Our results show that high BNP is an important predictor of early and late postoperative complications. The OR adjusted for postoperative atrial fibrillation was 3.8 (95% CI 1.45–10.38). Several risk factors are associated with the occurrence of atrial fibrillation after cardiac surgery; they include mainly old age, low ejection fraction, and heart valve surgery. The onset of this arrhythmia is most likely a consequence of hemodynamic, electrical, and histological atrial tissue abnormalities related to intraoperative changes [20, 21] and is associated with a higher incidence of early complications, such as congestive heart failure, stroke, renal dysfunction, infections, and neurocognitive impairment [22]. Postoperative atrial fibrillation increases hospital stays and health care costs [22, 23], so strategies to predict its occurrence may have important clinical and economic relevance by intensifying preoperative medical treatment, enabling the use of perioperative intensive prophylaxis schemes, or eventually leading to the use of intraoperative ablation techniques. These advancements could decrease the incidence of adverse events in this high-risk group and reduce hospital stay, costs, and morbidity associated with atrial fibrillation [24].
The incidence of atrial fibrillation in our group was 11.2%, which is not as high as that reported in other series [22–24]; however, several subgroups at high risk for this complication had to be excluded. During their stay at general ward hospitalization it is also possible that some patients had transient episodes of atrial fibrillation that were not documented. This lack of detection, however, would suggest that these events did not have clinical significance.
The risk attributable to elevated BNP (cutoff for the top quintile: 258 pg/mL) in our group is very similar to that reported in the Cleveland Clinic series [25] that included 187 patients undergoing cardiac surgery, whose records were retrospectively evaluated, revealing an OR of 3.7 when comparing the highest with the lowest quartile. Our data also show that preoperative BNP measurement is valuable in predicting a significant increase in the probability of postoperative low cardiac output: OR 3.46 (95% CI 1.53–7.80); in turn, this complication is closely related to the overall morbidity and mortality [26]. The ability to anticipate the need for prolonged postoperative myocardial inotropy has important clinical implications, as the measurement of BNP can be used to identify patients who require intensive preoperative medical therapy. Additionally, a common problem in practice is to define the appropriate time to carry out cardiac surgery in patients who require complex procedures; perhaps BNP measurement could be useful in these patients. In this series, other unadjusted postoperative outcomes were also significantly related to high BNP: hospital stay, ICU stay, and number of transfused RBC units; each of these has been associated with increased postoperative morbidity and mortality. Our results open the possibilities of designing research studies that incorporate BNP measurement as a routine part of the preoperative evaluation and comparing this strategy with the standard evaluation, in terms of reducing postoperative adverse events.
Regarding late events, for the pre-specified primary outcome, a significant interaction was found between BNP and diabetes, with a significant increase in risk when BNP was elevated in diabetic subjects. This increase in risk among diabetic subjects with elevated BNP is most likely multifactorial in origin. Elevated BNP in diabetic subjects, with or without microalbuminuria, could represent the presence of an early stage of diabetic nephropathy, in which the level of creatinine is not yet affected [27, 28]. It is possible that even mild degrees of renal dysfunction significantly impact postoperative morbidity. In the Steno-2 study of 160 type 2 diabetic subjects with microalbuminuria, higher baseline N-terminal-proBNP was associated with longer duration of diabetes, older age, higher systolic blood pressure, and impaired kidney function [29]. In our series, although diabetics tended to have higher morbidity and mortality at 12 months (primary outcome variable: 31.5% in diabetics vs. 29.1% in non-diabetics), this difference was not statistically significant. However, subjects with diabetes had other conditions associated with risk: older age (63.9 years vs. 59.0 years, p = 0.0001) and lower ejection fraction (50.0% vs. 52.6%, p = 0.017). These data could indicate that the subgroup of diabetic patients with elevated BNP have a higher comorbidity, which might explain our results. Moreover, diabetic heart disease is associated with a range of morphological changes, including myocyte hypertrophy, perivascular fibrosis, and accumulation of extracellular matrix amorphous protein. Over time, these processes contribute to the development of left ventricular hypertrophy, coronary artery disease, and congestive heart failure, each of which is associated with elevated BNP. Even if the ejection fraction is normal, occult cardiomyopathy characterized by impaired relaxation and a stiff left ventricle may precede manifestations of heart disease in patients with diabetes [30].
Finally, the hidden causes of nephropathy and more advanced heart disease may not be the only reason elevated BNP is associated with a significant increase in postoperative risk; other factors may include hypertension, increased extracellular volume, and pulmonary hypertension. Because all of these can be considered markers of cardiac risk, it is clear that the non-specific elevation of BNP can provide a useful indication of overall cardiovascular risk in this population.
Our data do not confirm that inflammatory markers were significantly and independently associated with the assessed outcomes; for hsCRP, the composite outcome had an OR of 1.53 (95% CI 0.80–2.91). For the same outcome, LEUCO had an OR of 1.41 (95% CI 0.78–2.55).
