FROM PREDICTION TO INTERPRETATION: MACHINE LEARNING-BASED INSIGHTS INTO SUICIDAL IDEATION AMONG ADHD PATIENTS
Author(s)
SAMUEL AYEMERE, MSc, PharmD1, Maryam Pathan, BS2, Mohammad A. Al-Mamun, PhD3;
1West Virgina University, MORGANTOWN, WV, USA, 2West Virginia University, Morgantown, WV, USA, 3University of West Virginia, Morgantown, WV, USA
1West Virgina University, MORGANTOWN, WV, USA, 2West Virginia University, Morgantown, WV, USA, 3University of West Virginia, Morgantown, WV, USA
OBJECTIVES: This study investigates the elevated risk of suicidal ideation (SI) among individuals with attention deficit hyperactivity disorder (ADHD) and employs machine learning (ML) techniques to identify and evaluate key contributing factors.
METHODS: This study utilized electronic health records from TriNetX to conduct a retrospective longitudinal study of all patients with ADHD who had available medication information and were receiving care within West Virginia healthcare organizations between 2007 and 2023. Given the low prevalence of SI in the data, synthetic minority oversampling technique (SMOTE) was applied to address the class imbalance issue. Logistic regression, Random Forest and Extreme Gradient boosting (XGBoost) models were developed and compared on the SMOTE-adjusted dataset. SHAP (Shapley Additive exPlanations) analysis was used to interpret the most influential factors contributing to the model predictions.
RESULTS: A total of 11,906 patients with ADHD were included in the study. Suicidal ideation was observed in 4.26% of the cohort. The XGBoost model trained on SMOTE-adjusted data with a sampling ratio of 0.5 demonstrated the best overall performance with an accuracy of 0.85, an F1 score of 0.82, a precision of 0.7, an AUCROC of 0.96, and an AUPRC of 0.93. SHAP analysis identified depression or anxiety as the major factor that leads to SI, followed by bipolar disorder, age <18 years, male gender, and stimulant use.
CONCLUSIONS: Age <18 years, male gender, stimulant use, psychiatric comorbidities like depression or anxiety and bipolar disorders were identified as key factors associated with SI in ADHD. Integrating ML-based SI risk prediction in clinical decision-making among individuals with ADHD may support earlier risk stratification, targeted monitoring, and timely intervention.
METHODS: This study utilized electronic health records from TriNetX to conduct a retrospective longitudinal study of all patients with ADHD who had available medication information and were receiving care within West Virginia healthcare organizations between 2007 and 2023. Given the low prevalence of SI in the data, synthetic minority oversampling technique (SMOTE) was applied to address the class imbalance issue. Logistic regression, Random Forest and Extreme Gradient boosting (XGBoost) models were developed and compared on the SMOTE-adjusted dataset. SHAP (Shapley Additive exPlanations) analysis was used to interpret the most influential factors contributing to the model predictions.
RESULTS: A total of 11,906 patients with ADHD were included in the study. Suicidal ideation was observed in 4.26% of the cohort. The XGBoost model trained on SMOTE-adjusted data with a sampling ratio of 0.5 demonstrated the best overall performance with an accuracy of 0.85, an F1 score of 0.82, a precision of 0.7, an AUCROC of 0.96, and an AUPRC of 0.93. SHAP analysis identified depression or anxiety as the major factor that leads to SI, followed by bipolar disorder, age <18 years, male gender, and stimulant use.
CONCLUSIONS: Age <18 years, male gender, stimulant use, psychiatric comorbidities like depression or anxiety and bipolar disorders were identified as key factors associated with SI in ADHD. Integrating ML-based SI risk prediction in clinical decision-making among individuals with ADHD may support earlier risk stratification, targeted monitoring, and timely intervention.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR209
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
No Additional Disease & Conditions/Specialized Treatment Areas