Machine Learning for Prediction of Childhood Mental Health Problems in Social Care

Background Rates of childhood mental health problems are increasing in the United Kingdom. Early identification of childhood mental health problems is challenging but critical to future psycho-social development of children, particularly those with social care contact. Clinical prediction tools could improve these early identification efforts. Aims Characterise a novel cohort of children in social care and develop and validate effective Machine Learning (ML) models for prediction of childhood mental health problems. Method We used linked, de-identified data from the Secure Anonymised Information Linkage (SAIL) Databank to create a cohort of 26,820 children in Wales, UK, receiving social care services. Integrating health, social care, and education data, we developed several ML models. We assessed the performance, interpretability, and fairness of these models. Results Risk factors strongly associated with childhood mental health problems included substance misuse, adoption disruption, and autism. The best-performing model, a Support Vector Machine (SVM) model, achieved an area under the receiver operating characteristic curve (AUROC) of 0.743, with 95% confidence intervals (CI) of 0.724-0.762. Assessments of algorithmic fairness showed potential biases within these models. Conclusion ML performance on this prediction task was promising but requires refinement before clinical implementation. Given its size and diverse data, the SAIL Databank is an important childhood mental health database for future work.