Background: Adolescent suicide attempts have well-established risk factors. The increased number of cases at Laniado Hospital's pediatric department over the past decade prompted our evaluation of the relative importance of these factors.
Objectives: To characterize adolescents presenting after suicide attempts and to use these features to develop a neural network for early identification of at-risk youth.
Methods: We conducted a retrospective study of suicide attempts among adolescents (11–18 years) admitted to Laniado Hospital's pediatric department (2015–2021). The study included 82 patients with a matched control group (n=82). We analyzed epidemiological, medical, and psychosocial characteristics, identifying statistically significant factors associated with suicide attempts. We then built a predictive model using neural networks.
Results: Significant risk factors for suicide attempts included living outside original home (odds ratio [OR] 6.71, P = 0.0002), female gender (OR 12.67, P = 0.0502), unmarried parents (OR=98.51, P < 0.0001), advanced age (95% confidence interval [95%CI] 0.477–1.583, P = 0.0001), higher hemoglobin (95%CI 0.689–1.81, P = 6.30 × 10-6), higher mean corpuscular volume (MCV) (95%CI 3.61–8.07, P = 0.014), and prior psychiatric diagnosis (OR 71.82, P < 0.0001). Non-Ashkenazi background was more common but not significant (P = 0.074). Our neural network model achieved 99.85% predictive accuracy.
Conclusions: Psychiatric history, unmarried parents, female gender, and living outside the home were the strongest risk factors for adolescent suicide attempts. We observed higher hemoglobin levels and MCV among affected individuals. Our neural network showed high accuracy (99.85%) in distinguishing between adolescents with suicide attempts from matched controls.