Background and objectives

Download PDF

Even today, cardiac surgery is associated with a significant risk of morbidity and mortality. Overall, the in-hospital mortality of patients undergoing myocardial revascularization surgery is 2.5%, and 22% have reported major morbidity events.1 Risk stratification plays an important role to estimate the outcome of cardiac surgery. The logistic EuroSCORE (logES) is one of the best risk models, and it estimates the postoperative mortality considering patient and procedural factors.2 However, this score overestimates the mortality, and the information needed for its calculation is not always available; EuroSCORE II, which is the more recent version, is currently being validated in several regions of the world.3

An association has been established between the preoperative hemoglobin or B-type natriuretic peptide (BNP) values and several adverse outcomes in cardiac surgery, especially atrial fibrillation (AF), postoperative low cardiac output (PLCO) and rehospitalization in the first year, with this risk being increased in patients with diabetes mellitus.4 However, proving statistical independence is not enough to conclude clinical utility in risk prediction. Additional studies are required to evaluate whether hemoglobin or BNP measurements could replace or provide additional prognostic information to existing methods, in terms of improving the prediction of postoperative adverse outcomes.

BNP may have advantages over logES (or may add value to logES) as a prognostic indicator; it is a simple, inexpensive and reproducible test, and it could improve the discriminatory power of logES when incorporating the preoperative hemoglobin levels. If so, the inclusion of these variables in the risk stratification schemes could reduce the morbidity and mortality.


The study was approved by the Research Ethics Committee of the Instituto de Cardiología-Fundaciòn Cardioinfantil. Informed consent was obtained from all patients. Five hundred and fourteen adults undergoing cardiac surgery were included (Figure. 01). Patients undergoing a cardiac transplantation, pulmonary thromboendarterectomy or transcatheter aortic valve implantation were excluded (because these procedures were not used for the original development of logES).

Can the Predictive Power of Euroscore Be Improved by Adding Preoperative Hemoglobin or B-type Natriuretic Peptide (BNP) Values

Figure. 01 Flowchart

The objectives of the study were the following: 1) To determine whether the predictive power of logES in cardiac surgery would be improved with the addition of hemoglobin and/or BNP variables and 2) To determine whether BNP alone or in combination with hemoglobin would be a better predictor of postoperative complications than logES (both objectives were assessed in the entire cohort and in the subgroup of diabetic patients). All data were collected prospectively using standardized definitions. The BNP plasma concentration was determined before surgery.

The primary endpoint of the study was the evaluation of the in-hospital and 6-month mortality. The morbidity and mortality outcomes were evaluated in the following two time periods:

In-hospital events:

Reoperation, death, PLCO, acute renal failure, cerebrovascular event, AF, acute myocardial infarction and prolonged stay in the intensive care unit (PSICU) or hospital.

Events at 6 months:

Death or rehospitalization, acute coronary syndrome, cerebrovascular event and further revascularization or valve repair procedures.

Data Analysis:

The statistical analysis focused on assessing the extent to which the inclusion of the hemoglobin and/or BNP variables in logES had an added effect in terms of a better discrimination between the subjects with and without outcomes. A preliminary step was the direct comparison between the predictive power of hemoglobin and/or BNP and logES. For each of the outcomes, receiver operating characteristic (ROC) curves were constructed, as follows:

AUC (Area Under the Curve) (*) of logES


AUC of Hemoglobin

AUC of BNP + Hemoglobin

AUC of logES + BNP

AUC of logES + Hemoglobin

AUC of logES + BNP + Hemoglobin

(*) 95% confidence interval (CI) in all of them

The ROC-AUC was used to determine the discriminatory power of logES. To construct the ROC curves in which hemoglobin or BNP should be incorporated into logES, a risk coefficient was calculated for hemoglobinand BNP, which was obtained after evaluating each of these variables by univariate analysis for each of the desired outcomes. The coefficients for each of the variables that constitute logES are known;5 thus, once the hazard ratios were calculated for hemoglobin and BNP (for each outcome), these variables were included in the sequence and for each analysis of the logES model. The Hosmer-Lemeshow test was used to assess the extent to which the predicted probability for the occurrence of the outcomes coincided with that observed during the follow-up and the behavior of the model (any of the possible AUC listed) in all possible ranges of risk. The Hanley-McNeil formula was used to calculate the sample size.


