The Role of Big Data Analytics in Predicting Suicide

This chapter reviews the long history of using electronic medical records and other types of big data to predict suicide. Although a number of the most recent of these studies used machine learning (ML) methods, these studies were all suboptimal both in the features used as predictors and in the analytic approaches used to develop the prediction models. We review these limitations and describe opportunities for making improvements in future applications. We also review the controversy among clinical experts about using structured suicide risk assessment tools (be they based on ML or older prediction methods) versus in-depth clinical evaluations of needs for treatment planning. Rather than seeing them as competitors, we propose integrating these different approaches to capitalize on their complementary strengths. We also emphasize the distinction between two types of ML analyses: those aimed at predicting which patients are at highest suicide risk, and those aimed at predicting the treatment options that will be best for individual patients. We explain why both are needed to optimize the value of big data ML methods in addressing the suicide problem.