Suicide is a leading cause of death globally. People who develop mental health conditions or suicidal thoughts are at higher risk of dying from suicide and will often seek help from their general practitioner (GP). Part of the role of the GP in this situation is to think about who is most at risk of suicide. One reason this is difficult is that only a small number of people with thoughts of suicide act on them. Another is that there is no accurate tool to help show what makes someone higher risk than another. Finding a way to accurately predict which people are most at risk of dying from suicide could result in them getting the help they need sooner, and potentially prevent these deaths.
Though the risk factors for suicide have been well studied in people who are under the care of mental health teams or who have harmed themselves in the past there is not much research looking at the risk factors for people who are under the care of their GP. Only a quarter of people who die from suicide have seen a psychiatrist in the year before their death, compared to 9 out of 10 people who have seen their GP. By exploring what factors make people who see their GP about their mental health more likely to die from suicide we hope to inform GP’s assessment of suicide risk and see which combination of risk factors can best predict who is most likely to die.
Objective
To improve our understanding of suicide risk in primary care attenders.
Primary outcome
Suicide will be defined as a recording of death by self-injury on the linked Office for National Statistics (ONS) Death Registry using ICD-10 codes X60-X84 “Intentional self-harm and event of undetermined intent" or Y87.0 “sequelae of intentional self-harm”.
Primary exposures
We have chosen a list of exposures a priori from a literature review of risk factors and discussion between co-authors who have a range of clinical and epidemiological backgrounds.
Study design
This study will have a cohort design.
Methods
Methods will compare regression (Cox Proportional Hazards for time to event and linear regression for risk) models to a range of machine learning techniques to produce predictive models for suicide. The machine learning models currently planned will be random forest and neural networks however this is a rapidly changing field and we are currently considering the use of ensemble techniques. Classification trees will be used to look for novel risk factors.
Linked data
To establish hospital admissions for self-harm and outpatient appointments linked Hospital Episode Statistics (HES) outpatient data will be used. Index of Multiple Deprivation linkages will be used to establish the impact of deprivation on risk. ONS Death Registry data will be used to define the outcome as well as being used to validate the cause of death being recorded as suicide in the core CPRD dataset.
Death from suicide.
Irene Petersen - Chief Investigator - University College London ( UCL )
James Bailey - Corresponding Applicant - University College London ( UCL )
Irene Petersen - Collaborator - University College London ( UCL )
Irwin Nazareth - Collaborator - University College London ( UCL )
Muhammad Qummer ul Arfeen - Collaborator - University College London ( UCL )
Natalie Fitzpatrick - Collaborator - University College London ( UCL )
Patricia Schartau - Collaborator - University College London ( UCL )
Seena Fazel - Collaborator - University of Oxford
Muhammad Qummer ul Arfeen - Collaborator - University College London ( UCL )
Natalie Fitzpatrick - Collaborator - University College London ( UCL )
HES Outpatient;ONS Death Registration Data;Patient Level Index of Multiple Deprivation Domains;Practice Level Index of Multiple Deprivation Domains