Development and Validation of a Novel Predictive Model for the Early Differentiation of Cardiac and Non-Cardiac Syncope

Abstract

Background: The diagnosis of cardiac syncope remains a challenge. This study sought to develop and validate a diagnostic model for the early identification of individuals likely to have a cardiac cause.

Methods: 877 syncope patients with a determined cause were retrospectively enrolled at a tertiary heart center. They were randomly divided into the training set and validation set at a 7:3 ratio. We analyzed the demographic information, medical history, laboratory tests, electrocardiogram, and echocardiogram by the least absolute shrinkage and selection operator (LASSO) regression for selection of key features. Then a multivariable logistic regression analysis was performed to identify independent predictors and construct a diagnostic model. The receiver operating characteristic curves, area under the curve (AUC), calibration curves, and decision curve analysis were used to evaluate the predictive accuracy and clinical value of this nomogram.

Results: Five independent predictors for cardiac syncope were selected: BMI (OR 1.088; 95% CI 1.022-1.158; P =0.008), chest symptoms preceding syncope (OR 5.251; 95% CI 3.326-8.288; P <0.001), logarithmic NT-proBNP (OR 1.463; 95% CI 1.240-1.727; P <0.001), left ventricular ejection fraction (OR 0.940; 95% CI 0.908-0.973; P <0.001), and abnormal electrocardiogram (OR 6.171; 95% CI 3.966-9.600; P <0.001). Subsequently, a nomogram based on a multivariate logistic regression model was developed and validated, yielding AUC of 0.873 (95% CI 0.845-0.902) and 0.856 (95% CI 0.809-0.903), respectively. The calibration curves showcased the nomogram's reasonable calibration, and the decision curve analysis demonstrated good clinical utility.

Conclusion: A diagnostic tool providing individualized probability predictions for cardiac syncope was developed and validated, which may potentially serve as an effective tool to facilitate early identification of such patients.

Keywords: cardiac syncope; diagnosis; nomogram; prediction model; syncope.

Figure 1

Selection of key features related to cardiac syncope by the least absolute shrinkage and selection operator (LASSO) regression. (A) The LASSO regression coefficient trendlines of the 36 candidate features. (B) Determination of the tuning parameter λ by 10-fold cross-validation. Two vertical lines were drawn, representing a more concise model within one standard error. The tuning parameter λ = 0.032 was selected under the 1-SE criteria with 10 non-zero coefficients included.

Figure 2

The nomogram for predicting the probability of cardiac syncope.

Figure 3

The ROC curves of the nomogram in both the training and validation sets.

Figure 4

Assessment of the nomogram model for diagnosing cardiac syncope. Calibration plots in the training set (A) and the validation set (B), decision curves in the training set (C) and the validation set (D).

Figure 5

The screen shot of the dynamic web-based nomogram established to predict the probability cardiac syncope (https://cardiacsyncope.shinyapps.io/DynNomapp/).