(PS5-1) Understanding Clinician Decision-Making about Suicide Risk

Abstract Text:

Introduction: Over 700,000 people die by suicide each year, making it a leading cause of death among all ages. Although our ability to predict suicide is at near chance levels, clinicians in high acuity settings must decide which patients are at risk for suicide every day. The ability to detect those at risk is key because approximately 50% of people who die by suicide have a clinical encounter in the month before their death. However, the literature lacks consensus about whether clinicians are accurate predictors of suicide and what factors inform their risk assessments. This study aimed to characterize clinician prediction accuracy and model what may be guiding their predictions about suicide risk.

Method: Participants were 2,000 patients presenting to the emergency department (ED) and referred for evaluation by the Acute Psychiatry Service (APS) due to concerns about psychiatric distress. The treating clinician for each patient completed a brief survey that included rating the likelihood (0%, 10%, 20%, 30% ...100%) that this patient would make a suicide attempt within the next 1- and 6-months if untreated. Clinicians also evaluated their confidence in this rating on a scale of 1-5. The occurrence of a suicide attempt within 1- and 6-months after the baseline assessment was established via follow-up surveys and review of patient medical records.

Results: Clinician 1-month and 6-month predictions were moderately positively correlated with whether patients had reported suicidal ideation in the past week (r = 0.31, r = 0.31), as well as whether patients had made a suicide attempt in the past week (r = 0.37, r = 0.35). There was a weak positive correlation between clinician confidence and 1-month and 6-month predictions (r = 0.06, r = 0.06). Knowledge of suicidal ideation (SI) and suicide attempts (SA) in the past week significantly increased clinician 1-month predictions (β = 1.2, p < 0.001; β = 1.6, p < 0.001). Higher confidence was significantly associated with lower 1-month prediction scores (β = -0.22, p < 0.001). Six-month predictions demonstrated the same trends. The proportion of SA made was largely uniform across clinicians’ likelihood rating scores for 1-month and 6-months. In contrast, proportion of suicide attempts increased as patient likelihood rating scores increased for both 1-month and 6-months. Clinician 1-month and 6-month prediction did not predict whether participants made a suicide attempt (β = - 0.27, p = 0.2; β = - 0.06, p = 0.10). Patient 1-month and 6-month predictions did significantly predict suicide attempt (β = 0.1, p = 0.009; β = 0.1, p = 0.009).

Conclusion: Patients are better at predicting their future suicide attempts than clinicians. Clinician predictions seem to be informed by the presence of patients’ suicidal thinking, but their resulting predictions about patient behavior are not accurate. Future work must help to better understand what clinicians are basing their predictions on, and how to help make them more accurate.