Evaluating machine learning algorithms for prediction of treatment response for sleep disturbances in patients with schizophrenia: A post-hoc analysis from a randomized controlled trial

Area of Science:

• Psychiatry
• Computational Medicine
• Pharmacology

Background:

• Schizophrenia often involves sleep and circadian rhythm disturbances.
• Predicting treatment response in schizophrenia is crucial for effective management.
• Machine learning offers potential for personalized treatment prediction.

Purpose of the Study:

• To develop and compare machine learning algorithms for predicting treatment response in schizophrenia patients with sleep disturbances.
• To identify key clinical and biological factors influencing treatment outcomes.

Main Methods:

• A post-hoc analysis of a randomized controlled trial (NCT03075657) involving 120 schizophrenia patients.
• Development and comparison of multiple machine learning models: random forest, k-nearest neighbors, extreme gradient boosting, R part, and logistic regression.
• Utilized R language with specialized packages (mlbench, caret, MASS, rPART) for model creation and analysis.

Main Results:

• Logistic regression emerged as the best-fit model, achieving 90% prediction accuracy.
• Key predictors identified included predominant symptom domain (positive/negative), baseline urinary melatonin levels, and global Pittsburgh Sleep Quality Index (PSQI) score.
• The model demonstrated a specificity of 0.93, sensitivity of 0.45, and an Area Under the Receiver Operating Characteristic Curve (ROC) of 0.78.

Conclusions:

• Machine learning models, particularly logistic regression, show significant promise in predicting treatment response in schizophrenia.
• The identified predictors offer insights into patient stratification for targeted interventions.
• Translating these machine learning models into clinical practice could enhance patient care for sleep disturbances in schizophrenia.