Population-based study of rare epilepsy incidence in a US urban population.

OBJECTIVE: This study was undertaken to estimate incidence of rare epilepsies and compare with literature. METHODS: We used electronic health record text search to identify children with 28 rare epilepsies in New York City (2010-2014). We estimated cumulative incidence and compared with literature. RESULTS: Eight of 28 rare epilepsies had five or more prior estimates, and our measurements were within the published range for all. The most common were infantile epileptic spasms syndrome (1 in 2920 live births), Lennox-Gastaut syndrome (1 in 9690), and seizures associated with tuberous sclerosis complex (1 in 14 300). Fifteen of 28 had fewer than five prior estimates, and of these, we provided additional estimates for early infantile developmental and epileptic encephalopathy (1 in 32 700), epilepsy with myoclonic-atonic seizures (1 in 34 100), Sturge-Weber syndrome plus seizures/epilepsy (1 in 40 900), epilepsy in infancy with migrating focal seizures (1 in 54 500), Aicardi syndrome plus seizures/epilepsy (1 in 71 600), hypothalamic hamartoma with seizures (1 in 225 000), and Rasmussen syndrome (1 in 450 000). Five of 28 rare epilepsies had no prior estimates, and of these, we provided a new estimate for developmental/epileptic encephalopathy with spike-and-wave activation in sleep and/or continuous spikes and waves during sleep (1 in 34 100). Data were limited for the remaining 12 rare epilepsies, which were all genetic epilepsies, including PCDH19, CDKL5, Alpers disease, SCN8A, KCNQ2, SCN2A, GLUT1 deficiency, Phelan-McDermid syndrome, myoclonic epilepsy with ragged-red fibers, dup15q syndrome, ring chromosome 14, and ring chromosome 20. SIGNIFICANCE: We estimated the incidence of rare epilepsies using population-based electronic health record data and literature review. More research is needed to better estimate the incidence of genetic epilepsies with nonspecific clinical features. Electronic health records may be a valuable data source for studying rare epilepsies and other rare diseases, particularly as genetic testing becomes more widely adopted.

Creating rare epilepsy cohorts using keyword search in electronic health records.

OBJECTIVE: Administrative codes to identify people with rare epilepsies in electronic health records are limited. The current study evaluated the use of keyword search as an alternative method for rare epilepsy cohort creation using electronic health records data. METHODS: Data included clinical notes from encounters with International Classification of Diseases, Ninth Revision (ICD-9) codes for seizures, epilepsy, and/or convulsions during 2010-2014, across six health care systems in New York City. We identified cases with rare epilepsies by searching clinical notes for keywords associated with 33 rare epilepsies. We validated cases via manual chart review. We compared the performance of keyword search to manual chart review using positive predictive value (PPV), sensitivity, and F-score. We selected an initial combination of keywords using the highest F-scores. RESULTS: Data included clinical notes from 77 924 cases with ICD-9 codes for seizures, epilepsy, and/or convulsions. The all-keyword search method identified 6095 candidates, and manual chart review confirmed that 2068 (34%) had a rare epilepsy. The initial combination method identified 1862 cases with a rare epilepsy, and this method performed as follows: PPV median = .64 (interquartile range [IQR] = .50-.81, range = .20-1.00), sensitivity median = .93 (IQR = .76-1.00, range = .10-1.00), and F-score median = .71 (IQR = .63-.85, range = .18-1.00). Using this method, we identified four cohorts of rare epilepsies with over 100 individuals, including infantile spasms, Lennox-Gastaut syndrome, Rett syndrome, and tuberous sclerosis complex. We identified over 50 individuals with two rare epilepsies that do not have specific ICD-10 codes for cohort creation (epilepsy with myoclonic atonic seizures, Sturge-Weber syndrome). SIGNIFICANCE: Keyword search is an effective method for cohort creation. These findings can improve identification and surveillance of individuals with rare epilepsies and promote their referral to specialty clinics, clinical research, and support groups.

Multicenter Assessment of Sturge-Weber Syndrome: A Retrospective Study of Variations in Care and Use of Natural History Data.

BACKGROUND: We summarize the history of individuals with Sturge-Weber syndrome (SWS) to inform clinical trial design and identify variations in care. METHODS: We performed retrospective chart review of individuals with SWS from centers in New York City. We characterized data quality using a novel scoring system. For 13 clinical concepts, we evaluated if data were present and if they were of high quality. RESULTS: We included 26 individuals with SWS (58% female; median age at initial visit 7 years; absolute range 1 month to 56 years]). Twenty-two had nevus flammeus, 13 glaucoma, four homonymous hemianopia, and 15 hemiparesis. Nineteen of 21 had at least one confirmed seizure with a known first seizure date, all before 24 months. Most (18 of 26, 69%) epilepsy was controlled. A plurality (10 of 23, 43%) had either normal cognitive function or mild cognitive delays. Aspirin use varied by site (P = 0.02)-at four sites, use was 0% (zero of three), 0% (zero of four), 80% (four of five), and 64% (nine of 14). Data were present for more than 75% of cases for 11 of 13 clinical concepts (missing: age of diagnosis, age of glaucoma onset). There were gaps in level of detail for motor impairments, glaucoma severity, seizure history, cognition, and medication history. CONCLUSIONS: Clinical charts have important gaps in the level of detail around core SWS clinical features, limiting value for some natural history studies. Any clinical trial in SWS designed to prevent epilepsy should begin in the first year of life. Variations in use of aspirin suggest de facto clinical equipoise and warrant a comparative effectiveness study.

Automated identification and quality measurement for pediatric convulsive status epilepticus.

