abstract
Background
The optimal management strategy for postoperative atrial fibrillation (POAF) following coronary artery bypass grafting (CABG) remains debated. While rhythm control aims to restore sinus rhythm, its impact on clinical resource utilization and long-term outcomes remains unclear. This study aimed to investigate the association between an initial rhythm control strategy (as an escalation to rate control) and clinical outcomes compared to rate control alone.
Methods
This retrospective cohort study identified 663 adult patients who developed POAF after isolated CABG from the MIMIC-IV database. We used two statistical approaches to control for confounding: a primary analysis on a 1:1 propensity score-matched (PSM) cohort and a sensitivity analysis using inverse probability of treatment weighting (IPTW) on the entire cohort. The primary outcome was sinus rhythm (SR) conversion within 24 h. Secondary outcomes included length of ICU stay, acute kidney injury (AKI), and long-term mortality.
Results
The rhythm control strategy was associated with a higher 24-hour SR conversion rate compared to rate control (PSM cohort: 71.7% vs. 32.3%; OR, 2.65; P = 0.010). However, patients in the rhythm control group had a significantly longer ICU length of stay (median 3.42 vs. 3.08 days; P = 0.026). No significant differences were found in AKI incidence or long-term mortality. Subgroup analysis indicated that the benefit of SR conversion was attenuated in patients with diabetes (P for interaction < 0.05). Findings were consistent in the IPTW sensitivity analysis.
Conclusion
Although effective for early rhythm restoration, an initial rhythm control strategy is associated with longer ICU stays and no survival benefit compared to rate control.
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1186/s12872-026-06317-3.
Keywords: Postoperative atrial fibrillation, Coronary artery bypass grafting, Rhythm control, Rate control
Introduction
Postoperative atrial fibrillation (POAF) is the most prevalent arrhythmia following cardiac surgery, affecting 20%-40% of patients after coronary artery bypass grafting (CABG) [1]. POAF is associated with adverse outcomes such as increased risk of stroke, prolonged hospitalization, elevated healthcare costs, and both short- and long-term mortality [2–4].
The pathophysiology of POAF is multifactorial, involving surgical trauma, systemic inflammation, autonomic imbalance, and electrolyte disturbances [5, 6]. Management typically involves rate control or rhythm control strategies [7]. Rate control focuses on stabilizing hemodynamics by managing ventricular response, while rhythm control aims to restore and maintain sinus rhythm, with potential benefits in atrioventricular synchrony and ventricular function demonstrated in computational models [8]. However, antiarrhythmic drugs may have significant adverse effects.
The optimal initial management strategy for POAF after CABG remains controversial. Major trials in non-surgical atrial fibrillation (e.g., AFFIRM) found no survival benefit for routine rhythm control, supporting rate control as the standard for chronic AF [9, 10]. However, the transient nature of POAF may limit the applicability of these findings [5]. Notably, a large randomized trial in POAF (CTSNet) found no significant difference in hospital stay or long-term mortality between rate and rhythm control [11], though interpretation is complicated by high crossover rates and the inclusion of low-risk patients, as noted in systematic reviews [12].
Current guidelines offer limited consensus on the preferred initial strategy for new-onset POAF after cardiac surgery [7]. While rhythm control is physiologically appealing, its superiority over rate control regarding clinical recovery and long-term prognosis remains unproven in the real-world setting. Using a large critical care database, this study investigated the association between an initial rhythm control strategy and clinical outcomes, including ICU length of stay and long-term survival, in patients with POAF after isolated CABG.
Methods
Study design and data source
This retrospective cohort study utilized data from the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.1) database. MIMIC-IV is a publicly available resource containing de-identified clinical data from patients admitted to the intensive care unit (ICU) at Beth Israel Deaconess Medical Center (BIDMC), Boston, Massachusetts, between 2008 and 2022 [13, 14]. The database was established with approval from the Institutional Review Boards (IRBs) of BIDMC and the Massachusetts Institute of Technology (MIT); the requirement for individual patient consent was waived due to the de-identified nature of the data. Access to the dataset was obtained by the author, Xianglin Long, following completion of the U.S. National Institutes of Health (NIH) online training course on human research participant protection (Certificate No. 68,822,776). This study is a retrospective observational analysis and does not meet the definition of a clinical trial; therefore, trial registration is not applicable.
Study population
We included adult patients (aged ≥ 18 years) who underwent isolated CABG and subsequently developed new-onset POAF during their index ICU stay. POAF was defined as the first-ever documented episode of atrial fibrillation or flutter after surgery. The diagnosis required confirmation by either a 12-lead electrocardiogram (ECG) or a continuous ECG monitoring strip revealing an irregularly irregular rhythm without discernible P waves, lasting for a minimum of 30 s. To ensure a true new-onset cohort, patients with any documented history of atrial fibrillation or flutter prior to the index surgery were excluded.
Inclusion criteria were receipt of either a rate control or rhythm control pharmacological strategy within 24 h of POAF onset. Exclusion criteria were: (1) pre-existing diagnosis of atrial fibrillation or flutter prior to hospital admission; (2) concomitant cardiac procedures, including valve surgery; (3) absence of rate or rhythm control therapy; or (4) multiple ICU admissions during the same hospitalization, with only the first ICU admission considered.
