Development and validation of a predictive model for febrile seizures

Abstract

AbstractFebrile seizures (FS) are the most prevalent type of seizures in children. Existing predictive models for FS exhibit limited predictive ability. To build a better-performing predictive model, a retrospective analysis study was conducted on febrile children who visited the Children's Hospital of Shanghai from July 2020 to March 2021. These children were divided into training set (n = 1453), internal validation set (n = 623) and external validation set (n = 778). The variables included demographic data and complete blood counts (CBCs). The least absolute shrinkage and selection operator (LASSO) method was used to select the predictors of FS. Multivariate logistic regression analysis was used to develop a predictive model. The coefficients derived from the multivariate logistic regression were used to construct a nomogram that predicts the probability of FS. The calibration plot, area under the receiver operating characteristic curve (AUC), and decision curve analysis (DCA) were used to evaluate model performance. Results showed that the AUC of the predictive model in the training set was 0.884 (95% CI 0.861 to 0.908, p < 0.001) and C-statistic of the nomogram was 0.884. The AUC of internal validation set was 0.883 (95% CI 0.844 to 0.922, p < 0.001), and the AUC of external validation set was 0.858 (95% CI 0.820 to 0.896, p < 0.001). In conclusion, the FS predictive model constructed based on CBCs in this study exhibits good predictive ability and has clinical application value.