OBJECTIVE: Treatment delays for refractory convulsive status epilepticus (RCSE) are associated with worse outcomes. In the United States, treatment for pediatric RCSE is slower than guidelines recommend. To address this gap, the American Academy of Neurology and Child Neurology Society (AAN/CNS) developed a quality measure: the percentage of RCSE patients that receive third-line treatment within 60 minutes. We aimed to develop computable phenotypes for convulsive status epilepticus (CSE) and RCSE to automate calculation of the quality measure. METHODS: From an observational cohort of children presenting to the emergency department for seizures or epilepsy, we identified presentations of RCSE and its precursors: CSE and benzodiazepine-resistant status epilepticus (BRSE). These served as a gold standard for computable phenotype development. Using multivariate analyses, we constructed and evaluated statistical models for case identification. We then evaluated adherence to the AAN/CNS RCSE quality measure. RESULTS: From 664 charts, we identified 56 patients with CSE, 36 with BRSE, and 18 with RCSE. Four predictors were used: International Classification of Diseases (ICD) codes, and receiving first-, second-, or third-line agents shortly after presentation to the emergency department (ED). Combinations of these predictors identified CSE with 84% sensitivity and 81% positive predictive value (PPV), BRSE with 67% sensitivity and 89% PPV, and RCSE with 94% sensitivity and 85% PPV. Median (interquartile range [IQR]) time to treatment for first-line agent was 13 (5-27) minutes for CSE, second-line for BRSE was 24 (9.5-43.5) minutes, and third-line for RCSE was 52 (27-87) minutes. Sixty percent of RCSE patients received a third-line agent within 60 minutes of ED arrival. SIGNIFICANCE: RCSE and its precursors can be identified automatically with high fidelity allowing automated calculation of time to treatment and the RCSE quality measure. This has the potential to facilitate quality improvement work and improve care for RCSE.

Clinical and electrographic features of persistent seizures and status epilepticus associated with anti-NMDA receptor encephalitis (anti-NMDARE).

Based on a multicenter cohort of people with anti-NMDA receptor encephalitis (anti-NMDARE), we describe seizure phenotypes, electroencephalographic (EEG) findings, and anti-seizure treatment strategies. We also investigated whether specific electrographic features are associated with persistent seizures or status epilepticus after acute presentation. In this retrospective cohort study, we reviewed records of children and adults with anti-NMDARE between 2010 and 2014 who were included in the Rare Epilepsy of New York City database, which included the text of physician notes from five academic medical centers. Clinical history (e.g., seizure semiology) and EEG features (e.g., background organization, slowing, epileptiform activity, seizures, sleep architecture, extreme delta brush) were abstracted. We compared clinical features associated with persistent seizures (ongoing seizures after one month from presentation) and status epilepticus, using bivariate and multivariable analyses. Among the 38 individuals with definite anti-NMDARE, 32 (84%) had seizures and 29 (76%) had seizures captured on EEG. Electrographic-only seizures were identified in five (13%) individuals. Seizures started at a median of four days after initial symptoms (IQR: 3-6 days). Frontal lobe-onset focal seizures were most common (n=12; 32%). Most individuals (31/38; 82%) were refractory to anti-seizure medications. Status epilepticus was associated with younger age (15 years [9-20] vs. 23 years [18-27]; p=0.04) and Hispanic ethnicity (30 [80%] vs. 8 [36%]; p=0.04). Persistent seizures (ongoing seizures after one month from presentation) were associated with younger age (nine years [3-14] vs. 22 years [15-28]; p<0.01). Measured electrographic features were not associated with persistent seizures. Seizures associated with anti-NMDARE are primarily focal seizures originating in the frontal lobes. Younger patients may be at increased risk of epileptogenesis and status epilepticus. Continuous EEG monitoring helps identify subclinical seizures, but specific EEG findings may not predict the severity or persistence of seizures during hospitalization.

Automated detection of sudden unexpected death in epilepsy risk factors in electronic medical records using natural language processing.

OBJECTIVE: Sudden unexpected death in epilepsy (SUDEP) is an important cause of mortality in epilepsy. However, there is a gap in how often providers counsel patients about SUDEP. One potential solution is to electronically prompt clinicians to provide counseling via automated detection of risk factors in electronic medical records (EMRs). We evaluated (1) the feasibility and generalizability of using regular expressions to identify risk factors in EMRs and (2) barriers to generalizability. METHODS: Data included physician notes for 3000 patients from one medical center (home) and 1000 from five additional centers (away). Through chart review, we identified three SUDEP risk factors: (1) generalized tonic-clonic seizures, (2) refractory epilepsy, and (3) epilepsy surgery candidacy. Regular expressions of risk factors were manually created with home training data, and performance was evaluated with home test and away test data. Performance was evaluated by sensitivity, positive predictive value, and F-measure. Generalizability was defined as an absolute decrease in performance by <0.10 for away versus home test data. To evaluate underlying barriers to generalizability, we identified causes of errors seen more often in away data than home data. To demonstrate how small revisions can improve generalizability, we removed three "boilerplate" standard text phrases from away notes and repeated performance. RESULTS: We observed high performance in home test data (F-measure range = 0.86-0.90), and low to high performance in away test data (F-measure range = 0.53-0.81). After removing three boilerplate phrases, away performance improved (F-measure range = 0.79-0.89) and generalizability was achieved for nearly all measures. The only significant barrier to generalizability was use of boilerplate phrases, causing 104 of 171 errors (61%) in away data. SIGNIFICANCE: Regular expressions are a feasible and probably a generalizable method to identify variables related to SUDEP risk. Our methods may be implemented to create large patient cohorts for research and to generate electronic prompts for SUDEP counseling.