We retrospectively analysed collected data from the National Adult Cardiac Surgery Audit (NACSA), obtained from the National Institute of Cardiovascular Outcomes Research’s (NICOR) central cardiac database. The definitions of the database variables used for this study can be found at https://www.nicor.org.uk/national-cardiac-audit-programme/datasets/ in the NACSA Dataset v5.2. The NACSA registry prospectively collects demographic and pre-, peri-, and post-operative clinical data for all significant adult cardiac surgery procedures performed in the United Kingdom. Its central role is to benchmark surgical practice. The flow of the data from data input to analysis has been previously described (7). The data are entered locally and validated at the unit level by database managers before being uploaded through a web portal to NICOR.

Further validation is performed according to logical rules, and missing data reports are generated for primary variables (e.g., EuroSCORE risk factors, patient identifiers, and outcome data). The data are then forwarded to an academic healthcare informatics department for data cleaning. Duplicate records are removed, transcriptional discrepancies are re-coded, and clinical and temporal conflicts are resolved. Missing data are determined during the validation stages of the data transfer from individual centres. Missing and conflicting data for in-hospital mortality status are backfilled and validated via record linkage to the Office for National Statistics (ONS) census database. For the current study, missingness in the outcome variables in the matched population is as follows: mortality 1.3%, need for post-op dialysis 6.9%, post-operative CVA 9.2%, and length of hospital stay 0.3%. Missing data were handled by exclusion. The Health Research Authority (HRA) and Health and Care Research Wales (HCRW) approved the study in 2020 (IRAS project ID 257758), and a waiver for patients' consent was obtained. This study was conducted in accordance with the Declaration of Helsinki.

We included patients who had isolated CABG and had non-dialysis-dependent severe renal impairment pre-operatively defined as a creatinine clearance of less than 50 mL/min calculated using the Cockcroft–Gault equation (8) between 18 August 1998 and 31 March 2019. Figure 1 depicts the subsetting of the patient groups included in the analysis. We excluded re-do CABG and patients with end-stage renal failure on dialysis since this is a highly comorbid patient population that is the focus of another analysis. We also excluded patients with moderate renal impairment (creatinine clearance of less than 50 mL/min) since the previous evidence suggested the most pronounced benefit of OPCABG in this spectrum of patients.

The primary outcome was in-hospital mortality. Secondary outcomes included in-hospital transient ischaemic stroke (TIA), cerebrovascular accident (CVA), post-operative renal dialysis, return to theatre for bleeding or cardiac tamponade, and length of hospital stay.

Categorical variables were summarised as counts and percentages and compared using Pearson's χ² test or Fisher's exact test. A Shapiro–Wilk test was used to assess the normality of the distribution of continuous data. Our continuous data were non-normally distributed, were summarised as a median with an interquartile range (IQR), and analysed using the Wilcoxon rank-sum test. The P-values were adjusted for multiple tests using the Bonferroni method.

Propensity score matching was performed using 1:1 nearest neighbor matching without replacement using the following baseline characteristics from the NACSA dataset: age, female sex, pre-operative AF, pre-operative neurological dysfunction, hypertension, body mass index (BMI), pre-operative creatinine, recent myocardial infarction, pulmonary disease, Canadian Cardiovascular Society (CCS) angina grade (0–4), New York Heart Association Heart Classification (1–4), pulmonary HTN, diabetes (diet, oral medications, or insulin), left ventricular impairment (good, moderate, and poor), peripheral vascular disease, and the urgency of surgery (elective, urgent, emergency, and salvage surgery) (Table 1).

After matching, all standardised mean differences (SMDs) for the covariates were checked using love plots, and the adequate balance was set to be below 0.1 (Figure 2). Matched variables were compared using a paired t-test for continuos data and a McNemar test for binary data (9).

