Clinical AND Statistical Prediction: Integrating Clinician Judgment and Statistical Models to Improve the Prediction of Suicidal Behavior

The suicide rate has increased over the past 20 years, as has the pace of research studies focused on suicide prediction, yet the ability to accurately predict suicidal behavior has not improved. There has been significant debate in the field over the past several decades regarding the use of clinical versus statistical prediction. Clinical prediction, rooted in clinicians' judgment and intuition, allows for contextual sensitivity and the detection of anomalies, whereas statistical prediction offers standardized, bias-resistant models derived from observed data. This article suggests that suicide prediction will advance most quickly and effectively through a synthesis of clinical AND statistical methods. Key avenues for future work toward this end are described, including those focused on quantifying effective clinical practices (eg, studying clinicians who excel at prediction, natural language processing of clinician notes) and testing the utility of combining statistical predictions with clinician decision-making practices. Creating and testing hybrid models may improve accuracy, increase clinician use, and ultimately reduce suicide deaths.