Developing Predictive Models to Enhance Clinician Prediction of Suicide Attempts Among Veterans With and Without PTSD

The limitations of self-report confine clinicians' ability to accurately predict suicides and suicide attempts (SAs). Behavioral assessments (e.g., Death Implicit Association Test [IAT]) may be a means of supplementing self-report and clinician prediction.

The authors aimed to build and test a predictive model of SAs that included established risk factors and measures of suicide risk, and Death IAT scores. The authors also sought to test the predictive validity of the SA model among subgroups of Veterans with and without PTSD.

Participants included 166 psychiatrically hospitalized Veterans.

A model that included patient prediction, age, and Death IAT scores improved upon clinician prediction of SAs during the six-month follow-up (C-statistic for clinician prediction = 73.6, 95% CI [62.9, 84.4] and C-statistic for model = 82.8, 95% CI [74.6, 91.0]). The model was tested in subgroups of Veterans with and without PTSD. Among Veterans without PTSD, the Death IAT and patient prediction predicted SAs above and beyond clinician prediction, while these variables did not significantly improve prediction among Veterans with PTSD (C-statistic for no-PTSD = 91.3, 95% CI [80.6, 1.00]; C-statistic for PTSD = 86.8, 95% CI [76.8, 96.8]). Building a separate model for Veterans with PTSD did not improve upon clinician prediction.

Findings indicate that predictive models may bolster clinician prediction of SAs and that predictors may differ for Veterans with PTSD.