Testosterone reactivity moderates the association between reactive aggression and suicidality in major depression
H. E. Baki, O. Aktan, E. Bodur, K. BaşarIntroduction
Reactive aggression (RA) is associated with suicidality; however, this relation has not been tested with behavioral measures of RA. Earlier, RA has been examined in relation to testosterone (T) and cortisol (C). In individuals with suicidal behavior, dysregulated cortisol responses have been reported, whereas the role of testosterone reactivity remains unclear.
Objectives
To investigate the associations of suicidal ideation (SI) and suicide attempt (SA) with laboratory-measured RA, baseline T and C levels and their reactivity, and clinical variables, and to test whether testosterone and cortisol reactivity moderate the association between RA and suicidality (SI/SA) in patients with major depression.
Methods
Patients with major depression (n=95; HAM-D≥14) were classified based on the Columbia-Suicide Severity Rating Scale into three groups: major depression (MD; n=42), suicidal ideation (SI; ideation in the past month; n=31), and suicide attempt (SA; attempt in the past month; n=22). Participants completed a 10-min Point Subtraction Aggression Paradigm (PSAP) for behavioral assessment of RA. Saliva samples were collected pre-task and 15 min post-task; T and C were measured by ELISA. Reactivity (ΔT%, ΔC%) was the percent change from baseline. STAXI-State Anger was administered pre- and post-task. Additional self-report measures included BPAQ, BIS-11, Fava AAQ, and INQ (perceived burdensomeness and thwarted belongingness). Analyses used a general linear model (GLM) and multinomial logistic regression; moderation was evaluated via interaction terms.
Results
Reactive aggression differed by group (
p
=.004,
η
²=.12), higher in SA than in SI and MD. State anger increased during PSAP (
p
<.001). Baseline hormone levels and hormonal reactivity (ΔT%, ΔC%) did not differ across groups (all p>.10). In the GLM predicting RA (sex, group, baseline T/C, ΔT%, ΔC%, group × ΔT%), the group × ΔT% interaction was significant (
F
=3.63,
p
=.031,
η
²=.08), driven by the SA group. Within SA, ΔT% positively predicted RA (
B
=0.034,
p
=.008) and explained 14.4% of the variance (Image 1). In the multinomial model, higher reactive aggression, higher depression severity, greater perceived burdensomeness, and lower baseline T were significant predictors of SA. SI was only associated with perceived burdensomeness. Adding RA × ΔT% to the model improved model fit (Nagelkerke
R²
=.583), and the interaction significantly predicted SA vs MD.
Image 1: Long description.