Predicting child fatalities among less-severe CPS investigations

This research is focused on the discovery of variables distinguishing Child Protective Services (CPS) maltreatment-related fatality cases from non-fatality CPS cases. Using only the cases that did not initially present severe features, information from CPS investigations prior to a child fatality was compared to information from investigations that did not result in a fatality. Logistic regression was used to distinguish less-severe non-fatality cases from less-severe fatality cases. Findings resulted in three categories of predictive indicators: 1) more actionable and indicative of reduced risk, 2) mostly more actionable and indicative of increased risk, and 3) less actionable and indicative of increased risk. Implications for the field in terms of risk assessment, practices, and training are discussed, as are the implications for future research.