Category: Suicide and Self-Injury


Data Mining to Compare Importance of Risk Factors in Predicting History of Nonsuicidal Self-Injury

Friday, November 17
12:00 PM - 1:30 PM
Location: Cobalt 502, Level 5, Cobalt Level

Keywords: Self-Injury | Statistics
Presentation Type: Symposium

Non-suicidal self-injury (NSSI) is predictive of poor mental health outcomes and as such, research is needed to better understand its etiology in order to inform prevention. Despite the growing body of literature examining risk factors for NSSI, major gaps remain. First, much research has examined correlates of any history of NSSI, without considering recency of the behavior. As a result, there is a gap in our understanding of how NSSI correlates may be differentially associated with former versus recent history of NSSI. Second, this body of literature has relied almost exclusively on analytic tools that permit a very limited number of variables to be considered simultaneously in solving classification problems. Exploratory data mining allows for many factors and their complex relations to be modeled simultaneously, which may be more appropriate for elucidating such a complex classification problem as identifying recent and former self-injurers.          

 The current study employed two exploratory data mining techniques (random forests and elastic net regression) to address these gaps in the literature. University students (N = 553) were administered measures assessing a total of 53 demographic, clinical, and environmental factors. Participants were also administered a measure assessing NSSI and were classified as having a recent (past one year) versus former (engagement prior to past one year) history of NSSI. Random forests (RFs) and elastic net regression were applied to identify variables that are important in the classification of recent and former NSSI.

 Results indicated that 15.8% (n = 87) of the sample endorsed recent NSSI and 15.2% (n = 84) of the sample endorsed former NSSI. In predicting recent NSSI history, anxiety, perceived burdensomeness, thwarted belongingness, and depressive symptoms comprised the most important features. In predicting former NSSI history, anxiety, body protection, body attitude, and bisexuality were identified as the most important features. Each of the important features identified by RFs were also selected in the elastic net regression analyses.          

Our results suggest that anxiety symptomatology is a highly important factor in identifying those with a former and a recent history of NSSI. This finding was surprising, given the modest body of research focused directly on the relationship between anxiety and NSSI. Future research should explore this association further, and investigate the specific symptom clusters of anxiety that may account for this association. While a former history of NSSI seems to be predicted best by body protection, body attitudes, and bisexuality, a recent history of NSSI is better predicted by current depressive symptoms and the interpersonal vulnerabilities of perceived burdensomeness and thwarted belongingness. Our findings suggest that depressive symptoms and interpersonal vulnerabilities may be candidate proximal risk factors for NSSI engagement and should be further investigated as such in short-term prospective studies. Given the divergent findings across our models, our findings highlight the significance of considering time frame of NSSI when investigating risk factors for NSSI. 

Taylor A. Burke

Doctoral Student in Clinical Psychology
Temple University


Send Email for Taylor Burke


Data Mining to Compare Importance of Risk Factors in Predicting History of Nonsuicidal Self-Injury

Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Data Mining to Compare Importance of Risk Factors in Predicting History of Nonsuicidal Self-Injury