Four hundred and ninety-two (492) patients were included. Tables. 01 and 02 show the baseline characteristics, presented for patients who survived or not, at hospital discharge and at 6 months; the univariate relationship between these characteristics and mortality is also shown.

Table. 01

Characteristics of patients and hospital outcomes.

CHARACTERISTIC All 492 (100%) Survivors 474 (96.34%) Non-survivors 18 (3.66% HR (95% CI) P-value
Age (Mean, SD) 64.1 (11.6) 63.9 (11.6) 69.6 (11.4) 1.051.00-1.10 0.039
Women n (%) 142 (28.8) 136 (28.6) 6 (33.3) 1.230.471-3.21 0.67
Diabetes mellitus n (%) 143 (29.07) 136(28.69) 7(38.89) 1.550.61-3.92 0.35
Arterial hypertension n (%) 351 (71.3) 336 (70.89) 15 (83.33) 20.59-6.83 0.26
Chronic pulmonary obstructive disease n (%) 41 (8.3) 39 (8.2) 2 (11.1) 1.370.32-5.77 0.66
Endocarditis n (%) 8 (1.6) 5 (1.05) 3 (16.6) 12.14.34-33.6 0.0001
Hemoglobin (Mean, SD) 14.16 (2.24) 14.26 (2.19) 11.6 (2.08) 0.610.49-0.751 0.0001
BNP (Mean, SD) 227.02 (388.07) 221.54(385.14) 371.38(446.82) 1.000.99-1.00 0.10
Creatinine > 2.3 n (%) Creatinine > 2.3 n (%) 18 (3.8) 4 (22.2) 6.12.18-17.02 0.001
Extracardiacarteriopathy n (%) 23 (4.6) 20 (4.2) 3 (16.6) 4.071.26-13.09 0.018
Neurological dysfunction n (%) 5 (1.02) 4 (0.84) 1 (5.56) 5.720.93-35.15 0.059
Previous cardiac surgery n (%) 39 (7.93) 35 (7.38) 4 (22.22) 3.311.14-9.59 0.027
Unstable angina n (%) 5 (1.02) 4 (0.84) 1 (5.56) 5.720.93-35.15 0.059
Moderate LV dysfunction n (%) 223 (45.33) 214 (45.15) 9 (50.00) 1.200.48-2.98 0.68
Severe LV dysfunction n (%) 36 (7.32) 34 (7.17) 2 (11.11) 1.580.37-6.61 0.52
Recent myocardial infarction n (%) 255 (51.83) 247 (52.1) 8 (44.4) 0.740.29–1.85 0.52
Pulmonary hypertension. n (%) 41 (8.33) 37 (7.81) 4 (22.22) 3.141.08-9.10 0.035
Surgery, other than revascularization n (%) 168 (34.15) 155 (32.7) 13 (72.22) 5.011.81-13.82 0.002
Critical preoperative state 15 (3.05) 11 (2.32) 4 (22.2) 9.083.39-24.33 0.0001
Thoracic aorta surgery 44 (8.94) 42 (8.86) 2 (11.11) 1.270.30-5.35 0.74
Renal replacement therapy (dialysis) 17 (3.46) 13 (2.74) 4 (22.22) 7.982.93-21.70 0.0001

The cohort included mainly men (71.2%), with a mean age of 64.1 ± 11.6 years. Most patients (65.8%) underwent an isolated myocardial revascularization; the other patients are distributed as shown in Table. 03. A complete follow-up was obtained for 99% of patients (Figure. 01).

