Validation of the diagnosis of Lennox-Gastaut syndrome: Challenges and lessons learned

Lennox-Gastaut Syndrome (LGS) is a severe developmental and epileptic encephalopathy characterized by multiple seizure types, cognitive impairment, and distinct electroencephalographic (EEG) findings. However, evolving diagnostic criteria, heterogeneous clinical features, and inconsistent electronic health record (EHR) documentation complicate accurate diagnosis and cohort identification. As part of a Patient-Centered Outcomes Research Institute-sponsored study developing EHR-based computable phenotypes for LGS, we established a reference standard through retrospective chart review and epileptologist classification.

A multicenter retrospective chart review was conducted across seven pediatric epilepsy centers. Stratified sampling using PCORnet data selected patients based on the International Classification of Diseases, 10th Revision diagnosis code for LGS or other intractable epilepsy diagnoses between 2016 and 2022. Clinical teams abstracted seizure history, EEG findings, developmental status, and treatment history. Two epileptologists independently reviewed each case as classic LGS, pragmatic LGS, or not LGS. Discordant cases underwent adjudication by a senior epileptologist. Cases with incomplete or unclear data underwent repeat abstraction and review.

Of 997 identified patients, 935 underwent classification after exclusions. Initial epileptologist disagreement occurred in 105 cases (11%, unweighted Cohen kappa = 82%). Following adjudication, 130 cases underwent reabstraction to retrieve additional EEG reports and clinical details. Rereview reclassified 67 cases initially considered "not LGS": 33 classic LGS, 12 pragmatic LGS, and 22 excluded for insufficient data. Key EHR challenges included inconsistent EEG reporting, incomplete seizure type characterization, fragmented documentation of cognitive impairment, unclear age at onset, and variable terminology across institutions and providers. Slow spike-and-wave EEG findings were the most frequently missing diagnostic element.

Validating LGS diagnoses using EHR data is complex and resource intensive. Standardized terminology, structured documentation, and rigorous multicenter validation may improve diagnostic accuracy, research cohort identification, computable phenotype development, and longitudinal care for individuals with LGS. Our results highlight important operational lessons for multicenter rare epilepsy studies and underscore the need for improved diagnosis pathways and longitudinal care systems for those with LGS and their families.