QLD Child and Youth Mental Health Service Users: Identifying the young people who are at risk of self-harm is a priority for the mental health specialist service providers – especially when it comes to high stakes decisions such as when to discharge a patient back to their General Practitioner. Tools that provide early identification of those at risk (and conversely those patients who are at minimal risk) will enhance the existing mental health services by providing decision-support tools for use along with clinical judgement as to the most appropriate interventions and prevention strategies. Yet, service providers find it challenging to predict suicidal ideation due to the broad range of drivers related to social, biological, psychological, and contextual aspects. In this project we explore the usefulness of machine learning to overcome this challenge considering the ability to incorporate high-dimensional interactions between a large set of variables. Using a clinical episode dataset spanning from 2010-2018 from Queensland Paediatric Community Mental Health Services, we have developed a machine learning model to identify the patients who are at the risk of suicidal ideations. Preliminary modelling using LASSO and Random Forest, shows acceptable AUC (78%-81%) and the potential to support the clinical decision making. However, the utility of such tools depends on clinical acceptability and social licence. The next stage of this study is a qualitative exploration with clinicians using past scored cases to understand whether the tool is valuable and would have led to different clinical decisions if the score had been available at the time.

Aboriginal and Torres Strait Island Health Service Users: The existing health predictive models tend to only flag patients at risk of narrow physical health outcomes or resource use such as re-admissions. In this project we want to explore whether the data exists to support a broader concept of good health, including patients reporting their own authentic perceptions of health and wellbeing which better reflects the Aboriginal and Torres Strait Island idea of what wellness is. In collaboration with the Institute for Urban Indigenous Health (IUIH), we have investigated the possible use of machine learning tool to help identify patients at an Aboriginal and Torres Strait Islander clinic who could benefit from proactive services. We have built preliminary predictive risk models targeting a few mental health and wellness outcomes that is available in the primary care data related to the demographics, medical history, medication history, previous appointments, alcohol use and fruit and vegetable consumption of patients. There are two next steps of this project. First is to deploy the tool at the clinic and conduct an RCT to verify whether using the tool for recruiting patients for additional programs or outreach can help mitigate poor mental health and wellness outcomes. The second is to work with the clinic doctors to advocate at State and Federal level and we need to start collecting more holistic measures of wellness that are better suited to the community values so that we can train future machine learning tools with more useful outcomes than the sort of proxy outcomes that are currently available to us.

Project members

Dr Gayani Tennakoon Mudiyanselage

Research Fellow in Data Scienc
Institute for Social Science Research

Professor Rhema Vaithianathan

Professor (Social Data Analytics)
Institute for Social Science Research

Dr Joemer Maravilla

Research Fellow
Institute for Social Science Research