Exposure and group definition
Eligible patients were stratified into two groups based on the initial pharmacological strategy administered within 24 h of POAF onset. Treatment decisions were made at the discretion of the attending physician based on clinical judgment. The rate control group included patients who received one or more rate control agents (beta-blockers [excluding sotalol], non-dihydropyridine calcium channel blockers, or digoxin) without any rhythm control agents. The rhythm control group comprised patients who received at least one rhythm control agent (e.g., amiodarone, propafenone, dofetilide) in addition to rate control therapy. Details of the specific antiarrhythmic medications administered in both groups, before and after propensity score matching, are provided in Supplementary Table S1.
Outcomes
The primary outcome was successful conversion to and maintenance of sinus rhythm (SR) within 24 h of treatment initiation. A successful SR conversion event was defined as the first documentation of sinus rhythm on continuous ECG monitoring that was sustained for a minimum duration of one hour. This duration threshold was prespecified to differentiate stable rhythm restoration from transient, hemodynamically insignificant sinus episodes. Secondary outcomes included successful SR conversion at 6 and 12 h using the same definition, as well as the incidence of postoperative acute kidney injury (AKI), long-term mortality, and lengths of ICU and hospital stay.
Postoperative AKI was defined according to the Kidney Disease: Improving Global Outcomes (KDIGO) criteria within the first 7 days after surgery. For our primary analysis, we focused on the incidence of clinically significant AKI, defined as KDIGO stage 2 or 3. This outcome was chosen to capture moderate-to-severe kidney injury while excluding milder, often transient, stage 1 events [15].
Data collection
A comprehensive set of variables was extracted for all eligible patients from the MIMIC-IV database using structured query language (SQL) executed through Navicat Premium (version 17). Extracted data included baseline demographics (age, gender, body mass index (BMI)), comorbidity history (diabetes, congestive heart failure, myocardial infarction, chronic pulmonary disease, and renal disease) identified by International Classification of Diseases, Ninth and Tenth Revision (ICD-9/10) codes, and severity of illness at ICU admission measured by the Sequential Organ Failure Assessment (SOFA) score. Baseline laboratory values (white blood cell count, creatinine, potassium) and information on critical care interventions, including use of vasoactive drugs at POAF onset, were also collected.
Statistical analysis
Handling of missing data
The completeness of baseline covariate data was evaluated prior to analysis (Supplementary Table S2). Most variables were fully observed. For covariates with missing values, primarily laboratory results such as calcium (12.52% missing) and BMI (6.79% missing), multiple imputation by chained equations (MICE) was applied under the missing-at-random assumption. Ten imputed datasets were generated. All subsequent analyses, including propensity score estimation and matching, were performed on each imputed dataset, with final results pooled according to Rubin’s Rules to account for imputation-related uncertainty.
Descriptive statistics and baseline comparisons
Baseline characteristics and clinical outcomes were compared between treatment groups. Continuous variables were summarized as mean ± standard deviation (SD) for normally distributed data, or as median [interquartile range (IQR)] for non-normally distributed data, and compared using the student’s t-test or Wilcoxon test, respectively. Categorical variables were reported as counts (percentages) and compared using the chi-squared test or Fisher’s exact test, as appropriate.
Propensity score matching
To address selection bias and confounding in this observational study, propensity score matching (PSM) was performed. Propensity scores, defined as the probability of receiving the rhythm control strategy, were estimated for each patient using a multivariable logistic regression model that included baseline covariates (see Supplementary Table S3). A 1:1 nearest-neighbor matching algorithm was applied, using a caliper width of 0.1 standard deviations of the logit of the propensity score. Covariate balance before and after matching was assessed with standardized mean differences (SMDs); an absolute SMD less than 0.1 was considered to indicate negligible imbalance. All primary analyses were subsequently conducted on this matched cohort.
Sensitivity analysis
To assess the robustness of our findings, we performed a sensitivity analysis with inverse probability of treatment weighting (IPTW). Stabilized weights were calculated for each patient based on a propensity score model identical to that used in the primary analysis. Covariate balance after weighting was confirmed by ensuring all absolute SMDs were below 0.1. All primary and secondary outcomes were then re-analyzed in the weighted cohort using the appropriate weighted regression models (e.g., logistic, Cox, or general linear models).
Outcome analysis
All primary and secondary outcomes were assessed in the propensity score-matched cohort. For binary outcomes such as sinus rhythm conversion, odds ratios (ORs) were estimated using logistic regression. Time-to-event outcomes were analyzed with Kaplan-Meier curves and the log-rank test, while hazard ratios (HRs) were calculated using Cox proportional hazards models. Given that baseline covariates were balanced after matching, no further adjustment was performed in these models. The same analytical methods were applied to the IPTW-weighted cohort for the sensitivity analysis.
Subgroup and interaction analyses
Pre-specified subgroup analyses were conducted to explore potential treatment effect modification. Within each stratum of a subgroup variable, we estimated the OR for sinus rhythm conversion. To formally assess for heterogeneity, a P-value for the interaction between the treatment strategy and the subgroup variable was derived from a logistic regression model using a likelihood ratio test. These analyses were primarily performed on the propensity score-matched cohort. To maximize statistical power and ensure the robustness of any observed interaction, the analysis was then repeated on the IPTW-weighted full cohort as a sensitivity analysis.
All statistical analyses were performed using R version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria). A two-sided P-value < 0.05 was considered statistically significant. Proportional hazards assumptions for the Cox regression analyses were formally evaluated using Schoenfeld residuals, and no violations were detected (all global P > 0.05; Supplementary Table S4, Supplementary Figure S1).
