Diagnosing Schizophrenia With Machine Learning

"Diagnosing serious mental illnesses like schizophrenia can take up to 10 years after the first episode. As one can imagine, this can have very negative repercussions. Catching psychosis early on helps create a better prognosis. Although there are many cases where those who have a late diagnosis go on to lead fulfilling lives, the hope is that future patients can get to that place faster."

"A machine learning model was then developed on the existing text data to train the model to assess whether future patients would develop schizophrenia within five years. By doing so, the machine learning model used 1092 "predictors," which are factors related to the disorder that are seen in people who already have it. When predictors are identified in a cluster or increased severity (or even a decreased frequency), a machine learning model can make predictions that a person would have schizophrenia."

"Some of the major text predictors for identifying schizophrenia included: hearing voices, discharge or admission from an inpatient hospital visit, contact (or no contact) with female friends, inpatient admission, wake or sleep time in the morning, words like "play" and "the game" (like board games and interacting with hospital staff), and the ability to explain their symptoms, to name a few"

Diagnosing serious mental illnesses like schizophrenia can take up to 10 years after the first episode, worsening outcomes. Early detection of psychosis improves prognosis. A cohort of 24,449 participants from the Psychiatric Services of the Central Denmark Region (2013–2016) had hospital notes analyzed for text and keywords. A machine learning model was trained on existing clinical text using 1,092 predictors to assess risk of developing schizophrenia within five years. Predictors included hearing voices, inpatient admissions, contact patterns with female friends, morning wake or sleep times, words such as 'play' and 'the game', and ability to explain symptoms. Models showed better performance for schizophrenia than for bipolar disorder.

Read at Psychology Today