Predicting Suicidal Ideation in Community-based Older Adults Using Self-report Questionnaires with Machine Learning

Suicide is a significant public health issue, particularly among older adults, where the risk is heightened. Early identification of individuals at risk for suicidal ideation is essential for timely interventions, greatly improving prevention efforts. This study aimed to develop a predictive model for suicidal ideation in community-dwelling older adults using psychiatric self-report scales and machine learning classifiers.

A total of 238 older adults were assessed using the Patient Health Questionnaire-9 (PHQ-9), Generalized Anxiety Disorder-7, Perceived Stress Scale-10, and World Health Organization Quality of Life Scale - Abbreviated Version. A nested 5-fold cross-validation procedure repeated 100 times was used for feature selection and model evaluation. Various classifiers-including support vector machines, random forest, logistic regression, linear discriminant analysis, and gradient boosting-were employed.

As the number of PHQ-9 items increased from two to six, the area under the curve (AUC) rose from 0.835 to 0.892. When a set of nine features-selected based on feature stability across iterations-was used, the AUC further improved to 0.904. This progression indicates that inclusion of additional informative items enhances classification performance.

This study demonstrates that psychiatric self-report scales can effectively predict suicidal ideation risk in community-dwelling older adults. By utilizing efficient features, the predictive accuracy of the model can be enhanced, offering valuable insights for developing early identification systems for high-risk groups. These findings suggest that a community-based suicide prevention program could be promoted by implementing a screening system to identify individuals at high risk for suicidal ideation among the elderly.