Validation of an algorithm of time-dependent electro-clinical risk stratification for electrographic seizures (TERSE) in critically ill patients.

OBJECTIVE: The clinical implementation of continuous electroencephalography (CEEG) monitoring in critically ill patients is hampered by the substantial burden of work that it entails for clinical neurophysiologists. Solutions that might reduce this burden, including by shortening the duration of EEG to be recorded, would help its widespread adoption. Our aim was to validate a recently described algorithm of time-dependent electro-clinical risk stratification for electrographic seizure (ESz) (TERSE) based on simple clinical and EEG features. METHODS: We retrospectively reviewed the medical records and EEG recordings of consecutive patients undergoing CEEG between October 1, 2015 and September, 30 2016 and assessed the sensitivity of TERSE for seizure detection, as well as the reduction in EEG time needed to be reviewed. RESULTS: In a cohort of 407 patients and compared to full CEEG review, the model allowed the detection of 95% of patients with ESz and 97% of those with electrographic status epilepticus. The amount of CEEG to be recorded to detect ESz was reduced by two-thirds, compared to the duration of CEEG taht was actually recorded. CONCLUSIONS: TERSE allowed accurate time-dependent ESz risk stratification with a high sensitivity for ESz detection, which could substantially reduce the amount of CEEG to be recorded and reviewed, if applied prospectively in clinical practice. SIGNIFICANCE: Time-dependent electro-clinical risk stratification, such as TERSE, could allow more efficient practice of CEEG and its more widespread adoption. Future studies should aim to improve risk stratification in the subgroup of patients with acute brain injury and absence of clinical seizures.

Detecting abnormal electroencephalograms using deep convolutional networks.

OBJECTIVES: Electroencephalography (EEG) is a central part of the medical evaluation for patients with neurological disorders. Training an algorithm to label the EEG normal vs abnormal seems challenging, because of EEG heterogeneity and dependence of contextual factors, including age and sleep stage. Our objectives were to validate prior work on an independent data set suggesting that deep learning methods can discriminate between normal vs abnormal EEGs, to understand whether age and sleep stage information can improve discrimination, and to understand what factors lead to errors. METHODS: We train a deep convolutional neural network on a heterogeneous set of 8522 routine EEGs from the Massachusetts General Hospital. We explore several strategies for optimizing model performance, including accounting for age and sleep stage. RESULTS: The area under the receiver operating characteristic curve (AUC) on an independent test set (n = 851) is 0.917 marginally improved by including age (AUC = 0.924), and both age and sleep stages (AUC = 0.925), though not statistically significant. CONCLUSIONS: The model architecture generalizes well to an independent dataset. Adding age and sleep stage to the model does not significantly improve performance. SIGNIFICANCE: Insights learned from misclassified examples, and minimal improvement by adding sleep stage and age suggest fruitful directions for further research.

Decision analysis of intracranial monitoring in non-lesional epilepsy.

PURPOSE: Up to one third of epilepsy patients develop pharmacoresistant seizures and many benefit from resective surgery. However, patients with non-lesional focal epilepsy often require intracranial monitoring to localize the seizure focus. Intracranial monitoring carries operative morbidity risk and does not always succeed in localizing the seizures, making the benefit of this approach less certain. We performed a decision analysis comparing three strategies for patients with non-lesional focal epilepsy: (1) intracranial monitoring, (2) vagal nerve stimulator (VNS) implantation and (3) medical management to determine which strategy maximizes the expected quality-adjusted life years (QALYs) for our base cases. METHOD: We constructed two base cases using parameters reported in the medical literature: (1) a young, otherwise healthy patient and (2) an elderly, otherwise healthy patient. We constructed a decision tree comprising strategies for the treatment of non-lesional epilepsy and two clinical outcomes: seizure freedom and no seizure freedom. Sensitivity analyses of probabilities at each branch were guided by data from the medical literature to define decision thresholds across plausible parameter ranges. RESULTS: Intracranial monitoring maximizes the expected QALYs for both base cases. The sensitivity analyses provide estimates of the values of key variables, such as the surgical risk or the chance of localizing the focus, at which intracranial monitoring is no longer favored. CONCLUSION: Intracranial monitoring is favored over VNS and medical management in young and elderly patients over a wide, clinically-relevant range of pertinent model variables such as the chance of localizing the seizure focus and the surgical morbidity rate.