Table. 02

Characteristics of patients and outcomes at 6 months

Characteristic All 4 92 (100%) Survivors 459 (93.29%) Non-survivors 33 (6.71%) HR (95% CI) P-value
Age (Mean, SD) 64.1 (11.6) 63.6 (11.5) 70.1 (10.7) 1061.02-1.10 0.002
Women n (%) 142 (28.8) 131 (28.5) 11 (33.3) 1230.47-3.21 0.67
Diabetes mellitus n (%) 143 (29.07) 131 (28.5) 12 (36.3) 1550.61-3.92 0.35
Arterial hypertension n (%) 351 (71.3) 324 (70.5) 27 (81.8) 200.59-6.83 0.26
Chronic pulmonary obstructive disease n (%) 41 (8.3) 38 (8.2) 3 (9.0) 1370.32-5.77 0.66
Endocarditis n (%) 8 (1.6) 5 (1.1) 3 (9.1) 1214.3-33.6 0.0001
HB (Mean, SD) 14.16 (2.24) 1429(2.18) 1244(2.40) 0710.62-0.82 0.0001
BNP (Mean, SD) 227.02 (388.07) 22275 (SD: 390.15) 286.38 (SD: 358.26) 1000.99-1.00 0.36
Creatinine > 2.3 n (%) 22 (4.47) 17 (3.7) 5 (15.1) 612.18-17.02 0.001
Extracardiacarteriopathy n (%) 23 (4.6) 18 (3.9) 5 (15.1) 401.26-13.097 0.018
Neurological dysfunction n (%) 5 (1.02) 4 (0.8) 1 (3.0) 5720.93-35.15 0.059
Previous cardiac surgery n (%) 39 (7.93) 34 (7.4) 5 (15.1) 3311.14-9.59 0.027
Unstable angina n (%) 5 (1.02) 4 (0.8) 1 (3.0) 5720.93-35.15 0.059
Moderate LV dysfunction n (%) 223 (45.33) 208 (45.3) 15 (45.4) 1200.48-2.98 0.68
Severe LV dysfunction n (%) 36 (7.32) 33 (7.1) 3 (9.0) 1580.37-6.61 0.52
Recent myocardial infarction n (%) 255 (51.83) 238 (51.8) 17 (51.5) 0740.29-185 0.52
Pulmonary hypertension n (%) 41 (8.33) 37 (8.0) 4 (12.1) 3141.08-9.10 0.035
Surgery, other than revascularization n (%) 168 (34.15) 152 (33.1) 16 (48.4) 5-011.81-1382 0.002
Critical preoperative state 15 (3.05) 10 (2.18) 5 (15.15) 9083.39-24.33 0.0001
Thoracic aorta surgery 44 (8.94) 41 (8.93) 3 (9.09) 1270.30-5.35 0.74
Renal replacement therapy (dialysis) 17 (3.46) 13 (2.83) 4 (12.12) 7982.93-21.70 0.0001

Table. 03

Distribution of surgical procedures

Myocardial revascularization 324 65.8
Aortic valve replacement + other (not myocardial revascularization) 59 11.9
Myocardial revascularization + other 45 9.1
Aortic valve replacement 29 5.8
Procedures on mitral valve + other (not myocardial revascularization, not aortic replacement) 24 4.8
Procedures on isolated mitral valve 13 2.6
Other procedures 16 3.2

The mean BNP values were 221.54pg/mL (± 385.14) in survivors and 371.38 pg/mL (± 446.82) in non-survivors (odds ratio (OR): 1.00; 95% CI, 0.99-1.00; p=0.10); in patients undergoing an isolated coronary artery bypass surgery, the mean BNP values were 153.1 pg/mL (± 194.2), compared with 369.5 pg/mL (± 582.0) in other procedures. The same values for hemoglobin (HB) were 14.3 g/dL (± 2.2) in patients undergoing an isolated coronary artery bypass surgery and 13.7 g/dL (± 2.2) for other procedures.

The total mortality predicted with logES was 7.54% (± 9.3). The actual in-hospital mortality was 3.6%, and it was 6.7% at 6 months.

According to what has been described in the literature, the discriminatory power of the model is considered excellent if the ROC-AUC is > 0.80, very good if > 0.75 and good if > 0.70.6

logES showed an excellent discriminatory power in predicting the in-hospital mortality (ROC-AUC: 0.83; 95% CI, 0.75-0.92), and it was superior to BNP (ROC-AUC: 0.69; 95% CI, 0.59-0.79) and to hemoglobin (ROC-AUC: 0.25; 95% CI, 0.13-0.37); these differences were statistically significant. For the BNP + hemoglobin model, the ROC-AUC was 0.73 (95% CI, 0.62-0.85); the comparison against the ROC-AUC of logES alone showed no statistically significant differences. For the BNP + logESmodel, the ROC-AUC was 0.84 (95% CI, 0.75-92; no significant difference); for the hemoglobin + logES model, the ROC-AUC did not change (0.82; 95%CI, 0.70-0.94). Finally, the evaluation of both BNP and hemoglobin added to logES showed an ROC-AUC of 0.82 (95%CI, 0.70-0.94); no significant difference) (Figure. 02). The Hosmer-Lemeshow C statistic was 8.79 (p=0.36), indicating good calibration.

