For three Axis I disorders, that is, mood, alcohol use and substance use, the AORs indicated either insignificant effects or a decreased risk of future SRO. These results differ from the increased risks found in the unadjusted analyses, indicating confounding effects of correlated predictors of SRO, for example, prior SRO and comorbid psychiatric disorders.3–5 In further analysis that excluded prior SRO in the multivariate model, a positive predictive effect of mood disorder on future SRO (AOR=2.05, 95% CI 1.92 to 2.17) was observed, contrary to the reduced effect of mood disorder in the full model that adjusted for prior SRO (results available upon request). This finding exemplifies an instance when collinearity with a stronger predictor (eg, wave 1 SRO) overwhelmed the explanatory power of other predictors with weaker relationships. It is thus remarkable that significant effects of smoking on risk of SRO remained despite the evidence of effect suppression due to confounding. Ranked in decreasing order, the significant predictors of SRO risk in the present sample were: prior SRO, borderline personality disorder, ADHD, schizotypal disorder, current smoking, former smoking, avoidant personality disorder and selected demographic characteristics.
