– Department of Pharmacy, Medical Security Center, Chinese People’s Liberation Army General Hospital, Beijing, Beijing, China
Background: The hospital information system (HIS) contains massive information about epileptic seizures（ES). However, due to the variety of data types, it is difficult to extract. In this study, based on the adverse drug event active surveillance and assessment system-II(ADE-ASAS-II), we designed and established an automatic monitoring module for seizures from multiple aspects, including symptom descriptions, EEG examinations, neuroimaging, records of anti-epileptic drugs and diagnosis.
Objectives: To establish an automatic monitoring module for epileptic seizures in hospitalized patients based on the ADE-ASAS-II, and to provide an efficient data mining tool for real-world research on large samples of epileptic seizures.
Methods: This module used text classification technology and decision tree to mine the cases with seizures. First, collect keywords about seizures in guidelines, literature, spontaneous reports and electronic medical records. Screen and classify those keywords into four subsets as the criteria for the classification of decision tree nodes. Optimize the module by custom and shielding functions of ADE-ASAS-II with test data and determine the best keyword sets as the module's alarm rules. Then, expand the sample size for monitoring, check the stability of the module.Finally, concisely describe the gender, age, seizure types and causes of positive cases.
Results: 5557 inpatients as the test data were manually reviewed. Through repeatedly tests, 37 alarm keywords in each branch of the decision tree were determined, and 12 keywords were entered to the EMR title shielding list. The positive predictive value of the module is 13.86%, and the recall rate is 100%. We monitored 14,549 hospitalized patients in May 2021 for validating the stability of the module, and 90 patients with epileptic seizures were screened by inclusion and exclusion criteria, with a PPV of 14.6% and an incidence rate of 0.62%. 53 out of 90 cases were acute symptomatic seizures, and the most frequent seizure type is tonic-clonic. Nervous system neoplasm surgery, post stroke seizure, and encephalitis autoimmune were the most common identifiable etiologies.
Conclusions: The application of the automatic seizure monitoring module based on ADE-ASAS-II can eliminate more than 95% of irrelevant cases, which can efficiently, comprehensively and rapidly obtain target cases in the hospital, providing a reliable data mining tool for real-world research on large samples of epileptic seizures.
