The creation and validation of predictive models to assess the risk of unfavorable outcomes following hybrid total arch repair for Stanford type A aortic dissection

Abstract

Background: The objective of this study was to develop and validate a nomogram for the individualized prediction of adverse events in patients with Stanford type A aortic dissection (TAAD) undergoing hybrid total aortic arch repair.

Methods: From April 2019 to April 2022, we conducted a comprehensive review of the medical records of Stanford type A aortic dissection patients who underwent hybrid total aortic arch repair surgery at our hospital. Patients were separated into two groups based on whether or not a composite adverse event occurred following surgery. Using univariate and multivariate analyses of logistic regression, the prediction model was created. Construct risk prediction models utilizing nomograms and evaluate their precision, discrimination, and clinical utility.

Results: Age, platelets, serum blood urea nitrogen, and ascending aortic diameter were the variables included in the nomogram by univariate and multivariate analysis. The risk model performed well in internal validation, with an area under the curve (AUC) of 0.829. The calibration curve demonstrated good agreement between predicted and actual probabilities (Hosmer-Lemeshow test, P = 0.22). Clinical decision analysis curves demonstrate predictive nomograms' clinical utility.

Conclusion: This study created and validated a nomogram for predicting the risk of composite endpoint events in TAAD patients undergoing hybrid total aortic arch repair. The nomogram can help determine the severity of a patient's condition and provide a more personalized diagnosis and treatment.

Keywords: In-hospital compound adverse events; Nomogram; Stanford type A Aortic Dissection.

Fig. 1

Process of choosing and grouping patients is shown in a flowchart

Fig. 2

A nomogram for evaluating the risk of a composite endpoint event after hybrid aortic repair in people with Stanford type A aortic dissection. The four individual scores can be added up to create the total score, which can then be used to calculate the risk of composite endpoint corresponding probability of occurrence.

Fig. 3

The ROC curve used to evaluate the discriminative performance of the model, the area under the curve AUC was 0.829, and the optimal cutoff point suggested a sensitivity of 77.0% and a specificity of 74.3%.

Fig. 4

(A) The calibration curve for the prediction model shows the degree of agreement between the predicted risk and the actual outcome, with the predicted risk on the x-axis and the actual outcome on the y-axis. (B) Brier score and other correlation probability calibration values are displayed, and indicate a good probability calibration effect.

Fig. 5

(A) Clinical decision curve analysis of predictive models. The Y-axis represents the net benefit, the orange line depicts the predictive model, the “All” curve represents all intervention, the “NO” curve represents no intervention at all, and the curve shows that the model outperforms both “intervention” and “no intervention” treatment strategies. (B) The red line represents the predictive model, and the blue dashed line represents the actual data, in the clinical impact curve of the predictive model. The graph indicates that the predictive model is clinically useful.