Machine learning-driven simplification of the hypomania checklist-32 for adolescent: a feature selection approach

Area of Science:

• Psychiatry
• Adolescent Mental Health
• Machine Learning in Healthcare

Background:

• The Hypomania Checklist-32 (HCL-32) is a common tool for screening bipolar disorder.
• Its length poses challenges for adolescent assessments.
• Machine learning was used to create a shorter, adolescent-focused version.

Purpose of the Study:

• To develop and validate a shortened HCL-32 for adolescents.
• To improve the efficiency and accuracy of bipolar disorder screening in this population.

Main Methods:

• Analysis of HCL-32 data from 2,850 adolescents.
• Utilized random forest and gradient boosting machine algorithms for feature selection.
• Evaluated model performance using area under the curve (AUC) and receiver operating characteristic (ROC) analysis.

Main Results:

• An 8-item HCL-32 version was developed with high predictive accuracy (AUC=0.97).
• The scale effectively identifies core manic symptoms like increased energy, risk-taking, and irritability.
• Optimal cutoff scores were identified for specificity and balanced sensitivity/specificity.

Conclusions:

• The 8-item HCL-32 shows significant diagnostic utility for adolescent bipolar disorder.
• This shortened scale offers a more efficient and accurate screening tool.
• It addresses challenges in diagnosing bipolar disorder in adolescents, improving clinical assessment.