講演情報
[GA-2-01]Leveraging Financial Metrics as Predictive Biomarkers for Suicide Risk: An Integrative Machine Learning Study on Economic and Mental Health Interactions
*Sahnaz Vivinda Putri1, Prihantini Prihantini2, Andi Nursanti Andi Ureng3, Asfirani Zahaz4, Rifaldy Fajar5 (1. International University Semen Indonesia (Indonesia), 2. Bandung Institute of Technology (Indonesia), 3. Andini Persada College of Health Sciences (Indonesia), 4. Bonto-Bonto General Hospital (Indonesia), 5. Yogyakarta State University (Indonesia))
キーワード：
Suicide Risk Prediction、Financial Distress、Machine Learning、Economic and Mental Health Integration、Risk Stratification
Background and Aim: The interplay between financial distress and mental health is well-established, yet its predictive role in suicide risk remains underexplored. This study investigates financial metrics, such as unemployment trends and debt-to-income ratios, as biomarkers for suicide risk. By integrating economic and mental health data, the aim is to develop a machine learning model for early identification of high-risk individuals, enabling proactive interventions tailored to economically vulnerable populations.
Methods: Datasets utilized include the American Community Survey (ACS, 2018–2023) for employment and income metrics, the Behavioral Risk Factor Surveillance System (BRFSS, 2018–2023) for mental health indicators, and the Federal Reserve Economic Data (FRED, 2018–2023) for macroeconomic trends such as bankruptcy rates. A cohort of 150,000 individuals was constructed by harmonizing these datasets using geographic and temporal alignment across U.S. counties. Predictors included unemployment duration (>6 months), income-to-debt ratio, and self-reported mental distress. A machine learning pipeline was implemented, integrating Gradient Boosting Machines (GBMs), Random Forests (RFs), and Neural Networks (NNs), with hyperparameter tuning via grid search. Model performance was evaluated using 10-fold cross-validation, with precision-recall area under the curve (PR-AUC) as the primary metric for assessing predictive accuracy.
Results: Unemployment lasting >6 months increased suicide risk by 46% (OR: 1.46; 95% CI: 1.41–1.52; p < 0.001), while income-to-debt ratios >50% raised risk by 58% (OR: 1.58; 95% CI: 1.52–1.65; p < 0.001). High self-reported mental distress was the strongest predictor, increasing risk by 87% (OR: 1.87; 95% CI: 1.80–1.94; p < 0.001). Younger individuals (18–34 years) and single-income households faced disproportionately higher risks. The ensemble model achieved strong performance (PR-AUC: 0.92; AUROC: 0.90) and showed a 27% improvement over logistic regression models, with Gradient Boosting Machines contributing most to accurate predictions.
Conclusions: This study establishes financial distress as a quantifiable predictor of suicide risk. The findings emphasize the importance of integrated economic and mental health strategies to prevent suicide in high-risk populations.
Methods: Datasets utilized include the American Community Survey (ACS, 2018–2023) for employment and income metrics, the Behavioral Risk Factor Surveillance System (BRFSS, 2018–2023) for mental health indicators, and the Federal Reserve Economic Data (FRED, 2018–2023) for macroeconomic trends such as bankruptcy rates. A cohort of 150,000 individuals was constructed by harmonizing these datasets using geographic and temporal alignment across U.S. counties. Predictors included unemployment duration (>6 months), income-to-debt ratio, and self-reported mental distress. A machine learning pipeline was implemented, integrating Gradient Boosting Machines (GBMs), Random Forests (RFs), and Neural Networks (NNs), with hyperparameter tuning via grid search. Model performance was evaluated using 10-fold cross-validation, with precision-recall area under the curve (PR-AUC) as the primary metric for assessing predictive accuracy.
Results: Unemployment lasting >6 months increased suicide risk by 46% (OR: 1.46; 95% CI: 1.41–1.52; p < 0.001), while income-to-debt ratios >50% raised risk by 58% (OR: 1.58; 95% CI: 1.52–1.65; p < 0.001). High self-reported mental distress was the strongest predictor, increasing risk by 87% (OR: 1.87; 95% CI: 1.80–1.94; p < 0.001). Younger individuals (18–34 years) and single-income households faced disproportionately higher risks. The ensemble model achieved strong performance (PR-AUC: 0.92; AUROC: 0.90) and showed a 27% improvement over logistic regression models, with Gradient Boosting Machines contributing most to accurate predictions.
Conclusions: This study establishes financial distress as a quantifiable predictor of suicide risk. The findings emphasize the importance of integrated economic and mental health strategies to prevent suicide in high-risk populations.