Predictors of suicide attempt versus isolated suicidal ideation in major depressive disorder: Novel insights from LASSO-boruta screening and binary logistic regression.

OBJECTIVES: Phenotypic heterogeneity between suicidal ideation (SI) only and suicide attempt (SA) in major depressive disorder (MDD) suggests distinct suicide risk profiles, yet key clinical and symptomatic factors distinguishing SA from SI remain incompletely defined. METHODS: A total of 799 MDD patients were allocated into SI only (n = 407) and SA (n = 392) groups. Sociodemographic indicators, clinical characteristics and depressive symptoms were compared. Predictors were selected using combined Boruta and LASSO regression, followed by binary logistic regression to identify independent factors associated with SA. A predictive model was constructed and assessed for discrimination, calibration, and clinical utility using decision curve analysis (DCA). RESULTS: Compared with the SI-only group, the SA group had more recurrent depression, treatment-resistant depression (TRD), positive psychiatric family history, hospitalizations, and depressive episodes. For symptoms, the SA group showed higher prevalence of self-harm, rejection sensitivity, weight gain, and mood-congruent psychosis, but lower prevalence of sluggishness and other pain conditions. Independent positive predictors of SA included number of hospitalizations (OR = 1.45, 95%CI: 1.17-1.79), recurrent depression (OR = 1.97, 1.24-3.10), tricyclic antidepressant use (OR = 5.17, 1.16-22.95), self-harm (OR = 31.11, 19.61-49.31), and rejection sensitivity (OR = 1.97, 1.27-3.05). In contrast, unhappiness, agitation, demoralization, and derealization were inversely associated with SA (all P < 0.05). The predictive model showed excellent discrimination (AUC = 0.907,95%CI: 0.887-0.928), satisfactory calibration, and favorable clinical net benefit in DCA. CONCLUSION: Distinct phenotypic differences exist between MDD patients with and without suicide attempts. Self-harm and rejection sensitivity are key risk factors, and the developed predictive model provides reliable individualized risk assessment for clinical practice.