Background and rationale
Globally, 800,000 people die from suicide every year, a rate of 10.6 per 100,000 person-years. Suicide is the leading cause of death for those aged 20 to 34 in the UK with a rate of 11 deaths per 100,000.
Prediction of suicide is difficult; prediction tools have been developed based on well-established risk factors and traditional statistical methods, but have not been accurate enough to be widely adopted in healthcare settings and are not recommended by the National Institute for Health and Care Excellence.
The 2012 suicide prevention strategy and subsequent Health Select Committee's inquiry into suicide prevention highlights the role GPs play in identifying and managing suicide risk. Machine learning is an emerging technology in healthcare research and has been applied to answer questions using CPRD data previously but, to our knowledge, not to suicide risk assessment.
Previous machine learning models for suicide risk prediction have used the area under the receiver operator characteristic curve (AUC) to compare model performance. Models have shown good AUCs but when calculated, positive predictive value still remains low. We believe that model performance would be better assessed by calculating the net-benefit of the model. Calculating net-benefit involves understanding what intervention might be offered to patients who are considered at risk by the model and incorporates the risks involved with false positives and false negatives. We believe this will give a more interpretable model output and is the next step towards producing a viable clinical risk scoring tool
To build a machine learning model for suicide from Primary Care data which can produce a net-benefit to patients.
The primary outcome will be suicide.
James Bailey - Chief Investigator - King's College London (KCL)
James Bailey - Corresponding Applicant - King's College London (KCL)
Alex Dregan - Collaborator - King's College London (KCL)
Grigorios Loukides - Collaborator - King's College London (KCL)
Johnny Downs - Collaborator - King's College London (KCL)
Rina Dutta - Collaborator - King's College London (KCL)
Vibhore Prasad - Collaborator - King's College London (KCL)