Results
Patient population and baseline characteristics
The study inclusion and exclusion process is illustrated in the flowchart (Fig. 1). From an initial cohort of 94,458 patients in the MIMIC-IV ICU database, 743 patients developed new-onset POAF after CABG surgery. After applying exclusion criteria, 663 patients remained in the pre-matched cohort, comprising 101 patients in the rate control group and 562 patients in the rhythm control group. Following 1:1 PSM, the final analytic cohort consisted of 198 patients, with 99 patients in each group.
Fig. 1.
Flowchart of patient selection. The flowchart details the sequential application of inclusion and exclusion criteria to derive the final study cohort from the MIMIC-IV ICU database. Abbreviations: CABG, coronary artery bypass grafting; POAF, postoperative atrial fibrillation; AF, atrial fibrillation
In the unmatched cohort, baseline characteristics differed between groups, particularly age, with patients in the rate control group being older (median age 78.06 vs. 74.10 years; P = 0.003) (Table 1). After PSM, baseline covariates were well-balanced between the groups, with almost all absolute SMDs less than 0.1 (Table 2, Supplementary Table S5, and Figure S2). The distribution of propensity scores before and after PSM is shown in Supplementary Figure S3. Importantly, the baseline characteristics of the matched rhythm control cohort (99 patients) were highly comparable to those of the original unmatched pool (562 patients), with negligible differences observed across almost all covariates (Supplementary Table S6).
Table 1.
Baseline characteristics before PSM
| Level | Overall (N = 663) |
rate group (N = 101) |
rhythm group (N = 562) |
P-value | |
|---|---|---|---|---|---|
| Gender n (%) | Female | 190 (28.66) | 36 (35.64) | 154 (27.40) | 0.117 |
| Male | 473 (71.34) | 65 (64.36) | 408 (72.60) | ||
| Race n (%) | Asian | 13 (1.96) | 2 (1.98) | 11 (1.96) | 0.864 |
| Black | 15 (2.26) | 3 (2.97) | 12 (2.14) | ||
| Other | 108 (16.29) | 14 (13.86) | 94 (16.73) | ||
| White | 527 (79.49) | 82 (81.19) | 445 (79.18) | ||
| Age | 74.35 [67.57, 80.76] | 78.06 [71.03, 82.69] | 74.10 [66.86, 80.51] | 0.003 | |
| BMI | 28.40 [25.08, 32.17] | 27.73 [24.06, 31.35] | 28.56 [25.29, 32.27] | 0.164 | |
| SOFA | 5.00 [4.00, 8.00] | 6.00 [4.00, 8.00] | 5.00 [4.00, 8.00] | 0.139 | |
| Past history | |||||
| MI n (%) | No | 438 (66.06) | 70 (69.31) | 368 (65.48) | 0.526 |
| Yes | 225 (33.94) | 31 (30.69) | 194 (34.52) | ||
| CHF n (%) | No | 405 (61.09) | 56 (55.45) | 349 (62.10) | 0.249 |
| Yes | 258 (38.91) | 45 (44.55) | 213 (37.90) | ||
| CPD n (%) | No | 488 (73.60) | 71 (70.30) | 417 (74.20) | 0.486 |
| Yes | 175 (26.40) | 30 (29.70) | 145 (25.80) | ||
| Renal disease n (%) | No | 481 (72.55) | 67 (66.34) | 414 (73.67) | 0.162 |
| Yes | 182 (27.45) | 34 (33.66) | 148 (26.33) | ||
|
Malignant cancer n (%) |
No | 637 (96.08) | 101 (100.00) | 536 (95.37) | 0.054 |
| Yes | 26 (3.92) | 0 (0.00) | 26 (4.63) | ||
| Diabetes n (%) | No | 391 (58.97) | 53 (52.48) | 338 (60.14) | 0.183 |
| Yes | 272 (41.03) | 48 (47.52) | 224 (39.86) | ||
| Medication | |||||
| Vasoactive drug n (%) | No | 84 (12.67) | 12 (11.88) | 72 (12.81) | 0.923 |
| Yes | 579 (87.33) | 89 (88.12) | 490 (87.19) | ||
| Laboratory indicators | |||||
| Hemoglobin | 8.80 [7.90, 9.90] | 8.50 [7.70, 9.60] | 8.80 [7.90, 10.00] | 0.165 | |
| Platelets | 135.00 [107.00, 174.00] | 128.00 [99.00, 163.00] | 136.50 [108.00, 175.75] | 0.048 | |
| WBC | 15.00 [11.80, 19.20] | 14.80 [12.30, 19.80] | 15.05 [11.72, 19.20] | 0.777 | |
| Aniongap | 13.00 [11.00, 15.00] | 13.00 [12.00, 15.00] | 13.00 [11.00, 15.00] | 0.306 | |
| Bicarbonate | 22.00 [21.00, 24.00] | 22.00 [20.00, 24.00] | 22.00 [21.00, 24.00] | 0.216 | |
| Calcium | 8.30 [8.00, 8.70] | 8.30 [8.10, 8.70] | 8.30 [7.90, 8.80] | 0.46 | |
| Chloride | 110.00 [107.00, 112.00] | 110.00 [107.00, 112.00] | 110.00 [107.00, 112.00] | 0.879 | |
| Creatinine | 1.00 [0.80, 1.30] | 1.10 [0.80, 1.50] | 1.00 [0.80, 1.30] | 0.432 | |
| Sodium | 137.00 [135.00, 139.00] | 137.00 [136.00, 139.00] | 137.00 [135.00, 139.00] | 0.834 | |
| Potassium | 4.60 [4.30, 4.90] | 4.50 [4.30, 4.90] | 4.60 [4.30, 4.90] | 0.862 | |
| INR | 1.40 [1.30, 1.60] | 1.50 [1.40, 1.70] | 1.40 [1.30, 1.60] | 0.007 | |
Table 2.
