Category: Suicide and Self-Injury

Symposium

Which Affect States Are Most Strongly Associated With Suicidal Ideation? A Real-Time Monitoring, Exploratory Data Mining Study

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

Keywords: Suicide | Ecological Momentary Assessment | Statistics
Presentation Type: Symposium

Intensive real-time monitoring has become increasingly popular, especially for studying suicidal ideation. One challenge of this methodology is that repeatedly sampling more than a dozen items multiple times per day can become burdensome for participants who may drop out of the study, especially during stressful times (which are the times of most interest in this research). Accordingly, the first goal of this study was to use exploratory data mining to identify which variables were the most important to include in repeated assessments of suicidal ideation and its associated risk factors. Beyond knowing which variables are important, it may also be useful to know what scores on these variables can identify those at highest risk for suicidal ideation. Accordingly, our second goal was to use exploratory data mining to identify optimal cutoffs for predicting suicidal ideation.


Inpatients (N=53) hospitalized for suicide risk completed real-time assessments of suicidal ideation and 19 different affect labels (e.g., ashamed, hopeless, abandoned) four times per day for the duration of their inpatient stay (652 total responses). We used random forests and regression trees to address our first goal. Both methods converged upon 8 of the 19 variables (self-hatred, hopeless, afraid, desperate, abandoned, hostile, inspired, and determined) as most important in predicting suicidal ideation. We used regression trees to address our second goal. We found that the highest 21% of suicidal ideation scores where characterized by “self-hatred” ratings above 7.5 (out of 10). The highest 2% of suicidal ideation scores were characterized by “self-hatred” ratings above 7.5, “desperation” ratings above 8.5, and (somewhat counterintuitively) “determined” ratings above 7.


The results of this study serve to help reduce the number of items asked during real-time monitoring studies of suicidal ideation. They also provide combinations of variables that identify the highest risk responses, highlighting the critical role of self-hatred in defining high-risk responses and the potentially ambiguous nature of ratings of “determined “in predicting the highest risk responses.

Evan Kleiman

College Fellow
Harvard University

Presentation(s):

Send Email for Evan Kleiman


Assets

Which Affect States Are Most Strongly Associated With Suicidal Ideation? A Real-Time Monitoring, Exploratory Data Mining Study



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 Which Affect States Are Most Strongly Associated With Suicidal Ideation? A Real-Time Monitoring, Exploratory Data Mining Study