+61 3 5227 1266
Centre for Pattern Recognition and Data Analytics
School of Information Technology
Locked Bag 20000
GEELONG VIC 3220
S. Rana, S. Gupta, S. Venkatesh, M. Berk, R. Harvey, D. Phung, B. Saha, T. Nguyen, T. Truyen, and W. Luo. An Analysis of Suicide Risk Assessment. Technical Report, TR-PRaDA-03-12, 2012.
Risk assessment and in particular suicide risk assessment is one of the primary mandates of health services. Indeed, in many public facilities, risk assessment is one of the cardinal gate-keeper indicators in triage, determining access to, and nature of care. Risk assessment, in addition, has critical medicolegal consequences. The reliability and validity of suicide risk assessment, however, is an open question. Implicit in the routine use of suicide risk assessment is a tacit acceptance of its validity in terms of positive and negative predictive value. However, there is very little quality data available at a system-wide level on the validity of risk assessment, which is a substantive issue as risk management rests on this foundation. This paper utilises an array of state of the art machine learning tools, that take an agnostic and hypothesis generating view, and examines risk assessment data through a classification perspective. Machine learning has been recently proposed to be a valuable tool for hypothesis generation in psychiatry. The question we examine is: Given three classes of patients (never attempted suicide, attempted suicide and committed suicide), how good is the classification capacity of a sleuth of modern machine learning algorithms? Classification here means that after learning from training data, we can assign the correct labels for unseen cases. To examine this issue, we utilise 15 agnostic machine learning methods to examine associations between risk assessments and completed suicides in a large and representative public hospital database.
[ pdf ]
Deakin University CRICOS Provider Code: 00113B