Machine learning-based prediction of suicide risk using adult attention-deficit/hyperactivity disorder symptoms and depression indicators: insights from a nationally representative south korean survey

Abstract

Background: Suicide is a major global health issue. Evidence shows adult ADHD symptoms increase suicide risk, particularly with depression. We aimed to develop a predictive ML model. Methods: We analyzed data from the 2021 Korean National Mental Health Survey. We defined suicide risk as serious suicidal ideation. Employing complex survey-weighted logistic regression and random forest (RF) classifier with age and sex covariates, we predicted suicidal ideation. We applied the synthetic minority oversampling technique (SMOTE) to address class imbalance in model training. Results: The SMOTE-balanced RF model exhibited higher recall (approximate to 0.76) for suicide risk than logistic regression (recall 0.08-0.48), with balanced overall performance and minimal bias between classes. Inattention symptoms demonstrated the strongest associations with suicidal ideation (odds ratio approximate to 3.2, p < 0.001). Conclusion: In a nationwide sample, adult ADHD symptoms-particularly inattention-and depression indicators were significantly correlated with suicidal ideation. Compared to traditional methods, an ML model substantially enhanced the identification of individuals exhibiting suicidal ideation, while maintaining fairness. These findings indicate that incorporating adult ADHD screening and ML-based models into suicide prevention strategies can facilitate the early detection of high-risk individuals in the general population.