Integrating thyroid function and psychometric profiles for lifetime suicide-attempt risk stratification in bipolar...

Area of Science:

• Psychiatry and Mental Health
• Computational Biology and Bioinformatics
• Clinical Diagnostics

Background:

• Bipolar disorder is a severe mental illness with recurrent mood episodes and low diagnosis rates.
• Suicide attempts are a significant concern in bipolar disorder patients, necessitating improved risk stratification.
• Cross-sectional associations were used to explore predictors of lifetime suicide attempts.

Purpose of the Study:

• To develop and validate machine learning models for risk stratification of lifetime suicide attempts in bipolar disorder patients.
• To identify key clinical and biological markers associated with suicide attempts in this population.
• To provide tools for clinicians to identify patients at higher risk for early intervention.

Main Methods:

• Utilized machine learning techniques including Random Forests, Gradient Boosting, and Support Vector Machines.
• Employed LASSO logistic regression for variable selection and SMOTE for handling class imbalance.
• Conducted sensitivity analyses to mitigate reverse causality bias and subgroup analyses on euthyroid patients.

Main Results:

• Random Forests model demonstrated superior performance with an accuracy of 0.938 and AUC of 0.962.
• Top predictors identified were suicidal ideation, education level, hopelessness, retardation symptom severity, and thyroid stimulating hormone (TSH).
• Sensitivity and subgroup analyses confirmed the robustness of the identified predictors and model performance.

Conclusions:

• A robust machine learning model was developed for suicide attempt risk stratification in bipolar disorder.
• The model can assist clinicians in identifying at-risk individuals for timely intervention.
• Future prospective validation is needed to confirm clinical utility and establish temporal precedence.