Session: Risk Modeling As a Tool for Maltreatment Prevention and Child Protection (Society for Social Work and Research 22nd Annual Conference - Achieving Equal Opportunity, Equity, and Justice)

185 Risk Modeling As a Tool for Maltreatment Prevention and Child Protection

Schedule:
Saturday, January 13, 2018: 8:00 AM-9:30 AM
Marquis BR Salon 9 (ML 2) (Marriott Marquis Washington DC)
Cluster: Child Welfare
Symposium Organizer:
Emily Putnam-Hornstein, PhD, University of Southern California
The increased availability of administrative data has led to growing interest in tools and statistical models for predicting future adverse events. Predictive risk modeling (PRM) is one such class of tools. PRM is used to automatically generate a risk score for each individual in a given data system, providing a cost-effective means of population risk screening without requiring any additional data entry on the part of clinicians or other staff members. PRM requires: (1) comprehensive and updated administrative data on risk factors; (2) risk-scoring algorithms that are accurate and can be deployed in advance of the adverse event; and (3) effective and adequately resourced service interventions for individuals who are identified as at risk of negative outcomes.

The use of PRM is relatively advanced in health care, reflecting a growing realization that many patients have a multiplicity of risk factors and long-term conditions of which clinicians may not be aware. The same challenges that confront health care providers in identifying high-risk cases to offer more intensive intervention are also present in child protection. Factors that influence the likelihood of future abuse or neglect are multiply determined, dynamic, complex to weight, and not always readily observable. Although maltreatment risk dynamics are arguably even more complex than those in health and include important ethical considerations unique to child protection, the same principles of PRM that have been applied in the health services arena might be usefully applied to the prevention of initial and future episodes of child maltreatment. Indeed, identifying children at risk of adverse outcomes such as maltreatment early in the life course trajectory and offering effective preventive services is a central pillar of the public health approach to child protection, and recent literature has reflected the growing desire for a tool that accomplishes goals similar to PRM in child protection systems.

Although PRM holds promise as applied to varying levels of maltreatment prevention, its use also presents unique and challenging ethical considerations. This symposium assembles four papers that examine different approaches to PRM as a tool for maltreatment risk assessment. Finno-Velasquez will open the panel with a largely descriptive overview of the Structured Decision Making tool, a first generation actuarial risk screening tool widely used by child protection agencies in more than half of states. Putnam-Hornstein will then discuss the use of PRM for primary maltreatment prevention, describing a series of multivariable models built from vital birth records to identify children at high risk of maltreatment. Next, Prindle will introduce how generalized geographically weighted regressions might be employed as a method for identifying the regional influence of risk factors in maltreatment prediction models. Finally, Palmer will present an analysis for classifying maltreatment risk by using several different modern Machine Learning techniques. In addition to specific models and methodologies, all panelists will also discuss trade-offs in modeling accuracy versus interpretability, in addition to implementation and ethical considerations.

* noted as presenting author
Substantiation of Child Maltreatment Risks Identified By Reporters during the Hotline Assessment
Megan Finno-Velasquez, PhD, New Mexico State University; Emily Bosk, PhD, Rutgers University; Stephanie Cuccaro-Alamin, PhD, University of Southern California; Emily Putnam-Hornstein, PhD, University of Southern California
Can We Predict Child Maltreatment Birth? an Exploratory Model Using Californian Birth Records
Emily Putnam-Hornstein, PhD, University of Southern California; Lindsey Palmer, MSW, University of Southern California; John Prindle, PhD, University of Southern California; Rhema Vaithianathan, PhD, Auckland University of Technology
Region Specific Geospatial Models for Prediction of Maltreatment Risk
John Prindle, PhD, University of Southern California; Emily Putnam-Hornstein, PhD, University of Southern California; Rhema Vaithianathan, PhD, Auckland University of Technology
Predicting Risk of Child Maltreatment at Birth: A Comparison of the Classification Accuracy of Machine Learning Techniques
Lindsey Palmer, MSW, University of Southern California; Michael Tsang, MS, University of Southern California; John Prindle, PhD, University of Southern California; Tanya Gupta, MS, University of Southern California; Emily Putnam-Hornstein, PhD, University of Southern California
See more of: Symposia