Baseline characteristics after PSM
| Level | Overall (N = 198) |
rate group (N = 99) |
rhythm group (N = 99) |
P-value | |
|---|---|---|---|---|---|
| Gender n (%) | Female | 68 (34.34) | 35 (35.35) | 33 (33.33) | 0.881 |
| Male | 130 (65.66) | 64 (64.65) | 66 (66.67) | ||
| Race n (%) | Asian | 4 (2.02) | 2 (2.02) | 2 (2.02) | 0.608 |
| Black | 5 (2.53) | 3 (3.03) | 2 (2.02) | ||
| Other | 35 (17.68) | 14 (14.14) | 21 (21.21) | ||
| White | 154 (77.78) | 80 (80.81) | 74 (74.75) | ||
| Age | 76.82 [70.54, 82.67] | 77.72 [70.98, 82.66] | 76.76 [68.98, 82.66] | 0.418 | |
| BMI | 28.06 [24.55, 32.04] | 27.89 [24.32, 31.46] | 28.41 [24.64, 32.11] | 0.576 | |
| SOFA | 6.00 [4.00, 8.00] | 6.00 [4.00, 8.00] | 6.00 [4.00, 8.00] | 0.69 | |
| Past history | |||||
| MI n (%) | No | 140 (70.71) | 69 (69.70) | 71 (71.72) | 0.876 |
| Yes | 58 (29.29) | 30 (30.30) | 28 (28.28) | ||
| CHF n (%) | No | 112 (56.57) | 56 (56.57) | 56 (56.57) | 1 |
| Yes | 86 (43.43) | 43 (43.43) | 43 (43.43) | ||
| CPD n (%) | No | 139 (70.20) | 69 (69.70) | 70 (70.71) | 1 |
| Yes | 59 (29.80) | 30 (30.30) | 29 (29.29) | ||
| Renal disease n (%) | No | 135 (68.18) | 67 (67.68) | 68 (68.69) | 1 |
| Yes | 63 (31.82) | 32 (32.32) | 31 (31.31) | ||
|
Malignant cancer n (%) |
No | 196 (98.99) | 99 (100.00) | 97 (97.98) | 0.477 |
| Yes | 2 (1.01) | 0 (0.00) | 2 (2.02) | ||
| Diabetes n (%) | No | 103 (52.02) | 53 (53.54) | 50 (50.51) | 0.776 |
| Yes | 95 (47.98) | 46 (46.46) | 49 (49.49) | ||
| Medication | |||||
| Vasoactive drug n (%) | No | 25 (12.63) | 12 (12.12) | 13 (13.13) | 1 |
| Yes | 173 (87.37) | 87 (87.88) | 86 (86.87) | ||
| Laboratory indicators | |||||
| Hemoglobin | 8.50 [7.70, 9.67] | 8.60 [7.75, 9.65] | 8.40 [7.60, 9.65] | 0.448 | |
| Platelets | 129.00 [103.00, 163.00] | 129.00 [98.50, 163.50] | 129.00 [107.00, 161.50] | 0.736 | |
| WBC | 14.55 [11.53, 19.03] | 14.80 [12.20, 18.90] | 14.50 [10.75, 19.00] | 0.645 | |
| Aniongap | 13.00 [11.25, 15.00] | 13.00 [12.00, 15.00] | 13.00 [11.00, 15.00] | 0.922 | |
| Bicarbonate | 22.00 [20.00, 24.00] | 22.00 [20.00, 24.00] | 22.00 [21.00, 24.00] | 0.37 | |
| Calcium | 8.25 [7.93, 8.70] | 8.30 [8.00, 8.70] | 8.20 [7.90, 8.70] | 0.466 | |
| Chloride | 110.00 [107.00, 112.00] | 110.00 [107.00, 112.00] | 111.00 [107.00, 112.00] | 0.584 | |
| Creatinine | 1.00 [0.80, 1.40] | 1.10 [0.80, 1.45] | 1.00 [0.80, 1.30] | 0.803 | |
| Sodium | 137.00 [136.00, 139.00] | 137.00 [136.00, 139.00] | 138.00 [136.00, 139.50] | 0.206 | |
| Potassium | 4.60 [4.30, 4.90] | 4.50 [4.30, 4.90] | 4.60 [4.30, 4.90] | 0.989 | |
| INR | 1.50 [1.30, 1.60] | 1.50 [1.40, 1.70] | 1.50 [1.30, 1.60] | 0.277 | |
In the matched cohort, beta-blockers were the primary rate control agents, while amiodarone was the predominant agent in the rhythm control group; all patients in the rhythm control group received amiodarone. A detailed breakdown of antiarrhythmic drug administration in both unmatched and matched cohorts is provided in Supplementary Table S1.
