Predictive risk modelling as a tool for supporting decision making in social work with children and families at risk.

In this workshop we will discuss findings from a Danish research project (2017-2024) on predictive risk modeling's (PRM) impact on enhancing child and family welfare decisions. Through simulated scenarios, interactive group and plenary discussions, we aim to explore implications for practice, policy, and future research. The workshop will begin with an overview of the predictive risk model's development process. Subsequently, a simulated scenario will illustrate the application of our model to a specific case, providing insights into how social workers cope with professional discretion and judgment when assessing risk in child and family welfare services.

Background:
The number of notifications concerning vulnerable children experienced a 40% increase between 2015 and 2019. A risk assessment of these notifications is mandated by Danish law.
With an aspiration to enhance accessibility to existing data, this project harnesses comprehensive administrative data. The overarching research question in the project is to investigate how PRM can support social workers assessments of maltreatment in specific cases with notifications. The need for accuracy and reducing noise is a crucial in these decisions of potential further investigation.
The hypotheses of the project were that technological innovations have made it possible to develop tools such as predictive risk modelling that can increase consistency and reduce errors when social workers asses similar cases with children and families at risk.
This may reduce social worker bias and inequity. It also poses many risks for children and families of being subject for discrimination and perpetuating systemic biases. If critical human reflections are not applied when using PRM, it can reduce empowerment and autonomy for social workers and service users.

Methods:
With the aim of understanding practical implications and perceptions among social workers a multifaceted approach working with quantitative and qualitative data was chosen to refine the predictive model and the visual output of the decision support tool.

Findings:
Building on our initial data analysis, social workers express concerns that decision support tools may limit their discretionary power, even as they acknowledge the tools' potential to ensure alignment in their decision-making. The data suggest that uniformity in assessments is essential for fairness, but individual and organizational factors are often also emphasized. Social workers find that decision support tools illuminate these factors, enhancing transparency. This prompts discussions on the extent of information to be shared with families during collaboration. To increase transparency in decision-making questions could be asked about the competencies and responsibilities of social workers in understanding and communicating PRM in collaboration with families and also possibly to support family autonomy.

Conclusions and Implications:
After assessing models with different degrees of complexity, an XGBoost model was chosen, using out-of-home placement as an indicator for child maltreatment. This model achieved an accuracy rate of 84% in distinguishing between maltreatment and non-maltreatment cases.
Our findings suggest that social workers may find PRM useful in terms of collaborate with their colleagues, management, and service users to support decision-making with a predictive model. Despite the potential benefits, navigating legislative complexities and updating digital systems can make social work cautious, influenced by both uncertainties and conflicting stakeholder interests.