Extending schizophrenia diagnostic model to predict schizotypy in first-degree relatives

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Schizophrenia poses a significant public health challenge.
• Early detection and understanding of vulnerability are crucial for intervention.
• Resting-state functional magnetic resonance imaging (rs-fMRI) offers insights into brain connectivity.

Purpose of the Study:

• To apply a machine learning algorithm (EMPaSchiz) trained on schizophrenia patients and healthy controls to first-degree relatives of schizophrenia patients.
• To investigate if the model can predict schizophrenia vulnerability in individuals without active psychosis.
• To correlate the model's predictions with schizotypal personality scores.

Main Methods:

• Utilized a pre-developed machine learning algorithm, EMPaSchiz, trained on rs-fMRI data.
• Applied the EMPaSchiz model to a cohort of first-degree relatives of schizophrenia patients.
• Assessed participants for schizotypal personality traits.

Main Results:

• The EMPaSchiz model classified some relatives as having a higher probability of schizophrenia.
• Individuals classified by EMPaSchiz as high-risk exhibited significantly higher schizotypal personality scores.
• EMPaSchiz probability scores for schizophrenia status were significantly correlated with schizotypal personality scores.

Conclusions:

• Machine-learned diagnostic models can predict state-independent vulnerability to schizophrenia.
• The findings suggest EMPaSchiz can identify individuals at risk even before full clinical diagnosis.
• This approach holds promise for early identification and intervention strategies in at-risk populations.