Artificial intelligence as a precision tool for predicting risk of in-hospital death after aortic valve replacement

Background/Introduction

The prediction of risk of in-hospital death associated with cardiac surgery still has important gaps. In this scenario, the computational tools and mathematical techniques, the pillars of artificial intelligence, can represent an effective solution to this problem.

Purpose

To develop an in-hospital death prediction model for isolated aortic valve replacement (AVR) based on an artificial intelligence constituted by an artificial neural network (ANN).

Methods

352 patients consecutively submitted to isolated AVR between 2010 and 2020 were included. Altogether, 30 baseline variables were evaluated. Initially, the Extra Tree Classifier machine learning algorithm was used to select the attributes with the highest association with death. With the application of the algorithm, it was possible to identify the 11 variables with the greatest weight associated with in-hospital death. After selecting the variables and dividing the dataset into training (70%) and testing (30%), a risk prediction model was structured through an ANN with multiple layers. The ReLU activation function was used in the hidden layers and the SoftMax activation function was used in the output layer. As an optimizing function of the ANN, the Nadam function was used. In addition, a thousand cycles of propagation and data return (Epochs) were performed to induce machine learning based on the cyclic adjustment of the weights of each of the independent variables included in the model. Accuracy assessments were performed using the ROC curve in the test dataset. The model was developed using the Python programming language.

Results

A predictive accuracy of 93,6% (AUC 0,936) was observed for the occurrence of in-hospital death in the test dataset to the ANN. When comparing the performance of traditional risk scores, also tested only in the test dataset, we found that the ANN-based model was significantly superior to the scores (EuroScore I = 84,0% (AUC 0,840); EuroScore II = 84,4% (AUC 0,844), STS Score = 74,0% (AUC 0,740). The area under the curve of the model based on the ANN was significantly higher when compared to the areas of the scores using the DeLong test (p<0.05). When applying the same model only to patients aged 75 and over, the results were as follows: ANN AUC 0,877; ES1 AUC 0,652; ES2 AUC 0,590; STS AUC 0,663 (p<0,05).

Conclusion

The application of artificial intelligence modelling is feasible for the creation of prediction models in the health area. In this study, the accuracy of the ANN was significantly higher than that of the other traditional risk scores in the general sample and for patients with more advanced age. These findings demonstrate the great potential that representative datasets have when accessed through artificial intelligence techniques. The demand for massive volumes of information is mitigated when well-structured datasets with extreme data quality is used.

Funding Acknowledgement

Type of funding sources: None.

Artificial intelligence as a precision tool for predicting risk of in-hospital death after coronary artery bypass graft surgery

Background/Introduction

The prediction of risk of in-hospital death associated with cardiac surgery still has important gaps. In this scenario, the computational tools and mathematical techniques that constitute data science and provide machine learning, pillars of artificial intelligence, can represent an effective solution to this problem.

Purpose

To develop an in-hospital death prediction model for isolated CABG based on an artificial intelligence constituted by an artificial neural network (ANN).

Methods

3,124 patients consecutively submitted to isolated CABG between 2010 and 2020 were included. Altogether, 30 baseline and operative variables were evaluated. Initially, the Extra Tree Classifier machine learning algorithm was used to select the attributes with the highest association with death. With the application of the algorithm, it was possible to identify the 13 variables with the greatest weight associated with hospital death. After selecting the variables and dividing the dataset into training (70%) and testing (30%), a risk prediction model was structured through an ANN with multiple layers. The ReLU activation function was used in the hidden layers and the SoftMax activation function was used in the output layer to extract the specific probability of death and survival. As an optimizing function of the ANN, the Nadam function was used. In addition, a thousand cycles of propagation and data return (Epochs) were performed to induce machine learning based on the cyclic adjustment of the weights of each of the independent variables included in the model. Accuracy assessments were performed using the ROC curve in the test dataset. The model was developed using the Python programming language.

Results

After consolidating machine learning based on the training dataset with 70% of the general sample, it was possible to observe that through the artificial intelligence technique, a predictive accuracy of 83.86% (AUC 0.8386) was obtained for the occurrence of in-hospital death in the test dataset. When comparing the performance of traditional risk scores, also tested only in the test dataset, we found that the ANN-based model was significantly superior to the scores (EuroScore I = 71.4% (AUC 0.714); EuroScore II = 71.9% (AUC 0.719), STS Score = 71.1% (AUC 0.714). The area under the curve of the model based on the ANN was significantly higher when compared to the areas of the scores using the DeLong test (p<0.05)

Conclusion

The application of artificial intelligence modelling is feasible for the creation of prediction models in the health area. In this study, the accuracy of the ANN was significantly higher than that of the other traditional risk scores. These findings demonstrate the great potential that representative datasets have when accessed through artificial intelligence techniques. The demand for massive volumes of information is mitigated when well-structured datasets with extreme data quality is used.

Funding Acknowledgement

Type of funding sources: None.