Primary and secondary outcomes
The primary and secondary outcomes for the propensity score-matched cohort are detailed in Table 3. The primary outcome, SR conversion within 24 h, was achieved by significantly more patients in the rhythm control group compared to the rate control group (71.7% vs. 32.3%; OR, 2.65; 95% CI, 1.19–4.07; P = 0.010). Consistent with this finding, the rhythm control group also demonstrated significantly higher rates of SR conversion at 12 h (63.6% vs. 20.2%; P = 0.007) and at 6 h (47.5% vs. 17.2%; P = 0.032).
Table 3.
The association of rate or rhythm control medications with outcomes in the matched cohort
| Outcome | rate group (N = 99) | rhythm group (N = 99) | effect estimate (95% CI) | P-value |
|---|---|---|---|---|
| Primary outcome | ||||
| SR Conversion < 24 h, n (%) | 32 (32.3%) | 71 (71.7%) | OR 2.65 (1.19–4.07) | 0.010 |
| Secondary Outcome | ||||
| Sinus Rhythm Conversion | ||||
| SR Conversion < 12 h, n (%) | 20 (20.2%) | 63 (63.6%) | OR 2.45 (1.30–4.60) | 0.007 |
| SR Conversion < 6 h, n (%) | 17 (17.2%) | 47 (47.5%) | OR 2.10 (1.08–3.95) | 0.032 |
| Postoperative Complications | ||||
| Postoperative AKI, n (%) | 53 (53.5%) | 48 (48.5%) | OR 0.82 (0.47–1.43) | 0.477 |
| Mortality | ||||
| 1-year Mortality, n (%) | 11 (11.1%) | 9 (9.1%) | HR 1.19 (0.40–3.55) | 0.760 |
| 2-year Mortality, n (%) | 13 (13.1%) | 11 (11.1%) | HR 0.78 (0.35–1.73) | 0.535 |
| 3-year Mortality, n (%) | 15 (15.2%) | 13 (13.1%) | HR 0.76 (0.36–1.61) | 0.481 |
| Length of Stay | ||||
| Length of hospital stay, day | 8.79 [6.07, 14.46] | 10.32 [7.25, 15.01] | Median difference 0.67 (-0.80, 2.06) | 0.066 |
| Length of ICU stay, day | 3.08 [2.19, 4.21] | 3.42 [2.22, 5.76] | Median difference 0.37 (0.08, 1.02) | 0.026 |
Regarding major postoperative complications, there was no significant difference in the incidence of postoperative AKI between the two groups (48.5% vs. 53.5%; OR, 0.82; 95% CI, 0.47–1.43; P = 0.477). In terms of long-term survival, no significant differences were observed between the two strategies at 1-year, 2-year, or 3-year follow-up (Fig. 2). Similarly, no survival difference was found in the pre-matched cohort (Supplementary Figure S4).
Fig. 2.
Kaplan-Meier survival analysis for all-cause mortality. Kaplan-Meier curves comparing all-cause mortality between the Rate group and the Rhythm group in the matched cohort. Panels show survival analysis censored at A 1 year (365 days), B 2 years (730 days), and C 3 years (1095 days). P-values were calculated using the log-rank test
Patients in the rhythm control group had a significantly longer median length of ICU stay compared to those in the rate control group (3.42 vs. 3.08 days; P = 0.026), while the difference in total hospital length of stay did not reach statistical significance (10.32 vs. 8.79 days; P = 0.066). To explore this further, we stratified these outcomes by diabetes status and found that the trend of a longer ICU stay in the rhythm control group was present in patients both with and without diabetes (Supplementary Table S7).
To assess the robustness of these findings, a sensitivity analysis using IPTW was performed. The results of the IPTW analysis were consistent with the primary analysis, confirming a higher likelihood of SR conversion but no significant differences in long-term mortality or AKI with the rhythm control strategy. Notably, the association with a longer ICU stay also remained significant in the weighted analysis (Supplementary Table S8). After weighting, all baseline covariates were well-balanced (Supplementary Table S5 and Figure S5).
Subgroup analysis of sinus rhythm conversion
We performed pre-specified subgroup analyses to explore treatment effect heterogeneity for sinus rhythm conversion at 24 h. In the primary PSM analysis, significant interactions were observed for both diabetes status and SOFA score (Fig. 3). The rhythm control strategy was associated with significantly higher odds of sinus rhythm conversion in patients without diabetes (OR, 2.64; 95% CI, 1.36–3.71), whereas no significant benefit was observed in patients with diabetes (OR, 1.21; 95% CI, 0.39–2.86) (P for interaction = 0.041). A similar pattern of interaction was observed for patients stratified by SOFA score: the benefit of the rhythm control strategy was more pronounced in patients with a SOFA score < 7 (OR, 2.64; 95% CI, 1.44–3.96) compared to those with a SOFA score ≥ 7 (OR, 1.62; 95% CI, 0.74–3.05) (P for interaction = 0.043). No significant interactions were identified for other pre-specified subgroups.
Fig. 3.
Subgroup analysis of sinus rhythm conversion within 24 h. Forest plot displaying the odds ratios (ORs) and 95% confidence intervals (CIs) for the primary outcome in pre-specified subgroups. The analysis compares the Rhythm group to the Rate group (reference). P for interaction was calculated using a likelihood ratio test to assess for heterogeneity of treatment effect
To assess the robustness of these findings, a sensitivity analysis using IPTW was performed (Supplementary Figure S6). The interaction with diabetes status remained significant and consistent with the primary analysis (P for interaction = 0.044). However, the interaction with SOFA score was no longer statistically significant in the weighted analysis (P for interaction = 0.137), although the direction of the effect was similar. No significant interactions were observed for any subgroup at the 6-hour or 12-hour time points in either analysis (Supplementary Figures S6 and S7).
