Predictive modeling of suicidal ideation in patients with epilepsy.

OBJECTIVE

The prevalence of suicide in the United States has seen an increasing trend and is responsible for 1.6% of all mortality nationwide. Although suicide has the potential to broadly impact the entire population, it has a substantially increased prevalence in persons with epilepsy (PWE), despite many of these individuals consistently seeing a health care provider. The goal of this work is to predict the development of suicidal ideation (SI) in PWE using machine learning methodology such that providers can be better prepared to address suicidality at visits where it is likely to be prominent.

METHODS

The current study leverages data collected at an epilepsy clinic during patient visits to predict whether an individual will exhibit SI at their next visit. The data used for prediction consisted of patient responses to questions about the severity of their epilepsy, issues with memory/concentration, somatic problems, markers for mental health, and demographic information. A machine learning approach was then applied to predict whether an individual would display SI at their following visit using only data collected at the prior visit.

RESULTS

The modeling approach allowed for the successful prediction of an individual's passive and active SI severity at the following visit (r = .42, r = .39) as well as the presence of SI regardless of severity (area under the curve [AUC] = .82, AUC = .8). This shows that the model was successfully able to synthesize the unique combination of an individual's responses to important questions during a clinical visit and utilize that information to indicate whether that individual will exhibit SI at their next visit.

SIGNIFICANCE

The results of this modeling approach allow the health care team to be prepared, in advance of a clinical visit, for the potential reporting of SI. By allowing the necessary support to be prepared ahead of time, it can be better integrated at the point of care, where patients are most likely to follow up on potential referrals or treatment.