Figure. 02 Comparative ROC curves for the in-hospital mortality outcome, logES (gray line): area 0.83 vs. logES + BNP + hemoglobin (red line): area 0.83 (p=0.82). Reference (green line)

Figure. 02 Comparative ROC curves for the in-hospital mortality outcome, logES (gray line): area 0.83 vs. logES + BNP + hemoglobin (red line): area 0.83 (p=0.82). Reference (green line)

The PSICU was best predicted by logES alone, with an ROC-AUC of 0.72 (95% CI, 0.66-0.77). The addition of BNP, hemoglobin or their combination added no discriminatory power. BNP alone showed a good predictive power in discriminating the following two postoperative complications: PLCO and acute renal failure.In the first case, it showed a tendency to be better than all other models, with an ROC-AUC of 0.72 (95% CI, 0.67-0.77); the inclusion of BNP in logES or the combination with hemoglobin did not improve the discrimination. For PLCO, the cutoff showing the best diagnostic yield was 171 pg/dL. For acute renal failure, the ROC-AUC for BNP was 0.75 (95% CI, 0.76-0.84), with a cutoff of 186 pg/dL; the inclusion of BNP in logES improved the predictive power of the latter from 0.78 (95% CI, 0.71-0.85) to 0.80 (95% CI, 0.71-0.85), but with no significant difference.

The need for transfusion was best predicted by models that incorporated hemoglobin, i.e., logES + hemoglobin (0.71; 95% CI, 0.66-0.76), logES + BNP + hemoglobin (0.71; 95% CI, 0.66-0.75), or BNP + hemoglobin (0.70; 95% CI, 0.65-0.75), compared to logES alone (0.67; 95% CI, 0.62-0.71). Hemoglobin alone was not a good predictor of transfusion (0.30;95% CI, 0.25-0.35).

The predictive power of logES for the cumulative 6-month mortality was good (ROC-AUC: 0.73; 95% CI, 0.64-0.83). The inclusion of hemoglobin or BNP + hemoglobin in logES slightly improved its predictive power (ROC-AUC: 0.74; 95% CI, 0.64-0.83) for both models, but this variation was not significant, as shown in Figure. 03. The Hosmer-Lemeshow C statistic was 8.09 (p=0.42), indicating a good logES calibration for this outcome

Figure. 03 Comparative ROC curves for the 6-month mortality outcome, logES (gray line) area 0.73 vs. logES + BNP + hemoglobin (red line) area 0.74 (p=0.86). Reference (green line)

Figure. 03 Comparative ROC curves for the 6-month mortality outcome, logES (gray line) area 0.73 vs. logES + BNP + hemoglobin (red line) area 0.74 (p=0.86). Reference (green line)

The combined morbidity at 6 months was best predicted by logES alone (ROC-AUC: 0.70; 95% CI, 0.65-0.75). The addition of BNP, hemoglobin or their combination to the model did not significantly change its discriminatory capacity.

Subgroup of diabetic patients

The proportion of diabetic patients was 29.0% (n=143), with an in-hospital mortality of 4.9%, compared to 3.1% for the non-diabetic population (OR 1.55; 95% CI, 0.61-3.92; p=0.35).

The predictive power for the in-hospital mortality in this subgroup was excellent for logES alone (ROC-AUC: 0.95; 95% CI, 0.91-1.00). Hemoglobin, BNP or their combination had a lower predictive power; the addition of BNP to logES did not change its predictive power, as shown in Figure. 04. BNP, with a cutoff of 208 pg/dL, showed a sensitivity of 71% and a specificity of 70% for this outcome.

Figure. 04 Comparative ROC curves for the in-hospital mortality outcome, subgroup of diabetic patients, logES (gray line): area 0.95 vs. logES + BNP + hemoglobin (red line: area 0.84 (p=0.44). Reference (green line)

Figure. 04 Comparative ROC curves for the in-hospital mortality outcome, subgroup of diabetic patients, logES (gray line): area 0.95 vs. logES + BNP + hemoglobin (red line: area 0.84 (p=0.44). Reference (green line)

As in the overall sample, BNP discriminated the following two outcomes well: acute renal failure (ROC-AUC: 0.74; 95% CI, 0.59-0.89), with a cutoff of 400 pg/dL, and PLCO (ROC-AUC: 0.73;95% CI, 0.64-0.83), with a cutoff of 159 pg/dL. For both outcomes, there was no significant difference with logES.

