Background: Every year approximately 800,000 people die by suicide worldwide, accounting for 1-2 in every 100 deaths. It is always a tragic event with a huge impact on family, friends, community and professionals. Unfortunately, suicide prevention and the development of risk assessment tools have been hindered by the complexity of the underlying mechanisms and the dynamic nature of people's motivations and intent. Many of those who die by suicide have had contact with health services in the preceding year but identifying those most at risk is a challenge.
Objective:To explore the feasibility of using artificial neural networks (ANNs) based on routinely collected electronic health records (EHRs) to support the identification of those at high risk of suicide when in contact with health services.
Methods:Using the Secure Anonymised Information Linkage Databank UK, we extracted those who died by suicide between 2001 and 2015 and paired controls. Looking at primary (general practice: GP) and secondary (hospital admissions) EHRs, we built a binary feature vector coding the presence of risk factors at different times before the contact leading to death. Risk factors included: GP contacts and hospital admissions; diagnosis of mental health, injury and poisoning, substance misuse, maltreatment or sleep disorder; and prescription of opiates or psychotropics. We trained simple ANNs to differentiate between cases and controls and interpreted the output score as the estimated suicide risk score. We assessed system performance with 10x10 K-Folds repeated cross-valadiation and studied system performance and behaviour by representing the distribution of estimated risk across cases and controls and the distribution of factors across different estimated risks.
Results:We extracted a total of 2,604 suicide cases and 20 paired controls per case. Our system obtained an error rate of 26.78% ± 1.46 (64.57% of sensitivity and 81.86% of specificity). While the distribution of controls was concentrated around estimated risks < 0.5, cases where almost uniformly distributed between 0 and 1. Prescription of psychotropics, depression and anxiety and self-harm increased the estimated risk by ~0.4. At least 95% of those presenting these factors were identified cases.
Conclusions:Despite the simplicity of the implemented system, the proposed methodology obtained an accuracy similar to other published methods based on specialized questionnaire generated data. Most of the errors came from the heterogeneity of patterns shown by cases, some of which were identical to those of controls. Prescription of psychotropics, depression and anxiety and self-harm were strongly linked with higher estimated risk scores, followed by hospital admission and long-term drugs and alcohol misuse. Other risk factors such as sleep disorder and maltreatment had more complex effects.