Both Paper 1 and Paper 2 focus on mortality among youth involved with CPS and examine relevant risk factors using linked administrative data. Paper 1 is the first study to examine the risk of death among a population of youth eligible for extended foster. In the study, the CPS records of youth who were in care in 2010 and 2011 between age 10 and 18 (n=31,477) were linked to the stateÃ¢â‚¬â„¢s vital death records. The demographic and case characteristics of all youth and the nature of death among decedents were examined inferentially using generalized linear models and multinomial logistic regression. Among decedents (n=260), 45% were a result of homicide and 32% were due to accidents. The findings underscored the disparate risk for violent deaths in young adulthood among Black and Hispanic youth. Paper 2 examines the association between child and adolescent suicide and a history of child protective service (CPS) involvement. This study uses linked birth, death and CPS records to demonstrate that over half of children and youth born in California in 1999 or 2000 and died by suicide had contact with CPS prior to death. Children and youth with CPS history had three times the odds of suicide compared to children with no such history. There were no discernible differences between children with a history of substantiated child maltreatment or foster care placement compared to those with an allegation only. Results indicate that suicide risk is not only concentrated among the small, and vulnerable, group of foster youth, rather the much larger universe of children who are alleged victims of child maltreatment are at heightened risk as well.
To test the potential of using an algorithmic decision aid to proactively identify foster youth at high risk of aging out, Paper 3 examined machine learning algorithms to assess the risk of aging out based on early CPS engagement before age 12. Predictive models were developed using 28 years (1991-2018) of CPS records and three classification algorithms (penalized logistic regression, random forest, and gradient boosting decision tree) and evaluated using F1 score. The results propose the potential for predictive analytics to move child welfare agencies toward a more data-informed approach to evaluating risk and providing services.220.127.116.11 on 5-8-2020-->