Using natural language processing to identify patients with encephalopathy admitted to an intensive care unit

Delirium is a well-recognized form of encephalopathy, but its’ diagnosis remains under-recorded in hospital health records.
Using a machine learning approach and aligned to ICD-9 codes, researchers identified 7.5% of the 46,520 different patients in the MIMIC-III dataset with definite encephalopathy (group 1), 45% with possible encephalopathy (group 2), and 38% without encephalopathy (group 3). Patients with definite and possible encephalopathy had a longer hospital stay, a higher mortality rate and more co-morbid conditions than those without encephalopathy.
According to the authors, natural language processing tools result in a high probability of identifying patients with underrecognized critical illnesses, like encephalopathy.