Machine Learning Model Predicts Schizophrenia More Accurately Than Bipolar Disorder: JAMA

To forecast the diagnostic conversion to schizophrenia or bipolar disorder at five years, researchers used two machine learning models: elastic net regularized logistic regression and extreme gradient boosting (XGBoost). The performance of the model was assessed with the area under receiver operating characteristic curve (AUROC).

Key Findings

Conversion to schizophrenia or bipolar disorder was predicted by XGBoost model with an AUROC of 0.70 (95% CI, 0.70-0.70) for the training dataset and 0.64 (95% CI, 0.63-0.65) for the test dataset.

At a predicted positive rate of 4%, the model had:

• Sensitivity: 9.3%

• Specificity: 96.3%

• Positive Predictive Value (PPV): 13.0%

When specifically predicting schizophrenia, the model showed an increased AUROC of 0.80 (95% CI, 0.79-0.81), with:

• Sensitivity: 19.4%

• Specificity: 96.3%

• PPV: 10.8%

• For bipolar disorder, the model was less accurate, with an AUROC of 0.62 (95% CI, 0.61-0.63), sensitivity of 9.9%, specificity of 96.2%, and PPV of 8.4%.

The study authors concluded that machine learning algorithms, specifically XGBoost, were able to predict schizophrenia and bipolar disorder based on EHR information. With increased refinement and testing, AI models could be great assets in psychiatric facilities to help enhance early treatment and diagnosis and ultimately improve patient outcomes.

Reference:

Hansen L, Bernstorff M, Enevoldsen K, et al. Predicting Diagnostic Progression to Schizophrenia or Bipolar Disorder via Machine Learning. JAMA Psychiatry. Published online February 19, 2025. doi:10.1001/jamapsychiatry.2024.4702