Distinguishing different psychiatric disorders using DDx-PRS.

Area of Science:

• Psychiatric Genomics
• Polygenic Risk Prediction
• Differential Diagnosis

Background:

• Distinguishing between clinically related psychiatric disorders (schizophrenia, bipolar disorder, major depressive disorder) is challenging, particularly at disease onset.
• Existing polygenic risk score (PRS) methods are primarily designed for case-control comparisons, not for differentiating between related disorders.

Approach:

• Introduced Differential Diagnosis-Polygenic Risk Score (DDx-PRS), a novel method that jointly estimates posterior probabilities for multiple diagnostic categories.
• DDx-PRS models variance-covariance structures across disorders, integrating PRS for individual disorders and prior clinical probabilities.
• The method utilizes only summary-level training data, eliminating the need for tuning data, thus facilitating clinical implementation.

Key Points:

• Simulations demonstrated DDx-PRS is well-calibrated and effective in distinguishing diagnostic categories, outperforming marginal analysis approaches.
• Application to large-scale Psychiatric Genomics Consortium data (N > 1 million) showed DDx-PRS achieved robust performance (AUCs 0.59-0.68) for schizophrenia, bipolar disorder, major depressive disorder, and control groups.
• DDx-PRS performance is comparable to methods requiring tuning data and shows potential for clinical utility, especially with larger training datasets.

Conclusions:

• DDx-PRS is an effective and well-calibrated method for genetically distinguishing between clinically related psychiatric disorders.
• The method's ability to leverage summary-level data and its strong performance suggest significant potential for clinical application in challenging differential diagnoses.