Discussion
In this retrospective analysis of patients with new-onset POAF following isolated CABG, we observed that an initial rhythm control strategy was associated with significantly higher rates of early sinus rhythm restoration compared with rate control. This comparison conceptually represents the escalation to rhythm control therapy (frequently in combination with baseline rate-limiting agents) versus rate control alone. Nevertheless, this physiological success did not translate into improved clinical outcomes. Instead, the rhythm control strategy was associated with a longer ICU length of stay without conferring any detectable benefit in long-term survival. These findings were consistent across both PSM and IPTW analyses, as summarized in Fig. 4.
Fig. 4.
Graphical summary of the study design and key findings
The observed higher rate of SR conversion is consistent with the underlying mechanism of action. The use of a rhythm control agent, primarily amiodarone, offers a direct pharmacological approach to terminating arrhythmia, in line with existing evidence supporting amiodarone for both prophylaxis and treatment of POAF [16, 17]. Nonetheless, the clinical implications of early SR restoration remain complex. First, the preoperative use of rate controlling medications, such as beta-blockers or calcium channel blockers, may modulate postoperative baseline heart rates and potentially influence the subsequent response to pharmacological cardioversion. Second, our primary endpoint of 24-hour conversion is relatively narrow and may not reflect durable rhythm stabilization, potentially overestimating the practical clinical utility of immediate rhythm control. Although computational studies, such as that by Barrows et al. [8], predict that restoring sinus rhythm and atrioventricular synchrony improves left ventricular function, our empirical findings suggest that these physiological benefits may not be robust enough to translate into superior hard clinical outcomes in the immediate postoperative phase.
The dissociation between successful early rhythm conversion and long-term mortality observed in this study is consistent with existing high-level evidence, including the CTSNet randomized trial and a recent meta-analysis aggregating both randomized and real-world observational cohorts [11, 12]. This lack of statistical difference supports the hypothesis that POAF may serve primarily as an indicator of pre-existing patient vulnerability rather than as a direct, modifiable driver of long-term mortality [18]. Indeed, the perioperative course after CABG is heavily influenced by systemic inflammatory burdens [19, 20]. This is highlighted by contemporary prognostic models like the Hematological Inflammatory Gradient Score (HIGS), which demonstrate that systemic inflammatory states significantly predict early mortality and adverse outcomes after coronary bypass surgery [21]. Interventions targeting the arrhythmia alone, without addressing these upstream inflammatory substrates or general frailty, are likely insufficient to alter long-term survival trajectories. Furthermore, because of the relatively low number of events and the resulting wide confidence intervals after matching, our study was underpowered to detect subtle survival differences. Therefore, the absence of statistical significance should not be misconstrued as definitive evidence of clinical equivalence between the two strategies.
The finding of a longer ICU stay in the rhythm control group, despite a higher rate of SR conversion, is a critical and paradoxical observation. This may be partly attributable to increased monitoring requirements for antiarrhythmic drugs. Furthermore, it may reflect the direct physiological effects of the agents themselves, predominantly intravenous amiodarone, which is frequently associated with acute adverse events such as hypotension necessitating vasopressor titration, or drug-induced bradycardia [22]. Additionally, while we employed rigorous statistical adjustments, we cannot rule out the influence of confounding by indication, where clinicians selectively escalated sicker or more unstable patients to rhythm control, a limitation inherent in retrospective data. Ultimately, this dissociation between physiological rhythm restoration and expedited recovery underscores that for many patients, POAF may be a marker of systemic stress rather than the primary driver of their post-operative clinical course.
A noteworthy observation of this study is the heterogeneity in treatment effect for the primary outcome. The therapeutic benefit of the rhythm control strategy on sinus rhythm conversion was notably greater in patients without diabetes and in those with lower illness severity (SOFA < 7). The interaction with diabetes proved to be a robust finding, as it remained statistically significant in the IPTW sensitivity analysis. While the interaction with SOFA score was no longer significant in the IPTW analysis, the direction of the effect was consistent across both methods. However, these subgroup findings must be interpreted with caution. Multiple subgroup comparisons inherently elevate the risk of Type I error, and the relatively small size of our matched subgroups means these analyses are potentially underpowered and should be regarded as exploratory and hypothesis-generating rather than definitive. Mechanistically, conditions such as diabetes and severe critical illness are known to promote adverse atrial remodeling and systemic inflammation, which may render patients less responsive to pharmacological cardioversion [23, 24]. Collectively, these findings advocate for a more personalized, substrate-guided approach to POAF management, potentially reserving an aggressive rhythm control strategy for lower-risk patients.
Limitation
This study has several limitations.
First, as a retrospective, single-center study, it is susceptible to selection bias and confounding by indication. Although PSM and IPTW were utilized, they cannot fully account for unmeasured confounding because the MIMIC-IV database lacks structured records on preoperative outpatient medication history (such as chronic beta-blocker or calcium channel blocker use), several critical perioperative variables (e.g., preoperative left ventricular ejection fraction, left atrial size, hemodynamic instability severity, and temporary pacing dependency) as well as intraoperative parameters (e.g., on-pump versus off-pump status, cardiopulmonary bypass duration, and cross-clamp time). Consequently, our findings should be interpreted as clinical associations rather than definitive causal effects.
