Myrna R. Epstein, PhD, MPH, Epidemiology, Yolo County Health Department, 10 Cottonwood Street, Woodland, CA 95695, 530-666-8645, myrna.epstein@yolocounty.org and Mitchell Watnik, PhD, Statistics Department, University of California at Davis, UCD Statistical Laboratory, Davis, CA 95616.

PURPOSE: A prospective study was conducted to determine if data obtained from standard risk assessments conducted through the California Comprehensive Perinatal Services Program for low-income women could identify infants who were subsequently abused or neglected. METHODS: Records were matched to a state child welfare database when infants were 20-28 months old. Of the 637 records, 22% had incomplete information about individual risk factors. By excluding incomplete records, analyses disregard systematic differences that may exist between complete and incomplete cases and reduce the number of observations yet results are applied to the whole cohort. To include all records, multiple imputation procedures were used. Instead of filling in a single value for each missing value, multiple imputation procedures replace missing values with a set of plausible values, inferred from the range of existing values. The imputed data sets were analyzed using standard multivariate logistic regression to identify factors that would distinguish between mothers of maltreated infants and those with no reported maltreatment. RESULTS: Nearly 7% of the infants had substantiated incidents of abuse or neglect during their first two years. Based on medical, psychosocial and substance use perinatal assessments, the sensitivity rate for correctly identifying maltreated infants ranged between 74-79 percent. CONCLUSIONS: High risk families may be more likely to have episodic care and thus incomplete records. If their cases were excluded from analyses of risk, the results may not accurately reflect the true picture of the relationship between family risk factors and outcome of early abuse and neglect.

Learning Objectives:

• Identify an alternative method to retain records with missing values for use in multivariate analyses;
• Describe the strengths and limitations of using imputation methods for dealing with data sets with missing values

Keywords: MCH Epidemiology, Family Violence

Related Web page: www.yolocounty.org/org/health/reports.asp

Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.