Suicide was the 10th most common cause of death in 2013 in the U.S., according to the Centers for Disease Control and Prevention (CDC). That year there were 41,149 suicides: one every 13 minutes. We’ve written about Facebook’s use of artificial intelligence to prevent suicide, and now investigators at Vanderbilt University have gone further with AI and suicide.

Colin Walsh, assistant professor of Biomedical Informatics, Medicine and Psychiatry and Behavioral Science at  Vanderbilt published a study in Clinical Psychological Science that demonstrated how artificial intelligence machine learning algorithms can be trained to predict suicide risk. Researchers automated risk analysis by analyzing electronic health data in two rounds of testing. The first round analyzed 5,167 patient records of adults coded for self-injury who were seen between 1998 and 2015. Clinical experts reviewed the data and divided it into 3,250 people with suicide attempt history and 1,917 who had non-suicidal self-injury. The machine learning algorithm shuffled the full set of records repeatedly, creating decision trees that compared the AI prediction to the experts’. That was how the algorithm became expert itself. In the end, the algorithm was 80% accurate predicting a suicide attempt based on data from two years prior to a suicide attempt and 84% with data from one week prior to an attempt.  The second round of testing with 13,000 general population EHR records successfully predicted suicide attempts 84% of the time two years prior and 92% of the time one week prior to actual attempts.

The results of the predictive analysis are promising. The hope is to develop a way to implement the machine-learning models of attempted suicide into actual use with new patients. The group will also need to determine how the data will be used. For example, if the system predicts a suicide attempt sometime in the future, medical teams will need guidelines for how to use that information with individual patients.