Figure 2.

Schematic representation of mediation (based on multivariate regressions) models evaluating the hypotheses depicted in Figure 1: Models 1, 2 and 3. All models included the same variables as follows: combat exposure score (CES), posttraumatic stress disorder (PTSD) check list (PCL) cluster B score (re-experiencing), PCL cluster D+E score (summed scores of negativity and hyperarousal), Hamilton Depression Rating Scale (HAM-D) score without item 3 (depression without suicidality), suicidality score (summed scores on HAM-D item 3 and Montgomery-Asberg Depression Rating Scale item 10), religiosity level (high or non-high), nicotine dependence (moderate/high or none/low), age and presence of cardiovascular comorbidity (CV) or diabetes mellitus (DM). The roles of “independent”, “mediator” or “dependent” variable or covariates, changed across the models. All possible associations – direct or mediated – are adjusted for all variables in the models hence all are “adjusted” or independent. Associations are expressed as regression coefficients (β) with 95% confidence intervals. Covariate effects on mediator or dependent variables are omitted for clarity. Fonts and lines/arrows (full for direct associations, dashed for mediated associations) in black depict coefficients (associations) different from zero. Gray font/lines depict coefficients not different from zero.