In a retrospective review of 720 patients undergoing cardiac surgery, where a measure of hsCRP was available, Cappabianca et al. identified that CRP ≥ 0.5 mg/dL conferred a higher risk of in-hospital mortality and postoperative infections [31]. Van der Harst, after evaluating retrospectively a subgroup of 87 patients with coronary revascularization, found that in patients with hsCRP above the median (1.9 mg/L), the cumulative incidence of cardiovascular events with a follow-up at 7.3 years was 29%, compared with 9% in patients with levels below the median (p = 0.048), independent of other risk factors [32]. However, Gaudino et al., in a prospective study of 114 patients undergoing myocardial revascularization surgery, found that CRP > 5 mg/L did not predict in-hospital postoperative complications or influence the extent of inflammatory activation [33]. In a larger series, Biancari et al. reported that preoperative CRP ≥ 1.0 mg/dL carried a higher overall risk of global postoperative death (5.3% vs. 1.1%, p = 0.001), cardiac death (4.4% vs. 0.8%, p = 0.002), and low cardiac output syndrome (8.8% vs. 3.7%, p = 0.01) (47). Ahlsson [34] found no association between preoperative CRP and postoperative atrial fibrillation.
The discrepant findings in the above results could suggest that multiple underlying risk factors are involved in the associations between adverse postoperative outcomes and hsCRP level. Insufficient adjustment for confounding variables related to the load of comorbidities may lead to an overestimation of risk. In our series, we carefully controlled for confounders by excluding active infectious and inflammatory or neoplastic processes. Furthermore, in multivariate analysis, age, gender, ejection fraction, diabetes mellitus, recent ACS, obesity, previous cardiac surgery, using or not using cardiopulmonary bypass, and type of surgery were controlled. It is possible that the relationship between preoperative and postoperative outcomes of hsCRP can largely be explained by multiple underlying factors that elevate hsCRP and increase surgical risk. However, because the ORs that we found for most of the outcomes were close to 1.5 for hsCRP throughout the follow-up, it is also possible that the sample size was insufficient to determine a real association.
Lower preoperative Hb was associated with increased postoperative adverse events in a Hb dose-dependent manner. A lower preoperative Hb (below 13.2 g/dL) was associated with increased in-hospital mortality and at 12 months and more likely to require inotropic drugs and suffer acute renal failure, cerebrovascular events, high-grade ventricular arrhythmias, and require RBC transfusion in the postoperative period. The primary composite outcome was also more frequently observed in the presence of lower Hb. Some pre-specified outcomes occurred entirely or almost entirely in the lowest hemoglobin quintiles (in-hospital mortality, acute renal failure, cerebrovascular events, and ventricular tachycardia); given their rarity and the fact that outcomes were concentrated in only one or two quintiles, we not considered it appropriate to test for statistical significance.
The effect of low Hb on most of the outcomes was attenuated after adjustment for the described covariates. However, the independent relationship between the lowest Hb values and postoperative low cardiac output was not affected by adjusting for covariates, yielding an OR of 0.33 (95% CI 0.13–0.81). A significant association was found between low preoperative Hb and major postoperative outcomes: hospital stay, ICU stay, and number of units of RBCs transfused. As mentioned above, these variables are associated with postoperative morbidity and mortality.
Reduced Hb may contribute to worse outcomes through a higher peripheral and myocardial oxygen demand and development of left ventricular hypertrophy, mainly due to a secondary increased cardiac output [35]. An inverse relationship between Hb value and left ventricular hypertrophy has been reported in clinical studies of patients with chronic kidney disease [36]. In a subgroup of patients with chronic heart failure, an increase of 1 g/dL in Hb was associated with a decrease of 4.1 g/m2 in left ventricular mass in a 24-week follow-up [37]. These changes could explain the increased risk of arrhythmias and the occurrence of low cardiac output in our patients with low Hb.
Our results show that the need for RBC transfusion in the perioperative period is directly related to the value of preoperative Hb, even in the absence of a defined anemia; subjects with Hb < 13.2 g/dL required an average transfusion of 3.1 units of RBCs, while if hemoglobin was > 16.1 g/dL, the required RBC units dropped to 0.6. One study showed that both preoperative anemia and intraoperative transfusion of red cells were independent and additive factors for adverse outcomes. Patients with low preoperative Hb had a higher incidence of postoperative adverse events, but for the same level of Hb, the risk of postoperative complications increased significantly with RBC transfusion. In addition, they observed a direct relationship between the number of RBC units transfused intraoperatively and the incidence of adverse events [38]. This independent association between RBC transfusion and adverse outcomes has been repeatedly described [39–44].
In this series, the fact that polycythemic subjects were at higher risk of postoperative adverse events could not be confirmed. However, this is physiologically plausible, and the number of subjects with this feature in the sample was low (16 patients); thus, it cannot be concluded that there was no such association.
Our findings open the possibility of designing randomized studies to assess if the optimization of preoperative Hb concentration, through the use of erythropoietin and/or iron therapy, would reduce perioperative transfusion and the incidence of ventricular arrhythmia, acute renal failure, and postoperative low cardiac output.
Limitations
Although the sample size was calculated to detect an increase in risk by a factor of 2 for the variable Hb, the study may have been underpowered to detect the prognostic significance of hsCRP and LEUCO, most likely because inflammatory mechanisms are a weaker predictor of outcomes than Hb or BNP. The ORs found were slightly above 1.5; multicenter studies on a larger scale would be needed to achieve the required precision.
The incidence of atrial fibrillation may have been underestimated because continuous electrocardiogram monitoring is not routinely performed in postoperative patients after they are discharged from the ICU. Therefore, some patients may have had transient episodes of During their stay at general ward hospitalization. This lack of detection, however, would suggest that these events did not have clinical significance.
This study required multiple statistical analyses to assess the independent effects of four biomarkers on several postoperative outcomes. However, this does not increase the possibility of a type I error because the analysis of the data sets for each independent variable was done separately and all hypotheses were pre-specified.