Background. Although suicide is a leading cause of death, few studies have attempted to predict suicidal ideation prospectively using epidemiological samples or multivariate methods.

Method. Discrete-time event history analysis was used to model the onset of thoughts of death or self-harm (TD/SH) in a population-based sample of female twins (N=2164) using variables from the demographic, psychopathological, childhood adversity, personality and life event domains. Univariate, multivariate-within domain and multivariate-across domain regression analyses were performed.

Results. Most variables predicted TD/SH in the univariate analyses. However, the only variables to predict TD/SH independently were obsessive symptoms, childhood sexual abuse, rural residence, unemployment, older age, lifetime history of cocaine misuse and low levels of education, personal religious devotion and altruism, as well as divorce/separation, loss of confidant, assault, job loss and financial problems in the previous month. This final model explained 16% of the variance in TD/SH. Lifetime histories of major depression, panic disorder and alcohol misuse had no significant independent effect.

Conclusions. Many variables, from all five domains of risk factors, are associated with the risk of TD/SH, but many of these effects may be mediated by other risk factors. Proximal life events and psychopathology may have more independent effects than other domains. The overall ability of these risk factors to predict TD/SH is modest. We cannot rule out that differences between these analyses and previous reports were due to our use of TD/SH as the dependent variable instead of thoughts of committing suicide per se.