Predicting Diagnostic Progression to Schizophrenia or Bipolar Disorder via Machine Learning.

Area of Science:

• Psychiatry
• Machine Learning
• Health Informatics

Background:

• Schizophrenia and bipolar disorder diagnoses are often delayed, impeding timely treatment.
• Early intervention is crucial as these conditions typically emerge in late adolescence or early adulthood.

Purpose of the Study:

• To evaluate machine learning models for predicting diagnostic progression to schizophrenia or bipolar disorder.
• To assess the utility of routine electronic health record (EHR) data for early detection.

Main Methods:

• A cohort study utilized EHR data from psychiatric services in the Central Denmark Region.
• Machine learning models (logistic regression, XGBoost) were trained on clinical data including diagnoses, medications, and notes.
• Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC).

Main Results:

• The XGBoost model achieved an AUROC of 0.64 on the test set for predicting transition to schizophrenia or bipolar disorder.
• Schizophrenia was predicted with higher accuracy (AUROC, 0.80) than bipolar disorder (AUROC, 0.62).
• Clinical notes were identified as particularly valuable predictors.

Conclusions:

• Machine learning models can predict diagnostic transitions to schizophrenia and bipolar disorder using routine EHR data.
• Early prediction is feasible, potentially enabling earlier treatment initiation.
• Schizophrenia prediction demonstrates higher efficacy compared to bipolar disorder prediction.