Age matters: a narrative review and machine learning analysis on shared and separate multidimensional risk domains for early and late onset suicidal behavior.

There is considerable heterogeneity among late-life suicide attempters who can present stark differences in their suicidal trajectories. This work provides a narrative review of sources of heterogeneity of suicide risk in late-life depression and describes a quantitative study of the relative importance of multidimensional risk domains, in discriminating suicide attempters, split into early- and late-onset cases, from depressed non-attempters. The sample comprised 382 depressed middle-aged and older adults (aged 50 years or older, mean age = 63.5 years). Penalized binomial logistic regression and Random Forest models were fit using cross-validation in 100 versions of a training dataset of 83 variables, grouped into seven domains, to distinguish early- and late-onset suicide attempters from depressed non-attempters, and evaluated on testing datasets. Variable and domain importances were defined based on the frequency of each variable in the final models. Variables from the behavioral control and planning domain had high importance in differentiating both early-onset and late-onset attempters from depressed non-attempers. Early-life history as well as mood/anxiety/emotion regulation were important in distinguishing the early-onset group from depressed non-attempters, whereas social dynamics/interactions and cognition/decision-making were important in distinguishing the late-onset group from depressed non-attempters. This study underscores the importance of examining multiple risk factors for suicide together, and the advantages of considering known sources of heterogeneity in populations at risk in the development of more comprehensive and personalized suicide risk assessment tools and guidelines.