Second, the timing of treatment initiation varied within the first 24 h. While the exposure window was strictly defined, patients in the rate control group did not require therapy escalation, potentially reflecting a milder baseline clinical trajectory. Additionally, the precise delay from POAF onset to treatment initiation was not captured, limiting our ability to analyze the impact of earlier intervention.
Third, our primary endpoint of sinus rhythm conversion within 24 h is limited in scope. A transient conversion sustained for only one hour does not guarantee durable clinical stabilization or reflect long-term rhythm status at discharge, which may overestimate the practical clinical benefit of rhythm control.
Fourth, several factors constrain the generalizability and external validity of our findings. The substantial baseline imbalance between groups required a 1:1 matching ratio that discarded 82.4% of the rhythm control cohort (463 patients), meaning the matched pairs may not fully represent the overall population. Furthermore, the 2008–2022 database span introduces potential temporal confounding as practice guidelines and protocols evolved. The near-exclusive use of amiodarone (98%-100%) also reflects the localized practice patterns of a single institution (Beth Israel Deaconess Medical Center) rather than current global standards. Therefore, our findings require prospective validation in external, multicenter cohorts.
Conclusion
In patients with new-onset POAF following CABG, an initial rhythm control strategy was associated with higher rates of early sinus rhythm restoration and longer ICU stays, but showed no long-term survival benefit. While rhythm control remains a valid physiological intervention, our findings do not support its routine prioritization over rate control for improving clinical prognosis; furthermore, this lack of survival difference should not be interpreted as clinical equivalence. Further randomized trials are required to determine the optimal management strategy.
Supplementary Information
Acknowledgements
We thank the Home for Researchers editorial team (www.home-for-researchers.com) for their language editing service. Figures 1 and 4 were created in https://BioRender.com.
Declaration of generative AI and AI-assisted technologies in the writing process
During the preparation of this work, the authors used ChatGPT (OpenAI) to perform language polishing and improve the readability of specific sections of the manuscript, namely the Introduction and Discussion. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the final content of the publication.
Authors’ contributions
Xianglin Long: Methodology, Software, Writing - Original Draft, Investigation. Weiwu Xiong: Investigation. Yeshen Zhang and Lixiong Zeng: Validation, Visualization. Xiaoyan Wang: Validation, Writing - Review and Editing. Zhihui Zhang: Resources, Writing - Review and Editing, Funding acquisition. All authors contributed to the article and approved the submitted version.
Funding
This work was supported by the National Natural Science Foundation of China (Grant No. 82570448 to Zhihui Zhang).
Data availability
The datasets generated and analyzed in this study are available at the MIMIC-IV website: https://physionet.org/content/mimiciv/3.1/.
Code availability
The code that supports the findings of this study is available from the corresponding author upon reasonable request.
Declarations
Ethics approval and consent to participate
This study was conducted in accordance with the guidelines outlined in the Declaration of Helsinki. Ethical approval for the utilization of the MIMIC-IV database was secured from the Institutional Review Boards (IRBs) of both the Massachusetts Institute of Technology (MIT) and Beth Israel Deaconess Medical Center (BIDMC). The previously granted ethical approval for the MIMIC database encompasses the data employed in this research, thereby eliminating the requirement for additional ethical approval or informed consent.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Contributor Information
Xiaoyan Wang, Email: 747626363@qq.com.
Zhihui Zhang, Email: zhangzhihui0869@csu.edu.cn.
References
- 1.Eikelboom R, Sanjanwala R, Le M-L, Yamashita MH, Arora RC. Postoperative Atrial Fibrillation After Cardiac Surgery: A Systematic Review and Meta-Analysis. Ann Thorac Surg. 2021;111(2):544–54. 10.1016/j.athoracsur.2020.05.104. [DOI] [PubMed] [Google Scholar]
- 2.Kawczynski MJ, van der Heijden CAJ, Maessen JG, et al. Early postoperative atrial fibrillation is associated with late mortality after cardiac surgery: a systematic review and reconstructed individual patient data meta-analysis. J Cardiothorac Surg. 2025;20(1):265. 10.1186/s13019-025-03504-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Iwasaki Y, Ohbe H, Nakajima M, et al. Association Between Intraoperative Landiolol Use and In-Hospital Mortality After Coronary Artery Bypass Grafting: A Nationwide Observational Study in Japan. Anesth Analg. 2023;137(6):1208–15. 10.1213/ANE.0000000000006741. [DOI] [PubMed] [Google Scholar]
- 4.Xu H, Lv P, Li B, Zhong P, Tian P, Wu L. Histidine and the risk of incident atrial fibrillation and postablation recurrence: Evidence from 2 prospective cohorts. Cardiac Res. 2026;2(1):40–50. 10.1097/re9.0000000000000010. [DOI] [Google Scholar]