Impact of atrial fibrillation on in-hospital outcomes of coronary artery bypass graft surgery: an analysis by propensity score matching

Background/Introduction

Many patients referred for coronary artery bypass graft (CABG) surgery have persistent or permanent atrial fibrillation (AF). Despite the significant occurrence, the impact of this arrythmia on surgical outcomes remains uncertain.

Purpose

To assess the impact of persistent or permanent AF on post-CABG in-hospital outcomes.

Methods

A cohort of 3,124 patients undergoing to isolated CABG between 2010 and 2020 was evaluated. A propensity score pairing was applied, considering persistent or permanent AF as a dependent variable and another 19 baseline characteristics as independent variables in the regression model used to build the propensity score. Pairing was performed at a ratio of 3:1 – Group 1: 324 patients without persistent or permanent AF; Group 2: 108 patients with the documented diagnosis of persistent or permanent AF. The statistical plan also included normality analyses, descriptive and univariate analyses, binary logistic regression, ROC curves and DeLong test to compare de curves. The significance level adopted was 5%. The analyses were performed by the programming language Python.

Results

None of the baseline characteristics evaluated showed a significant difference between the groups, including the EuroScore II (Group 1: 1.54±1.45 vs Group 2: 1.49±1.59; p=0.990). Likewise, none of the analysed surgical characteristics showed a significant difference, indicating a very approximate pattern of complexity of the surgeries. The absence of differences demonstrated a high degree of homogeneity between the groups. The use of pairing by propensity score aimed to form two extremely similar study groups, which differed only in relation to the diagnosis of the arrhythmia under study. Among the outcomes evaluated, AMI (1.5 vs 6.5; p=0.013), MACCE (7.1% vs 14.8%; p=0.015) and death (1.5% vs 6.5%; p=0.013) had significantly higher incidences in Group 2, formed by patients with persistent or permanent AF. From the multivariate analysis, it can be identified that permanent AF was an independent risk predictor for the occurrence of in-hospital death (OR: 5.009; 95% CI 1.433–17.507; p=0.012). Finally, it was also possible to verify that the association of EuroScore II with persistent or permanent AF showed higher predictive accuracy than EuroScore II alone (ESII+FA = AUC 0.852 vs ESII alone = AUC 0.775, p<0,05).

Conclusion(s)

Patients with persistent or permanent AF had significantly higher incidences of AMI, MACCE and in-hospital death. Persistent or permanent AF was characterized as an independent predictor for the occurrence of death and the association with the EuroScore II resulted in a 9.9% increase in the predictive accuracy of the surgical risk score.

Funding Acknowledgement

Type of funding sources: None.

Preoperative anaemia is an independent predictor of 30-day mortality post-CABG and improves de predictive accuracy of EuroScore II

Background

The impact of preoperative anaemia on the results of cardiovascular surgery has already been demonstrated by some authors. Some of the studies observed worse surgical results in anaemic patients, however, in others, no difference was observed. In addition, the impact that the inclusion of anaemia or haemoglobin would have as a predictor in a risk model such as EuroScore II is not yet known.

Purpose

To evaluate the impact of preoperative anaemia on 30-day mortality post-CRM and compare the predictive accuracy of EuroScore II with and without the inclusion of haemoglobin levels in the model.

Methods

Single center cohort with 2168 patients consecutively included between January 2010 and December 2020. All procedures were performed with cardiopulmonary bypass. 32 baseline and operative characteristics were assessed. The primary outcomes were 30-day mortality and the EuroScore II predictive accuracy. Patients were stratified into two groups according to anaemia status. WHO Classification – haemoglobin: men <13 g/dL and women <12 g/dL. No Anaemia Group (1301–60.1%) – Anaemia Group (867–39,9%). Univariate analysis was performed to compare the characteristics of the groups, the occurrence of death in 30 days and to verify variables associated with mortality. Logistic regression analyses were used to assess predictors of mortality and generate a set of probabilities for assessing the predictive accuracy of EuroScore II with and without the addition of haemoglobin in the model. The probabilities generated through the regressions were analysed by ROC curves, which in turn were compared using the DeLong test. The level of significance was 5% and the statistical analysis was performed using Python 3.0.

Results

The anaemic patients were older and had higher prevalence of conditions, such as: diabetes, renal impairment, smoking, HF class III or IV, RBC transfusion and highest mean of EuroScore II (p<0.05 for all conditions). When comparing mortality in the groups, a significantly higher rate was observed in the anaemic group (2.2% vs 5.4%; p<0.001). In view of the heterogeneity of the groups, an adjusted logistic regression model was applied. The results of the multivariate model demonstrated that preoperative anaemia was an independent risk predictor for the occurrence of death in 30 days after CABG (B: 0.597; SE: 0.27; Wald: 5.2; OR: 1.82; 95% CI 1.09–3.04; p=0.022). In addition, we were able to verify that the addition of preoperative haemoglobin to EuroScore II resulted in a significantly higher predictive accuracy when compared to the predictive accuracy of the isolated score (AUCs: 0,732 vs 0.709; p=0.032).

Conclusions

Preoperative anaemia was an independent risk predictor for the occurrence of death in 30 days after CABG. We were also able to verify that the addition of the preoperative haemoglobin levels to EuroScore II resulted in a significantly higher predictive accuracy, improving the performance of the surgical risk model.

