Machine learning for prediction of childhood mental health problems in social care

Katherine Parkin and Efthalia Massou, Primary Care Unit, Department of Public Health and Primary Care. 

What we found... 
1. Machine learning performance in the prototypes was promising, though more work will be needed to develop these models into clinically usable tools. 
2. Linking data together from different sources helps improve prediction of young people's mental health problems. Data linkage allows us to include risk factors relating to clinical, social and environmental exposures. 
3. Further work is needed to ensure models are fair and equitable, avoiding the perpetuation of health inequalities.  
 
Why does it matter? 
Many young people with social care contact have unrecognised mental health problems and therefore don't get the support they need or get it too late. Barriers to accessing mental health support are high, but especially for this group. Accurate prediction tools could help social workers and social care staff identify which young people to refer for mental health support, hopefully improving care pathways between social care and mental health services. Importantly, this would not replace professional judgement but rather be a decision-support tool.  

Link to paper: Machine learning for prediction of childhood mental health problems in social care | BJPsych Open | Cambridge Core