Opracowanie i walidacja modelu do wczesnego przewidywania epilepsji u pacjentów z urazem mózgu na podstawie bazy danych MIMIC-IV
Traumatic brain injury patients based on MIMIC-IV database Construction and validation of early epilepsy prediction model
W skrócie
[Preprint - wstępne wyniki] Badacze utworzyli model przewidujący, czy pacjent z urazem mózgu rozwinie epilepsję w ciągu pierwszych 7 dni pobytu w szpitalu. Model wykorzystuje 11 rutynowych wskaźników klinicznych, takich jak skala Glasgow, temperatura ciała, nasycenie tlenem i wyniki badań laboratoryjnych. Testowanie modelu na prawie 1300 pacjentach wykazało jego dobrą dokładność (AUC 0,748), co sugeruje, że może być przydatny w praktyce klinicznej do identyfikacji pacjentów zagrożonych wczesną epilepsją i pomocy w planowaniu indywidualnego leczenia.
Oryginalny abstract (angielski)
Abstract Background Early epilepsy is a common complication after traumatic brain injury ( TBI ), which is more common. The purpose of this study is to try to establish a model with some conventional clinical indicators, and then to verify this model to see whether it can predict the possibility of early epilepsy in TBI patients. Methods The MIMIC-IV database provides data on TBI patients admitted from 2008 to 2019. These patients were selected according to stricter criteria, and then the data were thoroughly cleaned up. At the same time, epilepsy was detected in a hierarchical manner, so that a relatively high-quality data set can be prepared for the establishment of the model. Among the 92 possible variables, we first use the single factor method to screen, then check whether there is collinearity between them, and finally use LASSO regression to determine the important predictors. The whole process is divided into three stages. Next, the multivariate logistic regression is used to establish the prediction model, and then the Bootstrap method is used for internal verification. At the same time, a more comprehensive sensitivity analysis is made to verify the robustness of the model. The study was conducted in accordance with the Declaration of Helsinki. Results A total of 1293 patients were finally included, and the incidence of early epilepsy ( within 7 days, high-quality evidence ) was 8.58% ( 111 cases ). After three-step screening, 11 core predictive variables were obtained : GCS score ( OR = 0.922 ), head AIS score ( OR = 1.206 ), alcohol-related ( OR = 1.677 ), body temperature ( OR = 1.7667 ), oxygen saturation ( OR = 1.069 ), blood glucose ( OR = 1.006 ), number of operations before ICU ( OR = 1.409 ), sedation at admission ( OR = 2.094 ), platelets, potassium ions and imaging edema. The AUC of the model was 0.748 ( 95% CI : 0.699–0.797 ), the calibration was good ( Hosmer-Lemeshow p = 0.740 ), and the average AUC of Bootstrap validation was 0.759. The decision curve analysis showed that the model continued to produce positive net benefits within the clinically relevant risk threshold range ( 0.05–0.25 ). Sensitivity analysis showed that the model maintained stable performance under different time windows, different TBI severity subgroups and different outcome definitions. Conclusion The prediction model established in this study includes 11 variables. This model is still good in discriminant ability, and the calibration is also reliable. The overall performance is also relatively strong. These variables are indicators that will be collected in clinical routines, so it is more convenient and practical to obtain. It can be used to evaluate the risk of early epilepsy in TBI patients, and may also be helpful in formulating personalized treatment plans, which may have some improvement in the prognosis of patients.