Predicting Transitions to Schizophrenia Spectrum disorders: A machine learning approach using Danish national Registers and genetic data

Identifier:  TFM:2529

Handle: https://hdl.handle.net/20.500.11797/TFM2529

Authors:  Salas Zamora, Júlia

Abstract:
Schizophrenia-spectrum disorders are among the most severe psychiatric conditions, often preceded by milder diagnoses such as mood or anxiety disorders. Early prediction of diagnostic transitions improve outcomes. This project applies machine learning to predict transitions from ICD-10 F3–F7 diagnoses to schizophrenia-spectrum disorders (F2), using Danish national register data and polygenic risk scores (PRS). Features were derived from multiple registers, capturing clinical history, socioeconomic background, parental psychiatric status, medication, and genetic data. Logistic Regression and XGBoost models were trained and evaluated using ROC-AUC, sensitivity, specificity, and calibration. Both weighted and unweighted modeling strategies were explored to account for class imbalance.