Abstract:

Objective To analyze the risk factors for the development of epilepsy in children with febrile convulsions based on Logistic regression and decision tree models and to construct a predictive model. Methods A total of 210 children with their first occurrence of febrile convulsion were selected and followed up for half a year (whether epilepsy occurred or not). Eventually, 196 cases completed the follow-up, and 36 cases (18.37%) of children with febrile convulsions developed epilepsy. Children were divided into the epilepsy group (36 cases) and the non-epilepsy group (160 cases) based on whether they developed epilepsy or not. All children underwent the first electroencephalogram (EEG) examination within 72 hours after the febrile convulsion and a follow-up EEG examination three months later. The clinical data of the children were collected, including gender, age of onset, birth weight, maternal age at delivery, mode of delivery, anemia status, nature of the first convulsion, number of convulsions, status at the first convulsion, time from fever to the first convulsion, duration of convulsion (taking the longest value), family history of epilepsy and EEG results. Multivariate Logistic regression was used to analyze influencing factors of febrile convulsions developing into epilepsy. Modeler software was used to construct a decision tree risk prediction model for predicting the development of epilepsy in children with febrile convulsions. The receiver operating characteristic (ROC) curve was drawn, and the area under the curve (AUC) of different models was compared. Results Compared with the non-epilepsy group, there were higher proportion of complex febrile convulsions, the number of seizures was ≥2 times, time from fever to the first convulsion was<24 hours, duration of convulsion was ≥ 15 minutes and abnormal EEG results were higher in the epilepsy group (P<0.01). Multivariate Logistic regression analysis showed that complex febrile convulsions, the frequency of convulsions≥2 times, time from fever to the first convulsion<24 hours, convulsion duration ≥ 15 minutes and abnormal EEG were risk factors for the development of epilepsy in children with febrile convulsions (P<0.05). The decision tree model selected the number of convulsions, time from fever to the first convulsion, nature of the first convulsion and EEG as important variables for the development of epilepsy in children with febrile convulsions were important variables for the development of epilepsy in children with febrile convulsions, and with information gains of 0.47, 0.27, 0.14 and 0.13, respectively. The AUC value of the multivariate Logistic model was 0.838 (95%CI: 0.764-0.911), and the AUC value of the decision tree model was 0.849 (95%CI: 0.780-0.916). There was no significant difference in AUC between the two models. Conclusion The method combining decision tree algorithm and Logistic regression model to identify the risk factors for the development of epilepsy in children with febrile convulsions has certain predictive value and can provide a reference for clinical practice.

Key words: seizures, febrile, epilepsy, decision trees, Logistic models, electroencephalography, risk factors