Dynamic prediction characteristics of the Child Abuse Potential Inventory

Risk measures are commonly used to evaluate outcomes in child abuse prevention and intervention programs. This study examined whether pre-intervention to post-intervention changes on the Child Abuse Potential Inventory (CAP) Abuse Scale corresponded to actual changes in risk for future reports of maltreatment and evaluated the validity of several algorithms for classifying clinically significant change.

Participants in the study were 459 parents participating in any one of 27 community-based family preservation and family support programs. Most parents were low-income mothers with a variety of social risk indicators, about a third of whom would be classified as high-risk by the CAP Abuse Scale. Participants were administered the CAP at program enrollment, then at completion of the intervention (median time=150 days), then followed for an average of approximately 2 years for future official maltreatment reports. Dynamic predictive validity of the CAP Abuse Scale was modeled by comparing survival models using a time-dependent structure of pre- and post-intervention scores to identically structured models using only a pre-intervention score.

Pre-intervention CAP Abuse Scale scores demonstrated incremental future predictive validity. However, score changes failed to correspond to changes in likelihood of future abuse. Models using pre-intervention scores only were more predictive than time-dependent score models, and pre-intervention scores were better predictors than post-intervention scores of post-intervention CPS referrals. Common algorithms for classifying clinically significant change yielded results that could be counter-intuitive and misleading. For example, participants classified as improved on these algorithms were actually at similar or even higher risk than those classified as unchanged or worse.

The results strongly supported the static predictive validity of the CAP and the use of the CAP for screening purposes. The results did not support the dynamic predictive validity of the CAP. Results of exploratory analyses suggested the possibility that the changes observed on the CAP Abuse Scale reflected changes in subscales assessing subjective distress or parenting attitudes, which may be markers for initial risk but when changed, do not necessarily translate into actual changes in future maltreatment behavior. Although replication and extension are needed before drawing firm conclusions, the current study raises questions about the common practice of using risk instruments as proxy measures for child maltreatment risk in intervention and prevention programs.