Abstract
BACKGROUND: This study aimed to develop a risk prediction model based on association rule mining to predict recurrent febrile seizures (RFS). METHODS: This is a retrospective observational study that examined the medical records of 105 children who were followed up with febrile seizure (FS) in a tertiary paediatric emergency department between October 2022 and December 2023. Children were divided into RFS and simple FS groups. RFS was defined as seizures occurring more than once within 24 hours of the first FS in the same febrile illness. Risk factors associated with RFS were determined by univariate and multivariate analyses. χ(2), Mann-Whitney U, receiver operating characteristics (ROC), multiple logistic regression and Classification Based on Association Rules Algorithm (CBA) analyses were applied to the dataset to obtain high-level outputs. RESULTS: RFS was detected in 32 out of 105 cases with FS (30.5%). Potential risk factors contributing to the development of RFS were seizure duration, number of recurrent seizures, family history, body temperature, time from fever onset to seizure, time from seizure onset to arrival at the emergency department, hyponatraemia, osmotic pressure and low haemoglobin level. The CBA algorithm obtained a total of 11 classification rules for the two patient groups. Additionally, the cut-off values obtained from CBA and ROC analysis showed satisfactory consistency. The CBA model achieved 97% overall accuracy classification performance. CONCLUSION: The developed CBA model shows good predictive ability for RFS. The relevant model can be used as a risk estimation tool to identify children at risk of developing RFS.