The preoperative hemoglobin level does not predict low postoperative hemoglobin (< 10 g/dL;ROC-AUC: 0.25; 95% CI, 0.17-0.33) or the need for a transfusion (ROC-AUC: 0.28; 95% CI, 0.19-0.37). In predicting postoperative hemoglobin < 10 g/dL, the combination of BNP + hemoglobin was found to provide the best discrimination (ROC-AUC: 0.77; 95% CI, 0.69-0.84). Regarding the need for a transfusion, the best prediction was obtained by the combination of logES + hemoglobin + BNP, with an ROC-AUC of 0.71 (95% CI, 0.66-0.75).

The need for reoperation was best predicted by logES alone, whose predictive power was good, with an ROC-AUC of 0.74 (95% CI, 0.56-0.92). The addition of BNP or hemoglobin to logES did not change its predictive power.

Regarding the occurrence of complex ventricular arrhythmia in the postoperative period (tachycardia or ventricular fibrillation), the logES discrimination was good (ROC-AUC: 0.74; 95% CI, 0.51-0.96), improving with the addition of hemoglobin(ROC-AUC: 0.78; 95% CI, 0.57-0.99) or hemoglobin and BNP(ROC-AUC: 0.78; 95% CI, 0.57-0.99), but these were non-significant differences.

For outcomes such as AF, prolonged stay in the intensive care unit or in the hospital or readmission to an intensive care unit, none of the models showed a good predictive power.

During the events at 6 months, logES had an excellent predictive power for the mortality (AUC: 0.84; 95% CI,0.74-0.98) and cerebrovascular event in diabetic subjects (ROC-AUC: 0.80; 95% CI, 0.69-0.92); for the latter, the addition of BNP and hemoglobin increased the ROC-AUC to 0.82 (95% CI, 0.68-0.96), but with no significant difference, as shown in Figure. 05. The calibration obtained by the Hosmer-Lemeshow test for the 6-month mortality was good (5.19; p=0.73)

Figure. 05 Comparative ROC curves for the 6-month mortality outcome, subgroup of diabetic patients, logES (gray line): area 0.84 vs. logES + BNP + hemoglobin (red line): area 0.77 (p=0.43). Reference (green line)

Figure. 05 Comparative ROC curves for the 6-month mortality outcome, subgroup of diabetic patients, logES (gray line): area 0.84 vs. logES + BNP + hemoglobin (red line): area 0.77 (p=0.43). Reference (green line)


Neither BNP nor hemoglobin alone or in combination was a better predictor of the in-hospital or 6-month mortality than logES. Adding these variables to logES did not significantly change its predictive power. However, BNP discriminated well those patients who will present PLCO, with an ROC-AUC of 0.72. Because PLCO is an underlying problem of most postoperative complications, its prediction may help to establish specific preventive measures.

Attaran et al.7 reported an association between the BNP levels and postoperative complications, particularly the requirement of inotropic agents, renal dysfunction and increased mechanical ventilation time.

Natriuretic peptides have shown a moderate predictive reliability in cardiac surgery. 811 In a previous publication,4 we found an adjusted OR of 3.8 (95% CI, 1.45-10.38) for postoperative AF and of 3.46 (95% CI, 1.53-7.80) for PLCO in cardiac surgery patients with a preoperative BNP > 258 pg/dL.

In another study,12 BNP was an independent predictor of postoperative ventricular dysfunction (OR=1.92; 95% CI, 1.12-3.29), prolonged hospital stay (OR=1.42; 95% CI,1.18-1.72) and mortality (OR=1.89; 95% CI, 1.08-3.33). The authors classified BNP based on a value of 292 ng/L and found a hazard ratio (HR) of 1.89 (95% CI, 1.08-3.33) for the 5-year mortality; their analysis reported an ROC-AUC of 0.63, compared to 0.70 for the 3-year mortality in the study of Cuthbertson,13 which was also different from our results, where we found an ROC-AUC of 0.61 for the 6-month mortality in the entire cohort and 0.68 for the subgroup of diabetic patients.

The lack of a predictive superiority or equivalency compared to logES on the outcomes described does not imply the absence of association with the variables studied, as it only highlights the discriminatory power of logES.The risk variables defined in logES are well related to the mortality outcome, and introducing new variables that would improve its performance requires identifying independent factors associated with a higher risk. However, it is remarkable that a simple model consisting only of hemoglobin and BNP provides a good predictive power (ROC-AUC: 0.73; 95% CI, 0.62-0.85) for the in-hospital mortality.

