Predicting treatment resistance in schizophrenia patients: Machine learning highlights the role of early...

Area of Science:

• Psychiatry
• Computational Neuroscience
• Medical Informatics

Background:

• Identifying patients with treatment-resistant schizophrenia (TRS) is crucial for tailored therapies and understanding disease mechanisms.
• Machine learning offers potential for predicting TRS risk by analyzing complex patient data.

Purpose of the Study:

• To explore the impact of demographic and clinical factors on predicting treatment-resistant schizophrenia (TRS) using machine learning.
• To identify both known and novel risk factors associated with a high-risk TRS profile.

Main Methods:

• A retrospective study analyzed 500 patient records from the University Hospital Group for Paris Psychiatry.
• Natural Language Processing (NLP) extracted features from discharge summaries and medical narratives.
• Three machine learning models (XGBoost, logistic elastic net, logistic regression) were compared for TRS prediction.

Main Results:

• Key predictors for high-risk TRS included early age at first psychiatric contact, antipsychotic treatment interruptions due to non-adherence, absence of positive symptoms, educational problems, and adolescent mental disorders.
• Age at first psychiatric contact and medication non-adherence showed significant associations with TRS outcome.
• Machine learning models effectively leveraged NLP-processed clinical notes for TRS prediction.

Conclusions:

• Early identification of specific patient characteristics can predict treatment-resistant schizophrenia (TRS) risk.
• Findings support the use of early clinical and demographic features for TRS prediction.
• NLP combined with machine learning shows promise for clinical decision support in early TRS detection.