Identifying Patients at Risk for Suicidal Ideation and Key Factors Responsible by Means of a Self-Explaining Neural Network

Deis, Lukas

Identifying Patients at Risk for Suicidal Ideation and Key Factors Responsible by Means of a Self-Explaining Neural Network

Files

Deis, L. s4588827-BSc-Thesis-2021.pdf (1.23 MB)

Authors

Deis, Lukas

Issue Date

2021-02-14

Language

en

URI

https://theses.ubn.ru.nl/handle/123456789/15677

Abstract

Suicide is a major cause of death in all of Europe, and it is on the rise. Worldwide, suicide is the second most commom cause of death in the age group of 15 to 29 years (Bachman, 2018). Suicide experiences a treatment gap of around 50%, meaning that only half the people that die by suicide receive treatment beforehand (Bruffaerts et al., 2011). Thus, it is worthwhile to investigate whether an automated solution could be used to identify people in the general population that may be at risk for suicide. Not only being able to detect suicidal ideation, but also giving insight into the underlying issues, is key to better treatment (WHO, 2019b). Therefore this should be done using explainable methods. A self explaining neural network (SENN) was trained to predict if a person suffers from suicidal ideation and state which factors were important in that prediction. For this research data from the MIND-SET study was used, which is a study by the Radboudumc. It includes 705 participants, of which 574 suffer from common psychiatric disorders and 131 are healthy controls. The best performing model had an accuracy of 85.3% on the test set with a sensitivity of 79.1% and a specificity of 89.1%; the positive predictive value was approximately 81.538% and the negative predictive value was 87.5%. Most important risk factors are from two questionnaires. One is the Outcome Questionnaire, designed to capture a low quality of life. The second is the Inventory of Depressive Symptomatology, designed to give insight into depression. Some factors seem to significantly reduce the risk, too, such as a score describing a good mental health. Perhaps surprisingly, most other relevant risk reducing factors stem from the measurement of autism characteristics.