Prediction of child abuse: a prospective study of feasibility.

Feasibility of identifying risk for child abuse prospectively was determined by interviewing 1400 expectant mothers and predicting that 273 were high risk. Nonaccidental child injuries were subsequently reported to authorities for 6% of these versus 1% of the remaining 1127 families. Prediction was only effective for 24 months following interview. Thus prenatal prediction was feasible although the rate of false positive high risk assignment would limit practical application of the interview we used. As a first step to improve prediction accuracy, a group of the 20 strongest predictors was selected from the interview by regression analysis. The correlation of these with abuse was .44 compared to .15 for the original interview. Important predictors included subjective impressions of interviewers, residency transience, untruthfulness, disturbed childhood nurture, unwanted pregnancy, and perhaps conditions that increased parent-child exposure.