Invariance-based causal prediction to identify the direct causes of suicidal behavior

Abstract

Despite decades of research, the direct causes of suicide remain unknown. Some researchers have proposed that suicide is sufficiently complex that no single variable or set of variables can be determined causal. The invariance-based causal prediction (ICP) is a contemporary data analytic method developed to identify the direct causal relationships, but the method has not yet been applied to suicide. In this study, we used ICP to identify the variables that were most directly related to the emergence of suicidal behavior in a prospective sample of 2,744 primary care patients. Fifty-eight (2.1%) participants reported suicidal behavior during the following year. Of 18 predictors tested, shame was most likely to be directly causal only under the least restrictive conditions. No single variable or set of variables was identified. Results support the indeterminacy hypothesis that suicide is caused by many combinations of factors, none of which are necessary for suicide to occur.

Lower Bounds on the Error Probability for Invariant Causal Prediction

“…The recent invariant causal prediction (ICP) framework [1] has pioneered the study of leveraging invariance in datasets across different experimental settings (or environments) for identifying potential causal predictors (with various theoretical extensions and applications [2]- [5]). For linear models, the underlying assumption [1, Assumption 1] requires the existence of invariant (w.r.t.…”