Maryland Family Law Maryland family law opinions and commentary
In the ever-changing world of technology, professionals are considering how Predictive Risk Models (“PRMs”) can be applied in child welfare decision-making.
There are many decision-making points in child welfare, including hotline screening, the removal of children from family homes, quality assurance for service provision, supervision of the workforce and improving caregiver retention and recruitment that may benefit from the use of PRMs.
“Automated predictive risk models … [rely] on information already captured in case management systems. They generate real-time risk classifications that are more accurate, more consistent, and far less resource-intensive than traditional manual tools.”
PRMs systematically analyze the data entered by caseworkers and stored in the administrative record in child welfare case management systems, including details from prior reports of alleged maltreatment, investigation findings, service episodes and foster care placements.
With that longitudinal history of child maltreatment dynamics, a PRM instrument identifies patterns in historical data and estimates the likelihood that specific future events will occur.
On November 13, 2025, the President signed an Executive Order, “Fostering the Future for American Children and Families,” which mandates the expansion of the states’ use of technology, including artificial intelligence and predictive analytics, to improve caregiver matching and recruitment.
Although Maryland has made efforts to include technology in child welfare work, such as using it to connect youth in foster care to potential kin caregivers, it has not done so in a manner to include PRMs.
A few states have used PRMs as a screening tool for hotline decision making, decision making and supervision at all stages of a case, centralized intake and investigation workflow, and to support supervisory oversight of child maltreatment investigation.
In Maryland, screening a case that was newly reported to the local department of social services, for example, might provide a good starting point to apply a PRM.
A PRM may draw on the information already in the Department of Social Services’ information system and then be augmented by the newly reported information from the case being processed.
A PRM may execute a decision analysis and develop a consistent, objective recommendation for next steps, such as whether to implement an alternative response or commence a maltreatment investigation, and which could then be considered by a supervisor for action.
Caution, however, is warranted when using PRM programs; PRM’s analysis might be biased based on the question it is programmed to answer.
For example, an estimation of the likelihood of a specific child being subjected to future abuse again is different from an estimation of the likelihood that a child welfare agency would make a removal again.
Additionally, a PRM trained on data already in the system may “learn” patterns that reflect the system’s behavior rather than the patterns that predict actual harm.
For example, in a jurisdiction where most reports of alleged abuse or neglect are not acted upon by the child welfare agency, if the PRM were to consider previous abuse or neglect reports to calculate the probability of future harm to a child, the resulting recommendation might not predict a likelihood of harm.
Finally, a PRM, trained on historical data in a child welfare system, is unlikely to distinguish those actions taken by the child welfare agency that represent disparate practices.
Indigenous or Black women and children and those experiencing poverty are disproportionately subject to the scrutiny of child welfare systems.
Given these disparities, using historical data as a basis for analysis will only reinforce the disparities.
To address possible disparities, PRMs must be validated “against outcomes that are not solely derived from child welfare processes, such as hospitalizations for injuries and other objective indicators of serious harm.”
Clearly, a PRM tool cannot and should not replace the subtleties of decision-making that experienced child welfare staff utilize on a daily basis, but, such a tool, incorporated in a meaningful, may supplement that decision-making.
Joan Little is a Chief Attorney at Maryland Legal Aid.
