| Objectives According to the indicator system screened by the analysis of risk factors for early onset of Refractory Epilepsy(RE)in children,develop the risk prediction model of early onset of refractory epilepsy in children and evaluate and verify the model.Methods 1 Based on the case-control study design,retrospectively analyzed epilepsy patients newly diagnosed in the Pediatric Outpatient and Inpatient Wards of the Tangshan City Maternal and Child Health Hospital Affiliated to North China University of Technology from January 2012 to January 2018.Patients with refractory epilepsy were set as case groups,and patients diagnosed with non-refractory epilepsy were set as control groups.Information collection of epilepsy patients was completed by checking the case log records,interviewing parents and other guardians of children in person and on the phone.The information collected includes the general situation of the child,the age of the first seizure,the etiology of the epilepsy,the clinical characteristics of the seizure,the time from onset to drug treatment,the electroencephalogram performance before and after treatment,the MRI results of the brain before treatment,and the first anti-epileptic drug efficacy,whether epilepsy syndrome,past history,family history,etc.2 Based on the collected data,the unconditional logistic regression method was used to analyze the relevant risk factors for each indicator.Based on the OR value(Odds Ratio)and 95%confidence interval,an indicator system for the prediction model was established.3 The sample data collected from January 2012 to January 2017 was used as the modeling population,and the sample data collected from January 2016 to January 2018 as the verification population.Used Logistic regression and classification regression methods in the modeling population to establish a prediction model,verified the established prediction model among the verification population,and evaluated the prediction ability and prediction effect of the prediction model.4 Statistical processing was performed using IBM SPSS Statistics 22.0 software.Results 1 This study included a total of 326 study subjects,including 111 cases in the case group and 215 cases in the control group.According to the classification criteria,273 cases were classified as modeling population and 131 cases were classified as verification population.2 According to the single factor logistic regression analysis of the OR value and 95%confidence interval of each index,the final single factor analysis determines 13 related forecasting indexes,and the multifactor logistic regression analysis determines that 4 forecasting indexes enter the regression equation,respectively:X1 is the cause(assignment of idiopathic disease is 0,symptomatic or cryptogenic cause is 1),X2 is the change of attack type(no change in the assignment is 0,there is a change to 1),X3 is the EEG after half a year of treatment(the assignment is normal or the background abnormality is 0,the epileptiform discharge is 1),X4 is the effect of initial medication(the effective value is 0,the invalid value is 1).3 The built Logistic regression prediction model is P=1/(1+exp[-(-3.871+2.405X1+2.333X2+1.013X3+2.606X4)])among the modeled population.The area under the curve(AUC)of the model is 0.937.When the cutoff value is 0.296,the Youden index is the largest,for0.737,and the corresponding sensitivity and specificity are 0.908 and 0.829,respectively.The goodness-of-fit test of the prediction model P=0.921.The established classification tree model is the classification tree model including the efficacy of the first antiepileptic drug,the etiology of epilepsy,the MRI results before treatment,and the psychomotor development before treatment.The area under the curve(AUC)of the model in the modeled population is 0.930.When the cutoff value is 0.604,the Youden index is the largest,for 0.746,and the corresponding sensitivity and specificity are 0.786 and 0.960,respectively.The 10-fold cross-validation risk estimate is 0.103.It is found in the verification population that the AUC is 0.919.When the cut-off value is 0.169,the Youden index is the largest,for 0.710,and the corresponding sensitivity and specificity are 0.949 and 0.761,respectively.The estimated risk in the verification population is 0.107.After statistical analysis,the difference between the AUC of the logistic regression prediction model and the classification tree prediction model in the modeling population and the verification population is not statistically significant.Conclusions 1 Using Logistic regression and classification tree analysis,two early prediction models for the risk of children with refractory epilepsy are established.2 After verification and evaluation,it is found that both the logistic regression prediction model and the classification tree prediction model have strong prediction discrimination ability and high prediction accuracy,and can be used as complementary.Figure3;Table16;Reference 88... |
PDF Full Text Request