We systematically searched the following databases for RCTs indexed through March 2023: PubMed, Embase, Cochrane Library (CENTRAL), CINAL (via EBSCO), and Web of Science. Search strings contained the terms coronary artery bypass [as Medical Subject Heading], coronary artery bypass*, coronary artery bypass surgery, aortocoronary bypass*, coronary artery bypass grafting, CABG, myocardial revascularization, vascular grafting, coronary artery bypass graft; AND breathing exercises [as Medical Subject Heading], breathing exercise*, respiratory muscle training, inspiratory muscle training, expiratory muscle training, inspiratory muscle strength, respiratory exercise, inspiratory muscle train*, respiratory train, ventilatory train, breathing train, respiratory therapy, IMT, RMT (Supplementary Appendix 1). In addition, we manually searched the reference lists of relevant studies.

To be included in this review and meta-analysis, studies had to (1) examine patients at least 18 years old who underwent CABG either pre- or postoperatively; (2) apply a randomized controlled design in which the IMT arm was compared to an arm that received sham IMT, physical therapy, or usual care; (3) report data on at least one of pneumonia, pleural effusion, or atelectasis; and (4) the language of the publication was not limited. Studies were excluded if the full text was unavailable or if they were reviews or observational studies.

Two authors independently extracted the following data using a data extraction form developed a priori: (1) study characteristics, including author name, title, comparison arms, and year of publication; (2) population characteristics, including age, sex, and sample size; (3) details of interventions, including training type, frequency, session duration, and intensity; and (4) outcomes. Primary outcomes were rates of pneumonia, pleural effusion, and atelectasis. Secondary outcomes were maximum inspiratory pressure (MIP), maximum expiratory pressure (MEP), Pm_peak/Pi_max, and length of hospitalization (LOS). Other outcomes such as exercise capacity were meta-analyzed when relevant data were reported. If outcome data were unclear or not reported, we contacted the authors in an attempt to obtain missing data.

Two authors independently assessed the risk of bias for included studies using the RoB 2.0 tools for randomized trials and included the following domains: randomization/allocation process, deviation from intended intervention, missing outcome data, outcome measurement, and selective outcome reporting. Visualization of RoB 2.0 was produced using robvis. Studies were judged to be at low, high, or unclear risk of bias. Similarly, discordance was dealt with by adjudication among the authors until consensus was reached by following the appropriate algorithms.

Statistical analyses were conducted using RevMan 5.3 software. Statistical heterogeneity in pooled results was assessed using the chi-squared test, Cochran's Q-test, and the inconsistency I² test, in which I² values of 25%, 50%, or 75% were considered cut-offs to indicate low, moderate, or high heterogeneity, respectively (22). We performed random-effect heterogeneity in the event of moderate or high heterogeneity. Where appropriate, results were reported as relative risk (RR) and associated 95% confidence interval (CI). The mean difference (MD) was used as the effect size if studies used the same tool to measure the outcome. We used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) system to assess the quality of the body of evidence associated with the following specific outcomes in our review and constructed a summary of findings for the main outcomes: pneumonia, atelectasis, and pleural effusion. This assessment considers the study methodological quality, directness of the evidence, heterogeneity of the data, precision of the effect estimates, and the risk of publication bias.

Database search yielded 1,390 potentially eligible articles, while manual searching of reference lists did not identify additional studies. After excluding 440 duplicates and 926 studies based on their titles and abstracts, the full text of 24 publications was examined, leading to the exclusion of 16 studies (Figure 1). In the end, a total of eight studies (23–30) met the inclusion criteria and were utilized for meta-analysis (Table 1).

These studies, each of which enrolled 20–276 patients, involved a total of 755 patients. Studies examined pre- and postoperative IMT, which usually involved the threshold IMT, and involved an intensity at 30%–60% of MIP once to twice daily for durations ranging from 3 days to 4 weeks (Table 2). Most studies described the details of random sequence generation, blinding during outcome assessment, incomplete outcome data, selective reporting, and other biases (Figure 2). However, few studies described details of allocation concealment or blinding of participants and investigators.

Figure 2 shows the results of the risk of bias assessment.

We noted a low heterogeneity between the trials, and small-study effects may be present. So we downgraded this outcome for inconsistency and publication bias. Eventually, we rated the quality of evidence as very low quality and the effect is uncertain.

Seven studies involving 738 participants evaluated the effect of IMT on MIP (23–25, 27–30), and it was significantly better in the IMT group. Similar results were obtained in separate meta-analyses of the four studies involving preoperative IMT (MD 16.55 cmH₂O, 95% CI 13.86–19.24, P < 0.00001); or the remaining three studies involving postoperative IMT (MD 8.99 cmH₂O, 95% CI: 2.39–15.60, P = 0.008) (Figure 4A).

Meta-analysis of three studies (27, 29, 30) involving 144 patients associated postoperative IMT with significantly better MEP (MD 7.15 cmH₂O, 95% CI 1.52–12.79, P = 0.01; Figure 4B). Both meta-analyses were performed with fixed-effect models because heterogeneity was negligible (I² = 0%). Two trials (23, 25) reported the data about respiratory muscle endurance, which they assessed in terms of the ratio of Pm_peak to Pi_max (Table 3), but data heterogeneity were too high to permit meta-analysis (I²= 90%, P = 0.018).

Five studies (24, 25, 27–29) compared length of hospitalization between the IMT and control groups, but one (25) of them reported respective median and ranges [7 (5–41) vs. 8 (6–70) days], which we could not pool with the means and standard deviations reported in the other four trials. Meta-analysis of those four trials associated IMT with significantly shorter hospitalization (−1.71 days, 95% CI −2.56 to −0.87 days, P < 0.0001; Figure 5).

Three studies (26, 29, 30) examined the impact of IMT on performance in the 6 min walk test (6MWT). Due to the high heterogeneity (I²= 92%, P < 0.00001), a random effects model was used for pooled analysis. The results showed that IMT was not associated with significantly longer distance traveled (MD 60.03 m, 95% CI −9.01 to 129.08, P = 0.09; Figure 6). Due to the small sample size, subgroup analysis and meta-regression cannot be conducted.

Subsets of the eight RCTs reported on additional secondary outcomes. One trial (28) with 235 participants found no significant difference in postoperative quality of life between IMT and control groups, as measured using the Form 36 questionnaire or the EuroQol questionnaire with five dimensions and three levels. One trial (28) reported no significant difference between IMT and control groups in how much percentage of inspiratory vital capacity (IVC%), forced vital capacity (FVC%), and forced expiration volume in 1 s (FEV₁%) changed as a result of CABG (Table 3). Similarly, another study (27) reported improvement between IMT and control groups in tidal volume (P = 0.049) and vital capacity (P = 0.022). Meta-analysis of two studies (27, 29) involving 85 patients found that IMT did not significantly affect the peak expiratory flow (MD 54.36 L/min, 95% CI −5.85 to 114.58, P = 0.08; Figure 7).

The limitations of our systematic review and meta-analysis should also be noted. First, although the overall analysis suggests that IMT can reduce pneumonia and atelectasis, further studies are needed to examine IMT on different types of PPCs, including pleural effusion, mechanical ventilation more than 48 h, and pneumothorax. Additionally, preoperative and postoperative intervention period were included, and subgroup analysis was not performed, except MIP. Second, in the secondary outcome, the quality of life and pulmonary function (FVC, FEV₁) data cannot be completely extracted. Thus, meta-analysis was not done. Third, this review did not cover cost–benefit and safety analyses. Fourth, due to the intervention equipment, sessions and intensity were inconsistent, and subgroup analysis was not performed. More well-designed large research studies are still needed.