PLCO is one of the complications that were best related to the overall morbidity and mortality. We found that the best predictive power for this complication was obtained by BNP alone, with an ROC-AUC of 0.72 (95% CI, 0.67-0.77). The presence of PLCO is the leading cause of mortality in emergency revascularization surgery;14 the incidence of mortality in this population has been reported to range from 38% to 52.8%.15 Substantial related morbidity has also been described, especially related to the increased incidence of pulmonary complications, prolonged respiratory mechanical support, PSICU, myocardial infarction, cerebrovascular event and renal failure. 16,17 Clearly, the identification of risk factors for developing PLCO is essential for optimizing preoperative management, defining cardioprotection strategies, providing early intraoperative hemodynamic support and considering the prophylactic use of balloon counterpulsation. In the Barros series, 14 some factors associated with this complication were identified, such as advanced age, emergency surgery, coronary revascularization and left ventricular dysfunction.

The occurrence of acute renal failure in the postoperative period is one of the strongest indicators of mortality in all reports.17 In our study, the best discrimination in predicting renal failure was obtained with the addition of BNP to the logES model, improving the ROC-AUC from 0.78 (0.71-0.85) to 0.80 (0.73-0.86), although this difference did not reach statistical significance, which was most likely due to the low number of events (37; 7.5%). Based on our data, the use of a cutoff point for a preoperative BNP of 186 pg/dL would obtain a sensitivity of 72% and a specificity of 69%, which are values that could provide additional information for the preoperative evaluation, allowing early follow-up and intervention in renal dysfunction.

The fact that the need for a transfusion is better discriminated by models that combine logES or BNP with hemoglobin, and not by preoperative hemoglobin alone, may indicate that the associated comorbidities or hemodynamic deterioration are more important as an indication of the need for a postoperative transfusion than the absolute value of hemoglobin, at least in our series. A practical consequence of this observation is that an “ideal” level of preoperative hemoglobin could not exist. Consistent with the results of Kuilier,18 patients with higher comorbidities may have a lower threshold of tolerance to anemia. In that study, it was found that anemic patients undergoing cardiac surgery have an increased risk of postoperative adverse events and that the number of preexisting comorbidities substantially affects the tolerance of perioperative anemia.

Reduced hemoglobin levels may contribute to worsening outcomes through higher peripheral and myocardial oxygen demand and the development of left ventricular hypertrophy, primarily due to a resultant increased cardiac output.19 An inverse relationship between hemoglobin and left ventricular hypertrophy has been described in studies of patients with chronic kidney disease.20

In their series, Bell found an OR of 2.0 (95% CI, 1.4-2.8) for the composite outcome of in-hospital death, cerebrovascular event or acute renal failure in patients with preoperative anemia.21

Lower preoperative hemoglobin levels are associated with increased postoperative adverse events throughout all hemoglobin levels and in a dose-dependent manner. However, the effect of anemia on most outcomes is weakened by adjusting for relevant covariates.4 In a retrospective evaluation of over 36,000 patients,21 it was found that a preoperative hemoglobin level < 10 g/dL was not an independent predictor of mortality at 30 days,concluding that the association of anemia with elevated serum creatinine levels or the presence of heart failure could explain the association in unadjusted analyses.

Subgroup of diabetic patients:

For forecasting purposes, a significant interaction has been described between BNP and diabetes mellitus (DM),4 recording a significant increase in the risk of adverse outcomes at the 12-month follow-up, with BNP > 258 pg/mL in diabetic subjects (OR: 18.82; 95%CI, 16.2–20.5) vs. non-diabetics (OR: 1.26; 95% CI, 0.61-2.61). In the current series, the in-hospital mortality was best predicted by logES alone, with an excellent predictive power (ROC-AUC: 0.95; 95% CI, 0.91-1.00). Hemoglobin, BNP and their combination had a lower predictive power, and the addition of BNP to logES did not change its predictive power.

Similar to the entire cohort, the ability to anticipate a low postoperative hemoglobin level is one aspect that can be considered most useful; for this outcome, the combination of BNP + hemoglobin hada very good predictive power (ROC-AUC: 0.77). The logES + hemoglobin or logES + BNP + hemoglobin model was better suited than logES alone for predicting postoperative anemia. The combination of BNP + hemoglobin, but not hemoglobin alone, gave a good prediction of the need for a transfusion (AUC: 0.7 vs. AUC: 0.28).

