Development and validation of a risk prediction model for valve regurgitation in Behçet's disease

In Behçet's disease (BD), mild-to-severe valvular regurgitation (VR) poses a serious complication that contributes significantly to heart failure and eventually death. The accurate prediction of VR is crucial in the early stages of BD subjects for improved prognosis. Accordingly, this study aimed to develop a nomogram that can detect VR early in the course of BD.

One hundred seventy-two patients diagnosed with Behçet's disease (BD) were conducted to assess cardiac valve regurgitation as the primary outcome. The severity of regurgitation was classified as mild, moderate, or severe. The parameters related to the diagnostic criteria were used to develop model 1. The combination of stepAIC, best subset, and random forest approaches was employed to identify the independent predictors of VR and thus establish model 2 and create a nomogram for predicting the probability of VR in BD. Receiver operating characteristics (ROC) and decision curve analysis (DCA) were used to evaluate the model performance.

Thirty-four patients experienced mild-to-severe VR events. Model 2 was established using five variables, including arterial involvement, sex, age at hospitalization, mean arterial pressure, and skin lesions. In comparison with model 1 (0.635, 95% CI: 0.512-0.757), the ROC of model 2 (0.879, 95% CI: 0.793-0.966) was improved significantly. DCA suggested that model 2 was more feasible and clinically applicable than model 1.

A predictive model and a nomogram for predicting the VR of patients with Behçet's disease were developed. The good performance of this model can help us identify potential high-risk groups for heart failure. Key Points • In this study, the predictors of VR in BD were evaluated, and a risk prediction model was developed for the early prediction of the occurrence of VR in patients with BD. • The VR prediction model included the following indexes: arterial involvement, sex, age at hospitalization, mean arterial pressure, and skin lesions. • The risk model that we developed was better and more optimized than the models built with diagnostic criteria parameters, and visualizing and personalizing the model, a nomogram, provided clinicians with an easy and intuitive tool for practical prediction.