To estimate the on-pump CABG effect vs. off-pump CABG and its standard error, we fitted a logistic regression model with mortality, stroke, post-op renal dialysis, and return to theatre for bleeding as binary outcomes and the treatment (on-pump CABG) and matching covariates as predictors. We included the full matching weights in the estimation. The “glm ()” function was used to fit the outcome, and the comparisons() function in the “marginal effects” package was used to perform g-computation in the matched sample to estimate the average treatment effect on patients undergoing on-pump CABG. Finally, a cluster-robust variance was used to estimate standard errors with matching stratum membership as the clustering variable (10–12). We used a composite endpoint between CVA and TIA to estimate the effect of ONCABG vs. OPCABG on stroke rates.

The baseline characteristics of the population we analysed are summarised in Table 2. The median age was 78 years (35% female) for the OPCABG cohort and 77 years (37% female) for the ONCABG cohort (both P's > 0.9). The proportion of patients with very poor (1.3% vs. 0.3%), poor (3.8% vs. 2.8%), and moderate LV function (20% vs. 17%) was significantly higher amongst ONCABG patients (P = 0.01) compared to OPCABG patients. More patients with peripheral vascular disease were in the OPCABG group than in the ONCABG group (24% vs. 20%, P = 0.018). There were no differences in the remaining variables we used for matching (Table 1). The percentage of patients with one graft and two grafts was higher in the OPCABG group (12% vs. 1.8% and 12% vs. 1.8%, respectively, P < 0.001). Conversely, the percentage of patients receiving three and four grafts was higher in the ONCABG group (55% vs. 39% and 20% vs. 11%, respectively, P < 0.001). A very small proportion of patients underwent five grafts or more, which was higher in the OPCABG cohort than in the ONCABG cohort (4.2% vs. 2.3%, P = 0.004).

There was no difference in mortality, post-operative dialysis, CVA, TIA, or return to theatre for bleeding or tamponade (all P’s > 0.005). However, the total length of hospital stay was significantly higher by 1 day in the ONCABG cohort compared to that in the OPCABG cohort (12 vs. 11 days, P = 0.002).

After matching, there was no imbalance between the baseline variables (Table 1 and Figure 1). The percentage of patients with one and two grafts remained higher in the OPCABG cohort than in the ONCABG cohort, and the percentage with three and four grafts remained higher in the ONCABG cohort (P < 0.001).

After matching, we found no difference in in-hospital mortality (4.1% vs. 3.8%, P = 0.7), need for post-operative dialysis (6.8% vs. 8.7%, P = 0.1), CVA (1.4% vs. 0.7%, P = 0.3), or TIA (0.7% vs. 0.6%, P > 0.9) between the ONCABG and OPCABG cohorts. However, return to theatre bleeding rates were higher in the ONCABG cohort than in the OPCABG cohort (5.9% vs. 3.9%, P = 0.008), and the length of hospital stay in the ONCABG cohort remained higher compared to that in the OPCABG cohort (12 vs. 11, P = 0.008).

There was no effect of ONCABG vs. OPCABG on mortality (OR, 1.05, P = 0.78), post-operative stroke (OR, 1.7, P = 0.12,) or dialysis (OR, 0.7, P = 0.09); however, ONCABG was associated with an increased risk of bleeding (OR, 1.53, P = 0.03) Table 3.

The strength of our study is its reliance on a multicentre registry dataset that reflects real-world practice. The large size of the dataset allowed us to perform robust propensity matching and double adjustment using clustered-robust standard error regression. Nevertheless, residual confounding can remain despite the methods employed and the retrospective nature of our data, which will never equal a randomised controlled trial.

We found fewer grafts performed in the OPCABG group, consistent with well-known findings in the literature, including studies of patients with pre-operative renal dysfunction (6, 17). It would have been useful to assess the completeness of revascularisation and use these data for matching or in the prediction model. However, one of the study's limitations was high missingness and inaccuracy in pre-operative coronary disease that could not allow us to assess for incomplete revascularisation. Another limitation of our study was the lack of long-term survival data that prevented us from assessing whether the number of grafts impacts the long-term survival of patients with chronic renal dysfunction requiring CABG, which is another limitation of our study. Finally, the study spans a long period when significant improvements in cardiopulmonary bypass techniques have occurred, thus impacting any differences between the two techniques.