The Allegheny Family Screening Tool

Allegheny County’s Department of Human Services began using the Allegheny Family Screening Tool (AFST) in August 2016 in its child-welfare intake office. The county describes it as a predictive risk model supporting screeners responding to suspected abuse or neglect reports. It analyzes integrated data from the county data warehouse and estimates the likelihood of a child experiencing harm within two years. County policy describes the score as decision support alongside staff experience and clinical expertise, not a decision made by the tool alone. The public documentation shows how an institution’s data and workflow shape a score. The 2019 impact-evaluation summary describes a 1-to-20 Family Screening Score and an “auto screen-in” recommendation for certain high scores, while also explaining that the recommendation was not an automatic investigation mandate. It reports that several policy changes accompanied AFST implementation, including changes to mandatory field screens. Therefore, observed outcomes cannot be attributed to the score alone without a design that separates those changes. Critics have raised concerns that integrated service records may reflect poverty, service access, and surveillance as well as child safety. More records for one family can create a different measurement surface than for a similar family using private services. The county’s 2024 summary of research on AFST and comparable models reported reduced racial disparities in investigation rates in the studied context, especially among high-risk referrals. That finding does not establish that every model use is fair or that any particular score is correct; it describes evaluated outcomes under specific policies and periods. A responsible review asks what the tool predicts, what data it uses, how human screeners use its output, and which policy changes occur alongside it. Transparency should include validation, score meaning, limits, override practices, and impact evaluations. Families and affected communities need meaningful ways to question data and processes. A risk score is not proof of abuse and should not substitute for evidence or professional judgment.