German Aortic Valve Score in Risk Assessment for Surgical Aortic Valve Replacement in a Brazilian Center

Objective To test the German Aortic Valve (GAV) score at our university hospital in patients undergoing isolated aortic valve replacement (AVR). Methods A total of 224 patients who underwent isolated conventional AVR between January 2015 and December 2018 were included. Patients with concomitant procedures and transcatheter aortic valve implantation were excluded. Patients’ data were collected and analyzed retrospectively. Patients’ risk scores were calculated according to criteria described by GAV score. Sensitivity, specificity, and accuracy (area under the ROC curve [AUC]) were also calculated. The calibration of the model was tested by the Hosmer-Lemeshow method. Results The mortality rate was 8.04% (18 patients). The patients’ mean age was 58.2±19.3 years and 25% of them were female (56 patients). Mean GAV score was 1.73±5.86 (min: 0.0; max: 3.53). The GAV score showed excellent discriminative capacity (AUC 0.925, 95% confidence interval 0.882-0.956; P<0.001). The cutoff “1.8” turned out to be the best discriminatory point with the best combination of sensitivity (88.9%) and specificity (75.7%) to predict operative death. Hosmer-Lemeshow method revealed a P-value of 0.687, confirming a good calibration of the model. Conclusion The GAV score applies to our population with high predictive accuracy.


INTRODUCTION
The assessment of operative risk is mandatory for all cardiac procedures, since patients need to be informed preoperatively about the risks and surgeons must weigh up pros and cons of a certain procedure. In this scenario, risk scoring systems are used to predict and evaluate results. Although there are widely spread risk scores, such as the European System for Cardiac Operative Risk Evaluation (EuroSCORE) [1] , that have demonstrated good predictive accuracy in the field of cardiovascular surgery, the trend of the moment is for more specific scores to be applied to more specific contexts in cardiac surgery.
an event (in this case, death) from those who do not. The discriminative capacity of the model was estimated by means of the area under the ROC curve (AUC). Calibration of the GAV score was assessed by the Hosmer-Lemeshow test. The calibration is considered to be poor if the test is statistically significant. For the analysis, the Statistical Package for the Social Sciences (SPSS)® software (SPSS, Inc., Chicago, IL, United States of America), version 15.0, for Windows ® was used. P-values < 0.05 were considered statistically significant.

RESULTS
We evaluated 224 isolated AVR procedures in adult patients. The mortality rate was 8.04% (18 patients). The patients' mean age was 58.2±19.3 years and 25% of them were female (56 patients).
Mean GAV score was 1.73±5.86 (min: 0.0; max: 3.53). The GAV score showed excellent discriminative capacity (AUC 0.925, 95% confidence interval [CI] 0.882-0.956; P<0.001) ( Figure 2). The calibration of the model was tested by the Hosmer-Lemeshow method. The derived P-value of 0.687 confirmed a valid accordance of predicted and observed mortality, which means good calibration of the model.
The cutoff "1.8" turned out to be the best discriminatory point with the best combination of sensitivity (88.9%) and specificity (75.7%) to predict operative death (Table 1).
Könning et al. [2] published in 2013 the German Aortic Valve (GAV) score. It was designed for fair and reliable outcome evaluation, allows comparison of predicted and observed mortality for conventional aortic valve replacement (AVR) and transcatheter aortic valve implantation (TAVI) in low-, moderate-, and high-risk groups, enables a risk-adjusted benchmark of outcome, and fosters the efforts for continuous improvement of quality in aortic valve procedures.
Since the score has never been tested in Brazil, we aimed to validate the GAV score in patients who underwent conventional AVR at a Brazilian center.

METHODS
Patients who underwent conventional AVR between January 2015 and December 2018 were included in the study. Those who underwent concomitant procedures or TAVI were excluded. Data were collected and analyzed retrospectively. Primary endpoint was in-hospital mortality. Patients' GAV scores were calculated according to the criteria described by Kötting et al. [2] (Figure 1). The score is calculated through the sum of regression coefficients, which corresponds to a certain expected operative mortality.
Sensitivity and specificity were assessed through the receiver operating characteristic (ROC) curve. The discrimination measures the capacity of a model (in this case, the GAV score) to differentiate between the individuals of a sample who suffer

Limitation
The major limitations of our study were its non-randomized and retrospective design, single institution setting, and the fact that our hospital is a multi-surgeon one.

DISCUSSION
The presumption that a scoring system might be comprehensive enough for all patients and cardiovascular surgical procedures could not be further from the truth [3,4] . For instance, the widely used EuroSCORE was based on a data set consisting mainly of coronary artery bypass surgeries. Thus, such score might be less well adapted to aortic procedures than a specific score as the one evaluated in the present study. Such aspects have been highlighted by other authors as well [5][6][7][8] .
To the best of our knowledge, our study is the first one to report the results of the GAV score in a Latin American scenario. It is well known that predictive models work best in the series at the location where it was developed. For this reason, the GAV score fits best to the population in Germany. Nevertheless, despite the differences between German and Brazilian populations, the score also showed a very good discriminative capacity in our population.

CONCLUSION
The GAV score applies to our population with high predictive accuracy and could be used in our population to calculate operative risk.