- 5.Shah S, Chahil V, Battisha A, Haq S, Kalra DK. Postoperative Atrial Fibrillation: A Review. Biomedicines. 2024;12(9). 10.3390/biomedicines12091968. [DOI] [PMC free article] [PubMed]
- 6.Spicer S, Hasheminia A, Kandi S, Abu-Omar Y, Fernández ÁL, El-Diasty M. Cardiac and pericardial inflammatory changes and post-cardiac surgery atrial fibrillation. Cytokine Growth Factor Rev. 2025. 10.1016/j.cytogfr.2025.07.006. [DOI] [PubMed] [Google Scholar]
- 7.Van Gelder IC, Rienstra M, Bunting KV, et al. 2024 ESC Guidelines for the management of atrial fibrillation developed in collaboration with the European Association for Cardio-Thoracic Surgery (EACTS). Eur Heart J. 2024;45(36):3314–414. 10.1093/eurheartj/ehae176. [DOI] [PubMed] [Google Scholar]
- 8.Barrows RK, Augustin CM, Gsell MAF, et al. Rhythm control benefits left ventricular function compared with rate control in patients with atrial fibrillation: A computational study. Heart Rhythm O2. 2025;6(7):1011–9. 10.1016/j.hroo.2025.04.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Van Gelder IC, Hagens VE, Bosker HA, et al. A comparison of rate control and rhythm control in patients with recurrent persistent atrial fibrillation. N Engl J Med. 2002;347(23):1834–40. [DOI] [PubMed] [Google Scholar]
- 10.Wyse DG, Waldo AL, DiMarco JP, et al. A comparison of rate control and rhythm control in patients with atrial fibrillation. N Engl J Med. 2002;347(23):1825–33. [DOI] [PubMed] [Google Scholar]
- 11.Gillinov AM, Bagiella E, Moskowitz AJ, et al. Rate Control versus Rhythm Control for Atrial Fibrillation after Cardiac Surgery. N Engl J Med. 2016;374(20):1911–21. 10.1056/NEJMoa1602002. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Ahmed M, Belley-Coté EP, Qiu Y, et al. Rhythm vs. Rate Control in Patients with Postoperative Atrial Fibrillation after Cardiac Surgery: A Systematic Review and Meta-Analysis. J Clin Med. 2023;12(13). 10.3390/jcm12134534. [DOI] [PMC free article] [PubMed]
- 13.Johnson AEW, Bulgarelli L, Shen L, et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci Data. 2023;10(1):1. 10.1038/s41597-022-01899-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Goldberger AL, Amaral LA, Glass L, et al. PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals. Circulation. 2000;101(23):E215–20. [DOI] [PubMed] [Google Scholar]
- 15.Xiong C, Jia Y, Wu X, et al. Early Postoperative Acetaminophen Administration and Severe Acute Kidney Injury After Cardiac Surgery. Am J Kidney Dis. 2023;81(6). 10.1053/j.ajkd.2022.11.009. [DOI] [PubMed]
- 16.Gill J, Shah AG, Di Luozzo G, et al. Amiodarone Prophylaxis against postoperative atrial fibrillation in off-pump coronary artery bypass. Heart Lung. 2025;72:85–94. 10.1016/j.hrtlng.2025.03.009. [DOI] [PubMed] [Google Scholar]
- 17.Suero OR, Ali AK, Barron LR, Segar MW, Moon MR, Chatterjee S. Postoperative atrial fibrillation (POAF) after cardiac surgery: clinical practice review. J Thorac Dis. 2024;16(2):1503–20. 10.21037/jtd-23-1626. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Dobrev D, Aguilar M, Heijman J, Guichard J-B, Nattel S. Postoperative atrial fibrillation: mechanisms, manifestations and management. Nat Rev Cardiol. 2019;16(7):417–36. 10.1038/s41569-019-0166-5. [DOI] [PubMed] [Google Scholar]
- 19.Heo RH, Wang MK, Meyre PB, et al. Associations of Inflammatory Biomarkers With the Risk of Morbidity and Mortality After Cardiac Surgery: A Systematic Review and Meta-analysis. Can J Cardiol. 2023;39(11):1686–94. 10.1016/j.cjca.2023.07.021. [DOI] [PubMed] [Google Scholar]
- 20.Vervoort D, Caldonazo T, Doenst T, Fremes SE. Inflammatory Biomarkers and Cardiac Surgical Risk: Hitting the Mark? Can J Cardiol. 2023;39(11):1695–7. 10.1016/j.cjca.2023.09.004. [DOI] [PubMed] [Google Scholar]
- 21.Soylemez N, Karaca O, Toprak B, Bora R, Bilgic A, Yilmaz S. Hematological Inflammatory Gradient Score (HIGS): A Novel Predictor of Early Mortality After Coronary Artery Bypass Grafting. J Inflamm Res. 2025;18:17495–508. 10.2147/JIR.S571285. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Yang J, You M, Wang J, et al. Adverse events in different administration routes of amiodarone: a pharmacovigilance study based on the FDA adverse event reporting system. Front Pharmacol. 2025;16:1517616. 10.3389/fphar.2025.1517616. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Kantharia BK, Tabary M, Wu L, et al. Diabetes and Atrial Fibrillation: Insight From Basic to Translational Science Into the Mechanisms and Management. J Cardiovasc Electrophysiol. 2025. 10.1111/jce.16655. [DOI] [PubMed] [Google Scholar]
- 24.Weng C, Lin J, Liu Q, et al. Comparison of rhythm and rate control medications for new-onset atrial fibrillation in septic patients: MIMIC-IV database analysis. J Transl Med. 2025;23(1):512. 10.1186/s12967-025-06380-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets generated and analyzed in this study are available at the MIMIC-IV website: https://physionet.org/content/mimiciv/3.1/.
The code that supports the findings of this study is available from the corresponding author upon reasonable request.