The appearance of PLCO was better discriminated by BNP (ROC-AUC: 0.73; 95% CI, 0.64-0.83) or BNP + hemoglobin (ROC-AUC: 0.75; 95% CI, 0.66-0.83) than by logES (ROC-AUC:0.71 95% CI, 0.60-0.81), but these differences did not reach statistical significance. Barros et al.22 found that the post-CABG in-hospital mortality was 11.8% for diabetic patients, with a 14.8% incidence of PLCO. The study identified two risk factors for post-coronary revascularization mortality in diabetic patients: the use of cardiopulmonary bypass and PLCO. Rao et al.23 identified nine independent risk factors for the occurrence of PLCO, including the presence of DM, with an OR of 1.6. In their series, the operative mortality was higher in the patients who developed PLCO vs. those who did not (16.9% vs. 0.9%, p: 0.001).

The EuroSCORE was specifically developed to predict the in-hospital mortality; however, from the perspective of the patient, a scale only applicable to this outcome can be interpreted as a limited vision of the risk spectrum that the patients have to cope with when scheduled for cardiac surgery. Previous studies have shown that the morbidity events following discharge are high.24 Few studies have evaluated other relevant outcomes, especially the reliability of logES in predicting the postoperative long-term morbidity and mortality.

Based on the results of our population, logES can be considered to be a model that is not limited to the prediction of in-hospital mortality; its prognostic value is good for the in-hospital outcomes of morbidity, especially PSICU (ROC-AUC: 0.72; 95% CI, 0.66-0.77), acute renal failure (ROC-AUC: 0.78; 95% CI, 0.71-0.85) and PLCO (ROC-AUC: 0.71; 95% CI, 0.66-0.76). Regarding late outcomes, its ability to predict the 6-month mortality (ROC-AUC: 0.73; 95% CI, 0.64-0.83), cerebrovascular event (ROC-AUC: 0.78;95% CI, 0.70-0.85) or a combined morbidity outcome (ROC-AUC: 0.70;95% CI, 0.65-75) is very good.

In diabetic patients, the predictive power of logES is even higher, and the discrimination of reintervention, complex ventricular arrhythmias, acute renal failure or prolonged inotropic requirements are good to excellent. At 6 months, the discrimination is excellent for mortality and very good for cerebrovascular events.

logES can then be used to predict long-term complications in a broad spectrum of cardiac surgery patients. The identification of patients at high risk for morbidity and mortality is important to ensure that these patients have intensive follow-up; additionally, it would allow for a better preoperative preparation or modification of thesurgical procedure to obtain the best result.

However, the major limitation of logES in real-world practice is the high number of missing values for some variables. In the study by Noyes, 25 a proportion as high as 58% of the ejection fraction values and up to 77% of the systolic pressure values in the pulmonary artery were not available or were not measured. These two variables contribute significantly to the model; therefore, the absence of reports or subjective estimates should not be used for risk assessment. Hence, some considerations arise when evaluating the clinical utility of these biomarkers in clinical practice. BNP is a simple test that can be easily performed in most laboratories as part of the preoperative evaluation, and its cost is low compared to the costs of surgery and follow-up. The good discriminatory power of BNP to predict hemodynamic complications, acute renal failure and the need for a transfusion, even with an ROC-AUC value of 0.72, may significantly impact morbidity.

There are limitations to this study. Our patients have been evaluated and intervened in only one institution, and accordingly, the results need to be further evaluated in other centers. In addition, the inclusion of hemoglobin or BNP in the logES model required calculating hazard ratios in the univariate model; however, to be able to weigh the variables differently would imply building a new model, which is not viable at this stage of development in this line of research.


The discriminatory power of logES for mortality in our country is very good, especially for diabetics. Hemoglobin or BNP added no predictive power for this outcome. In most of the morbidity outcomes, the discrimination obtained with BNP is comparable to logES alone. The addition of these markers to logES results in a model with a good predictive power for the need of a perioperative transfusion.


This study was funded by a grant from Colciencias. Colciencias is an institution of the government of Colombia, which finances research projects. Except for the group of researchers, nobody had access to the cohort database. The study design, data analysis and conclusions have been carried out based entirely on the researchers criteria.