Diagnosing schizophrenia with routine blood tests: a comparative analysis of machine learning algorithms

Area of Science:

• Biomarker discovery
• Computational psychiatry
• Clinical diagnostics

Background:

• Schizophrenia diagnosis relies on clinical criteria, lacking objective biomarkers.
• Peripheral blood biomarkers and machine learning (ML) offer potential for improved diagnostic accuracy.
• Current diagnostic methods face challenges in objectivity and effectiveness.

Purpose of the Study:

• To develop and evaluate ML models for schizophrenia diagnosis using peripheral blood biomarkers.
• To identify optimal biomarker subsets for distinguishing schizophrenia patients from healthy controls.
• To assess the diagnostic performance and potential clinical utility of ML-based approaches.

Main Methods:

• Retrospective case-control study with 203 schizophrenia patients and 192 healthy controls.
• Extraction and imputation of routine hematological and biochemical parameters.
• Application of Grey Wolf Optimization (GWO) for biomarker selection and various ML models (RF, XGBoost, SVM, KNN, LR) with 10-fold cross-validation.

Main Results:

• XGBoost achieved the highest accuracy (95.90%) and specificity (95.54%) post-GWO optimization.
• Random Forest showed strong performance with 94.95% accuracy and 96.25% recall.
• Key distinguishing biomarkers included total protein, glucose, iron, creatine kinase, total bilirubin, uric acid, calcium, and sodium.

Conclusions:

• ML models utilizing routine blood parameters demonstrate high diagnostic accuracy for schizophrenia.
• The identified biomarkers and developed models offer a cost-effective approach compared to expensive methods.
• Further external validation is recommended to confirm generalizability and clinical applicability.