WASHINGTON – Using machine learning and data from 2,466 Apache tribal members, researchers developed a formula for predicting suicide attempts among youths aged 15-24 years with a “pretty high” level of accuracy, Emily E. Haroz, PhD, said at the annual conference of the American Association of Suicidology.
The area under the receiver operating characteristic curve for the best performing predictive model was 0.806, said Dr. Haroz, a psychiatric epidemiologist at Johns Hopkins University in Baltimore. This compared with an area under the curve of 0.57 for prediction based solely on prior attempts, which is a “very low” predictive score, Dr. Haroz noted. The area under the curve is a performance measure used widely to make predictions.
The new model developed by Dr. Haroz and her colleagues takes prior suicide attempts into account but also factors in other variables, including risk factors such as a diagnosis of posttraumatic stress disorder and recent substance use, and protective factors such as the importance of tribal activities. In addition, the new model requires validation. If validated, it will become part of the suicide surveillance system that has been active at the White Mountain Apache Tribe Reservation in Arizona since the Tribal Council mandated it in 2006. said Dr. Haroz, and contrasts with several past and existing surveillance programs that have been hospital based.Previously, colleagues of Dr. Haroz from the Johns Hopkins Center for American Indian Health reported that implementation of the Celebrating Life surveillance system in 2007 strongly linked with a dramatic reduction in suicide deaths on the Apache reservation during 2007-2012, compared with 2001-2006. Overall, Apache suicide deaths fell from 40/100,000 during the earlier period to 25/100,000 after the Celebrating Life program began, a 38% relative reduction (Am J Public Health. 2016. Dec; 106[12]:183-9). Among youths aged 15-24 years old, the suicide death rate dropped from 129/100,000 to 99/100,000, a 23% relative decrease.
The machine-learning process used data collected through the prospective surveillance program on suicide ideation, nonsuicidal self-injury, suicide attempts, binge substance use, and suicide deaths. The process also used data collected from surveillance participants for 55 different demographic and behavioral variables.“We don’t really know much about how to predict suicide risk,” Dr. Haroz noted, and most of what’s known has been based on retrospective data from hospital-based surveillance. Improved prediction models can help case managers more quickly identify and focus on people who are at highest risk, she added.
Other researchers from the Johns Hopkins Center for American Indian Health described a new intervention program they have developed in collaboration with the Apache tribal leaders that uses tribal elders to teach tribal culture, values, and beliefs to students 11-15 years old. Begun in 2015, “we see this as upstream suicide prevention,” said Victoria O’Keefe, PhD, a clinical psychologist with the Johns Hopkins program. “What is unique about the program is that the elders go into the classrooms,” for 45- to 60-minute sessions done monthly.