A new machine learning-derived screening measure for differentiating bipolar from unipolar mood disorders

Area of Science:

• Psychiatry and Mental Health
• Diagnostic Tools and Measurement

Background:

• Current screening measures for bipolar disorder may lack diagnostic sensitivity due to equivalent weighting of criteria.
• The Diagnostic and Statistical Manual of Mental Disorders, 5th Edition (DSM-5) also assigns equal weight to all diagnostic criteria.
• A more discriminating measure is needed to improve the accuracy of bipolar disorder classification.

Purpose of the Study:

• To develop a new diagnostic measure for bipolar disorder.
• To examine if a weighted rating scale is superior to an equivalent weighting scale for diagnosis.

Main Methods:

• An international sample of 165 bipolar patients and 29 unipolar patients completed a 96-item symptom assessment.
• Machine learning identified the twenty most discriminating items, with analysis focusing on the top ten.
• Items were scored using both equivalent and machine learning-generated weighted values.

Main Results:

• Both weighted and equivalent scoring achieved high classificatory accuracy (around 96%) in distinguishing bipolar from unipolar patients.
• Optimal cut-off scores were identified for the new measure.
• High accuracy was maintained when comparing bipolar I and bipolar II groups against the unipolar group.

Conclusions:

• The ten-item set provides a novel, highly accurate measure for evaluating bipolar disorder.
• This measure can assist clinicians in differentiating between bipolar and unipolar mood disorders.
• The findings support the utility of weighted symptom assessment in psychiatric diagnosis.