OBJECTIVES: Outcome research has been applied successfully in healthcare for years, but its application in social care is scarce. In the UK, over 65,000 children live with almost 55,000 foster families. Every 22 minutes there is a child entering the UK care system. In 2017, the Children’s Commissioner for England called for the development of cutting-edge outcome analysis to shine a light on the issue of stability, the most important aspect of children’s experience in care. Through the use of a retrospective database study, a real-world outcome product for foster-care can be developed using advanced machine learning to provide early warning to potential placement disruptions and to thereby maximise stability outcomes for fostered children.

METHODS: For each unique placement, 368 complex features were generated to unravel the hidden layers behind the obvious placement features through our systematic feature-extraction. Follow-up periods of 2-month, 5-month, 1-year and 2-year were set post index dates to reflect the tension period between a newly formed foster pair. Random forest classification method was used to train the operational stability model based on the calculated features list. For this pioneer study, only the placement stability at 2-month is predicted and sensitive information about the child is not disclosed.

RESULTS: CONCLUSIONS: This pioneer study offers real-world insights to suggest that machine learning provides the pragmatic tool for prediction of placement stability and early warning of fostering placement disruptions.