The use of artificial intelligence to spot suicides has come from Florida State University scientists, led by Jessica Ribeiro, and the groundbreaking results could assist healthcare providers with safeguarding people who are at risk from taking their own lives. Suicide rates vary across different countries and by demographic groups. In the U.S., for example, the suicide rate is 120 people per day, totaling some 45,000 per year. The rate has been rising since the year 1999.
The predictions about suicide have come via a special algorithm. This was developed by surveying a mass of data relating to patients: the electronic health records of about 2 million patients in the U.S. state of Tennessee. From these, 3,200 people had attempted suicide. The system began to work out what was different about the 3,200 people from the 2 million population. The computer program was able to assess how different variables interact with one another as a whole. From this, the combination of factors likely to lead to a person contemplating suicide could be tracked. The program then “learnt” which combination of factors were the most likely and was able to make predictions about new patients in terms of suicide risk.
As well has having a high predictive power over a two-year timeframe the accuracy of the predictions increases the closer a person comes to making a suicide attempt (up to 92 percent at one week before an attempt is made). The objective is to allow medics to assign a risk score to patients and to signal a ‘red alert’ to those patients considered to be at the biggest risk. This would allow a psychologist or a psychiatrist to respond rapidly. This is doubly important since the review of records suggest that most people who commit suicide have visited a medical professional in the months leading up to their suicide attempt. It also allows medical facilities to prioritize the most at risk cases.
The research is to be published in the journal Clinical Psychological Science, under the title “Predicting Risk of Suicide Attempts over Time through Machine Learning.”
