Identifying Suicidal Ideation in Older Patients With Chronic Pain Using Explainable Machine Learning: A Cross-Sectional Study

Abstract

Background:

Early identification of suicidal ideation (SI) in older adults with chronic pain is critical for improving mental health outcomes. However, methods for SI risk identification and stratification remain limited. Furthermore, traditional statistical methods often fail to capture the complex relationships among risk factors, making machine learning approaches particularly valuable.

Purpose:

To develop explainable machine learning models to identify SI risk in older adults with chronic pain.

Methods:

This cross-sectional study recruited 516 inpatients from a tertiary hospital in China, all of whom were older adults (aged ≥60 years) with chronic pain (duration ≥3 months) and a visual analog scale (VAS) score of ≥3. Participants completed a general data questionnaire, Beck Suicidal Ideation Scale, VAS, Pain Catastrophizing Scale, and Perceived Social Support Scale. Data were preprocessed using Min–Max Normalization and SMOTETomek. Variables were selected based on univariate analysis, Support Vector Machine-Recursive Feature Elimination, and LASSO regression. Machine learning models were developed and interpreted using the Shapley Additive Explanations method.

Results:

History of suicide, history of depression, suicidal behavior exposure, number of pain sites, pain duration, pain level, pain catastrophizing, and perceived social support were associated with SI. The Random Forest (RF) model, which exhibited the best performance, achieved an accuracy of 0.85 and an area under the curve of 0.89. Pain level, perceived social support, and pain catastrophizing were the 3 most influential contributors to SI risk.

Conclusions:

The explainable machine learning models facilitate early detection and risk stratification of SI for nurses. By identifying key SI risk factors, these findings support targeted interventions and routine clinical screening.