Background: The objective of this study was to validate the value of the Status Epilepticus Severity Score (STESS) in the prediction of the risk of in-hospital mortality in patients with nonhypoxic status epilepticus (SE) using a machine learning analysis. Methods: We included consecutive patients with nonhypoxic SE (aged ≥ 16 years) admitted from 2013 to 2021 at the Modena Academic Hospital. A decision tree analysis was performed using in-hospital mortality as a dependent variable and the STESS predictors as input variables. We evaluated the accuracy of STESS in predicting in-hospital mortality using the area under the receiver operating characteristic curve (AUROC) with 95% confidence interval (CI). Results: Among 629 patients with SE, the in-hospital mortality rate was 23.4% (147 of 629). The median STESS in the entire cohort was 2.9 (SD 1.6); it was lower in surviving compared with deceased patients (2.7, SD 1.5 versus 3.9, SD 1.6; p < 0.001). Of deceased patients, 82.3% (121 of 147) had scores of 3-6, whereas 17.7% (26 of 147) had scores of 0-2 (p < 0.001). STESS was accurate in predicting mortality, with an AUROC of 0.688 (95% CI 0.641-0.734) only slightly reduced after bootstrap resampling. The most significant predictor was the seizure type, followed by age and level of consciousness at SE onset. Nonconvulsive SE in coma and age ≥ 65 years predicted a higher risk of mortality, whereas generalized convulsive SE and age < 65 years were associated with a lower risk of death. The decision tree analysis using STESS variables correctly classified 90% of survivors and 34% of nonsurvivors after the SE, with an overall risk of error of 23.1%. Conclusions: This validation study using a machine learning system showed that STESS is a valuable prognostic tool. The score appears particularly accurate and effective in identifying patients who are alive at discharge (high negative predictive value), whereas it has a lower predictive value for in-hospital mortality.
Retrospective External Validation of the Status Epilepticus Severity Score (STESS) to Predict In-hospital Mortality in Adults with Nonhypoxic Status Epilepticus: A Machine Learning Analysis / Brigo, Francesco; Turcato, Gianni; Lattanzi, Simona; Orlandi, Niccolò; Turchi, Giulia; Zaboli, Arian; Giovannini, Giada; Meletti, Stefano. - In: NEUROCRITICAL CARE. - ISSN 1541-6933. - (2022). [10.1007/s12028-022-01610-3]
Retrospective External Validation of the Status Epilepticus Severity Score (STESS) to Predict In-hospital Mortality in Adults with Nonhypoxic Status Epilepticus: A Machine Learning Analysis
Lattanzi, Simona;
2022-01-01
Abstract
Background: The objective of this study was to validate the value of the Status Epilepticus Severity Score (STESS) in the prediction of the risk of in-hospital mortality in patients with nonhypoxic status epilepticus (SE) using a machine learning analysis. Methods: We included consecutive patients with nonhypoxic SE (aged ≥ 16 years) admitted from 2013 to 2021 at the Modena Academic Hospital. A decision tree analysis was performed using in-hospital mortality as a dependent variable and the STESS predictors as input variables. We evaluated the accuracy of STESS in predicting in-hospital mortality using the area under the receiver operating characteristic curve (AUROC) with 95% confidence interval (CI). Results: Among 629 patients with SE, the in-hospital mortality rate was 23.4% (147 of 629). The median STESS in the entire cohort was 2.9 (SD 1.6); it was lower in surviving compared with deceased patients (2.7, SD 1.5 versus 3.9, SD 1.6; p < 0.001). Of deceased patients, 82.3% (121 of 147) had scores of 3-6, whereas 17.7% (26 of 147) had scores of 0-2 (p < 0.001). STESS was accurate in predicting mortality, with an AUROC of 0.688 (95% CI 0.641-0.734) only slightly reduced after bootstrap resampling. The most significant predictor was the seizure type, followed by age and level of consciousness at SE onset. Nonconvulsive SE in coma and age ≥ 65 years predicted a higher risk of mortality, whereas generalized convulsive SE and age < 65 years were associated with a lower risk of death. The decision tree analysis using STESS variables correctly classified 90% of survivors and 34% of nonsurvivors after the SE, with an overall risk of error of 23.1%. Conclusions: This validation study using a machine learning system showed that STESS is a valuable prognostic tool. The score appears particularly accurate and effective in identifying patients who are alive at discharge (high negative predictive value), whereas it has a lower predictive value for in-hospital mortality.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.