Funding Acknowledgement

Type of funding sources: None. Figure 1. Haemoglobin vs 30-day mortalityFigure 2. Predictive accuracies – EuroScore II

EuroScore II is the best predictive model for the off pump coronary artery bypass graft surgery

Background

Coronary artery bypass graft surgery (CABG) is the most common cardiac surgery performed in the world and a significant part of these surgeries are performed without cardiopulmonary bypass (off pump). Although none of the main surgical risk scores include pump use in their prediction model, the scores are widely used in risk stratification, including for patients who will be submitted to off pump CABG.

Purpose

To analyse and compare the predictive accuracy of EuroScore I, EuroScore II and STS Score for 30-day mortality after off pump CABG.

Methods

Single-centre cohort with 943 patients consecutively submitted to off pump CABG between January 2010 and December 2020. 31 baseline and operative variables were analysed. The primary outcome was the occurrence of death in the first 30 days after the surgery. Descriptive analysis, normality for quantitative data and univariate inference were performed to compare proportions and means between the survival group (n=930) and death group (n=13). Next, three logistic regression models were performed. Each of them had 30-day mortality as a dependent variable and one of the scores as an independent variable. The probabilities generated by the three models were saved and analysed by ROC curves. Thus, it was possible to assess the predictive accuracy of each of the scores. Finally, the values of the areas under the curves were compared using the DeLong test. The level of significance was 5% and the analysis was performed using the Python 3.0 programming language.

Results

The mean age of the general group was 63 years old and there was a predominance of male patients (68.4%). The means of the three evaluated risk scores were significantly higher in the Death group (p<0,05). This pattern confirmed the findings of higher prevalence of several comorbidities in the death group. The 30-day mortality rate was 1.37%. Through the analysis of regressions and the probabilities generated through them, it was possible to verify that the predictive accuracy of EuroScore II was significantly higher than that of the other two scores. While the predictive accuracy of EuroScore II was 77.3%, the accuracy of two other scores was in the range of 69% (AUC EsI: 0.697; AUC EsII: 0.773; AUC STS: 0.695; p=0.029).

Conclusion

EuroScore II seems to be the most adequate surgical risk score for the assessment of mortality risk of patients who will undergoing to off pump CABG. The score had a predictive accuracy of 77.3%, almost 8% more than the other two scores. Therefore, although EuroScore II does not include in its model the use of cardiopulmonary bypass, it has a satisfactory accuracy to be used in clinical-surgical practice. On the other hand, the EuroScore I and the STS Score showed predictive accuracy not adequate for this type of surgery.

Funding Acknowledgement

Type of funding sources: None. Predictive accuracies of risk scores

Results of on-pump and off-pump coronary artery bypass graft surgery in 30 days: an analysis by propensity score matching

Background

The results of coronary artery bypass graft surgery (CABG) performed with and without the support of cardiopulmonary bypass have already been widely discussed and studied, including through a few large randomized clinical trials. Despite the efforts, the findings of these studies still generate controversy and doubts about the outcomes achieved by the two techniques. One of the contested points is the heterogeneity of the degree of specialization in the off-pump technique in relation to the surgical groups that participated in the studies.

Purpose

To compare the results in 30 days of on-pump and off-pump CABG.

Methods

A single centre cohort with 1,767 patients undergoing isolated CABG was initially evaluated (January 2013 – December 2018). 397 patients undergoing off-pump CABG and 1,370 patients undergoing on-pump surgery were identified. To obtain two completely homogeneous study groups, a propensity score matching was applied. For this, a logistic regression model was built with the variable use of CPB support as dependent variable. In the group of independent variables, 14 baseline and operative characteristics were included. The probabilities generated for each patient were used as scores to establish the match. To establish a pair, it was necessary to have three squares after the comma, with the fourth decimal place being the tiebreaker criterion in the pairing. In this way it was possible to obtain 332 pairs (N=664). The paired groups, on and off-pump, were compared by descriptive and univariate analysis and later a logistic regression model was applied to identify possible risk predictors and to verify the impact of CPB support on 30-day mortality. The level of significance was 5% and the analysis was performed using Python 3.0.

Results

None of the 29 baseline and operative characteristics showed a significant difference between the groups, demonstrating a high degree of homogeneity obtained from the propensity score matching, which enabled a solid comparison between the incidences of outcomes in 30 days. None of the analysed outcomes showed any difference between the groups on and off-pump, including AMI, stroke, major reoperation and death (1.5% vs 2.4%; p=0.401). Through regression analysis it was possible to establish that the use of CPB was not an independent predictor of risk for the occurrence of death (p=0.246).

Conclusion

After matching by propensity score, patients who underwent surgery with and without CPB had similar incidences of 30-day mortality. In addition, it was possible to verify that the use of CPB was not an independent predictor of risk for the occurrence of death in 30 days.

Funding Acknowledgement

Type of funding sources: None. Propensity score adjustment by group30-day outcomes vs CABG technique