Background and Objectives: to identify predictors of progression to refractory status epilepticus (RSE) using a machine learning technique. Methods: Consecutive patients aged >= 14 years with SE registered in a 9-years period at Modena Academic Hospital were included in the analysis. We evaluated the risk of progression to RSE using logistic regression and a machine learning analysis by means of classification and regression tree analysis (CART) to develop a predictive model of progression to RSE. Results: 705 patients with SE were included in the study; of those, 33 % (233/705) evolved to RSE. The progression to RSE was an independent risk factor for 30-day mortality, with an OR adjusted for previously identified possible univariate confounders of 4.086 (CI 95 % 2.390-6.985; p < 0.001). According to CART the most important variable predicting evolution to RSE was the impaired consciousness before treatment, followed by acute symptomatic hypoxic etiology and periodic EEG patterns. The decision tree identified 14 nodes with a risk of evolution to RSE ranging from 1.5 % to 90.8 %. The overall percentage of success in classifying patients of the decision tree was 79.4 %; the percentage of accurate prediction was high, 94.1 %, for those patients not progressing to RSE and moderate, 49.8 %, for patients evolving to RSE. Conclusions: Decision-tree analysis provided a meaningful risk stratification based on few variables that are easily obtained at SE first evaluation: consciousness before treatment, etiology, and severe EEG patterns. CART models must be viewed as potential new method for the stratification RSE at single subject level deserving further exploration and validation.
Progression to refractory status epilepticus: A machine learning analysis by means of classification and regression tree analysis / Meletti, S.; Giovannini, G.; Lattanzi, S.; Zaboli, A.; Orlandi, N.; Turcato, G.; Brigo, F.. - In: EPILEPSY & BEHAVIOR. - ISSN 1525-5050. - 161:(2024), pp. N/A-N/A. [10.1016/j.yebeh.2024.110005]
Progression to refractory status epilepticus: A machine learning analysis by means of classification and regression tree analysis
Meletti S.;Orlandi N.;
2024
Abstract
Background and Objectives: to identify predictors of progression to refractory status epilepticus (RSE) using a machine learning technique. Methods: Consecutive patients aged >= 14 years with SE registered in a 9-years period at Modena Academic Hospital were included in the analysis. We evaluated the risk of progression to RSE using logistic regression and a machine learning analysis by means of classification and regression tree analysis (CART) to develop a predictive model of progression to RSE. Results: 705 patients with SE were included in the study; of those, 33 % (233/705) evolved to RSE. The progression to RSE was an independent risk factor for 30-day mortality, with an OR adjusted for previously identified possible univariate confounders of 4.086 (CI 95 % 2.390-6.985; p < 0.001). According to CART the most important variable predicting evolution to RSE was the impaired consciousness before treatment, followed by acute symptomatic hypoxic etiology and periodic EEG patterns. The decision tree identified 14 nodes with a risk of evolution to RSE ranging from 1.5 % to 90.8 %. The overall percentage of success in classifying patients of the decision tree was 79.4 %; the percentage of accurate prediction was high, 94.1 %, for those patients not progressing to RSE and moderate, 49.8 %, for patients evolving to RSE. Conclusions: Decision-tree analysis provided a meaningful risk stratification based on few variables that are easily obtained at SE first evaluation: consciousness before treatment, etiology, and severe EEG patterns. CART models must be viewed as potential new method for the stratification RSE at single subject level deserving further exploration and validation.File | Dimensione | Formato | |
---|---|---|---|
1-s2.0-S152550502400386X-main.pdf
Open access
Tipologia:
Versione pubblicata dall'editore
Dimensione
1.33 MB
Formato
Adobe PDF
|
1.33 MB | Adobe PDF | Visualizza/Apri |
Pubblicazioni consigliate
I metadati presenti in IRIS UNIMORE sono rilasciati con licenza Creative Commons CC0 1.0 Universal, mentre i file delle pubblicazioni sono rilasciati con licenza Attribuzione 4.0 Internazionale (CC BY 4.0), salvo diversa indicazione.
In caso di violazione di copyright, contattare Supporto Iris