Develop ment and validation of a prognostic dynamic nomogram for in-hospital mortality in patients with Stanford type B aortic dissection

Abstract

Background: This study aimed to identify the risk factors for in-hospital mortality in patients with Stanford type B aortic dissection (TBAD) and develop and validate a prognostic dynamic nomogram for in-hospital mortality in these patients.

Methods: This retrospective study involved patients with TBAD treated from April 2002 to December 2020 at the General Hospital of Northern Theater Command. The patients with TBAD were divided into survival and non-survival groups. The data were analyzed by univariate and multivariate logistic regression analyses. To identify independent risk factors for in-hospital mortality, multivariate logistic regression analysis, least absolute shrinkage, and selection operator regression were used. A prediction model was constructed using a nomogram based on these factors and validated using the original data set. To assess its discriminative ability, the area under the receiver operating characteristic curve (AUC) was calculated, and the calibration ability was tested using a calibration curve and the Hosmer-Lemeshow test. Clinical utility was evaluated using decision curve analysis (DCA) and clinical impact curves (CIC).

Results: Of the 978 included patients, 52 (5.3%) died in hospital. The following variables helped predict in-hospital mortality: pleural effusion, systolic blood pressure ≥160 mmHg, heart rate >100 bpm, anemia, ischemic cerebrovascular disease, abnormal cTnT level, and estimated glomerular filtration rate <60 ml/min. The prediction model demonstrated good discrimination [AUC = 0.894; 95% confidence interval (CI), 0.850-0.938]. The predicted probabilities of in-hospital death corresponded well to the actual prevalence rate [calibration curve: via 1,000 bootstrap resamples, a bootstrap-corrected Harrell's concordance index of 0.905 (95% CI, 0.865-0.945), and the Hosmer-Lemeshow test (χ² = 8.3334, P = 0.4016)]. DCA indicated that when the risk threshold was set between 0.04 and 0.88, the predictive model could achieve larger clinical net benefits than "no intervention" or "intervention for all" options. Moreover, CIC showed good predictive ability and clinical utility for the model.

Conclusion: We developed and validated prediction nomograms, including a simple bed nomogram and online dynamic nomogram, that could be used to identify patients with TBAD at higher risk of in-hospital mortality, thereby better enabling clinicians to provide individualized patient management and timely and effective interventions.

Keywords: Stanford type B aortic dissection; in-hospital mortality; nomogram; prediction model; risk factors.

FIGURE 1

Flow chart of the study. LASSO, least absolute shrinkage and selection operator.

FIGURE 2

Predictors selection using LASSO regression. (A) LASSO coefficient profiles of all the clinical features. (B) Identification of the optimal penalization coefficient λ in the LASSO model with 10-fold cross validation and the 1-SE criterion. Results of LASSO regression screening for risk factors for in-hospital mortality. SE, standard error; LASSO, least absolute shrinkage and selection operator.

FIGURE 3

Nomogram model for predicting the risk of in-hospital death in patients with Stanford type B aortic dissection. Every factor in the nomogram got an individual score according to the value of factor, and a total score was obtained by summarizing the scores of the seven factors, which could be used to estimate the probability of the risk of in-hospital death in this type B aortic dissection patient.

FIGURE 4

Nomogram model predicts in-hospital death risk in patients with Stanford type B aortic dissection. Given values of the seven variables, the patient can be mapped onto the nomogram. Each red dot represents the value of each variable of the patient. As shown, a probability of 63.2% was the risk of in-hospital death in this type B aortic dissection patient. PE, pleural effusion; ICVD, ischemic cerebrovascular disease; HR, heart rate. *0.01 = p < 0.05; **0.001 = p < 0.05; ***p < 0.001.

FIGURE 5

ROC curve for evaluating the model’s discrimination performance. AUC of the ROC curve is 0.894 (95% CI, 0.850–0.938). ROC, receiver operating characteristic; AUC, area under the curve.

FIGURE 6

Nomogram calibration curve. The x-axis represents the nomogram-predicted probability, and the y-axis represents the actual probability of the nomogram. The “Ideal” line indicates perfect prediction by an ideal model. The “Apparent” line depicts the model’s performance, and the black solid line is bias-corrected by bootstrapping (B = 1,000 repetitions), indicating observed nomogram performance. C-index of the nomogram calibration curve is 0.905 (95% CI, 0.865–0.945).

FIGURE 7

DCA for the predictive nomogram. The y-axis represents the net benefit. The “None” line is the net benefit of intervening no patients. The “All” line is the net benefit of intervening all patients. Solid red line is the net benefit of intervening patients on the basis of the nomogram. The generated curve indicated that at a threshold probability ranging from approximately 4–88%, the nomogram model can be beneficial for making the decision to intervene. DCA, decision curve analysis.

FIGURE 8

CIC of the nomogram. Two horizontal axes show the correspondence between cost: benefit ratio and risk threshold. Of 1,000 patients, solid red line shows the total number of high-risk patients for each risk threshold. Dotted red line shows how many of those are with positive event. CIC, clinical impact curves.