Determining suicide risk in an adolescent population: Questionnaire-based approaches.

Area of Science:

• Psychiatry and Mental Health
• Computational Science
• Health Informatics

Background:

• Accurate suicide risk assessment is crucial for timely intervention.
• Questionnaires are common tools, but their length and optimal use require investigation.
• Comparing algorithmic and data-driven approaches can enhance predictive accuracy.

Purpose of the Study:

• To compare knowledge-based algorithms with machine learning models for suicide risk classification.
• To identify and reduce the number of essential questionnaire items for suicide risk assessment.
• To evaluate the performance of different approaches in predicting suicide risk levels.

Main Methods:

• A classification task was performed on a set of suicide risk questionnaires.
• Two methods were compared: expert knowledge-based algorithms and data-inferred machine learning models.
• Feature ablation was utilized to reduce questionnaire items while assessing impact on prediction.

Main Results:

• Machine learning models achieved an F1 macro average score of up to 85%, outperforming knowledge-based models.
• Significant reduction in questionnaire items was possible without compromising predictive performance.
• The Suicide Cognitions Scale-Revised was identified as the most impactful questionnaire for prediction.

Conclusions:

• Data-inferred machine learning models effectively predict suicide risk levels using a reduced set of questionnaires.
• Optimizing questionnaire length and leveraging machine learning can enhance clinical utility in suicide risk assessment.
• The findings support the use of machine learning for more efficient and accurate suicide risk screening.