Using Machine Learning and the HAMD-24 Scale to Predict Suicide Ideation in Depressed Patients

Area of Science:

• Psychiatry and Mental Health
• Computational Psychiatry
• Machine Learning in Healthcare

Background:

• Suicidal ideation is a significant concern in patients with depression.
• Early identification of at-risk individuals is crucial for effective intervention.
• The Hamilton Depression Scale (HAMD-24) is a widely used measure of depression severity.

Purpose of the Study:

• To identify factors associated with suicidal ideation in depression patients.
• To develop and evaluate machine learning models for predicting early suicide ideation risk.
• To determine the optimal machine learning algorithm for this prediction task.

Main Methods:

• Utilized data from 374 depression patients assessed with HAMD-24 and Beck Suicide Ideation (BSI) Questionnaire.
• Compared four machine learning models: Support Vector Machine (SVM), Naive Bayes Classification (NBC), Random Forest (RF), and Extreme Random Trees Classification (ERTC).
• Evaluated models using accuracy, precision, recall, F1 scores, Kappa, Matthew's correlation coefficients, and Area Under the Curve (AUC).

Main Results:

• The ERTC model demonstrated superior performance with 77.75% accuracy and 0.80 AUC.
• Key predictors of suicidal ideation included despair, guilt, inferiority complex, loss of work/interests, and depressive emotions.
• Patients with suicidal ideation were younger and less likely to be on antidepressants.

Conclusions:

• The ERTC model is effective for predicting suicidal ideation risk in depression patients.
• Early detection through this model can facilitate timely interventions and potentially reduce suicide rates.
• Findings provide a theoretical basis for developing new depression and suicide assessment scales.