Analysis of Semantic Drifting in Diagnostic Texts for Sleep Disorders

Along with the diagnostic process for sleep disorders, a diversity of clinical assessments are required to diagnose the concrete type of sleep disorder. For an accurate diagnosis, a variety of medical examinations are conducted, whose results are documented in clinical records. The textual records form the basis for disease categorization and cohort creation using the International Classification for Sleep Disorder (ICSD-III). However, textual records generated through patient-physician interaction may contain different types of biases caused by the heterogeneous nature of medical specialties and the used vocabulary in the documentation. In this work, we will analyze different types of semantic biases and driftings in the context of sleep disorder diagnosis using the Bern Sleep Database. The database contains documents describing the clinical history, referring reasons and results from different assessments (polysomnography (PSG), multiple sleep latency test (MSLT), maintenance of wakefulness test (MWT), actigraphy (Wrist, PLMS)) from more than 6000 patients. An enhanced text-based ICSD-III classification (Multichannel CNN and Hierachical attentive networks) using a standardized concept representation will be proposed. We will analyze the influence of semantic bias on the accuracy of this automatic classification.