Predicting suicide risk with machine learning

One in four people in the world will be affected by mental or neurological disorders at some point in their lives.

According to the World Health Organization, around 450 million people currently suffer from such conditions, placing mental disorders among the leading causes of ill-health and disability worldwide.  

The emerging field of Predictive Medicine in mental health has recently generated tremendous interest. An increasing body of genetic and imaging research shows that it is becoming possible to accurately forecast the onset of major psychiatric disorders, such as depression and schizophrenia before people become sick. 

Can machine learning and algorithms be harnessed meaningfully in the area of mental health? Can technology help predict mental illnesses and somehow positively contribute to the diagnoses or treatment? 

Global burden of suicide  

Globally, close to 800,000 people die by suicide every year, making suicide the 15th leading cause of death worldwide.  

The public health implications of such suicide events equate to more than 20 million attempted suicides per year worldwide.  

For each suicide, there are more than 20 suicide attempts. Suicide and self-harm attempts have cascading detrimental effects that negatively impact the individual’s family and their community.  

Can suicide be predicted and prevented effectively?  

Suicide prediction that can lead to prevention 

Early identification of individuals likely to cause harm to themselves will enable mental health caregivers to intervene early before a serious incident takes place. Early intervention means these at-risk patients can receive the right mental healthcare they need to avoid serious events, including suicide.  

study featured in the American Journal of Psychiatry sought to develop and validate models using electronic health records to predict suicide attempts and suicide deaths following an outpatient visit.  

Combining electronic health record (EHR) data and results from a depression questionnaire, the study showed that an accurate predictive analytics model can predict suicide risk in the 90 days following a mental health visit.  

The team found that in the 90 days following a mental health visit, suicide attempts and deaths among patients in the top one percent of predicted risk were 200 times more common than among those in the bottom half of predicted risk. 

The study identified that strongest predictors of a self-harm attempt included prior suicide attempts, the state of their current mental health, substance abuse diagnoses, use of prescribed psychiatric medications, inpatient or emergency room care and high scores on the depression questionnaire.  

More accurate predictive analytics can be determined, by drawing out meaningful data from existing patient datasets to build prediction models that analyse other risk factors such as early-life characteristics, personality type, psychosocial stress and social adversity and so on. 

Having a comprehensive view could also help mental healthcare providers identify whether they should help patents develop a personal safety plan or counsel them about reducing access to means of self-harm.  

Integrating prediction models and data sets for a better solution 

There have been several advances in this space of predictive analysis globally. But we can do better!  

Globally, a more accurate prediction model can be created if all health systems work together to integrate prediction models into an existing process for identifying and addressing suicide risk. 

Risk predictions can supplement clinical judgement and direct clinicians’ attention to where it is most needed. Predictions do not replace a clinical assessment, but they can help providers intervene with the right patients at the right time.  

How Orion Health can help make accurate decisions at the point of care 

The Orion Health Intelligence platform uses aggregated data and intelligent machine learning to draw on information from entire populations to treat and manage a person’s health.  

Machine learning models help eliminate guesswork at the point of care, augmenting clinical decision-making by providing consistent automated processes. Mental healthcare providers will be empowered with the information they need to identify at-risk populations.   

Interested in learning more about Orion Health Intelligence?

The next blog in this series will look at Predictive Medicine and the power of pre-operative risk prediction.