Improving Risk Estimation in Observational Research on Child Maltreatment: Establishing Methods for the Effective Control of Contamination

PROJECT ABSTRACT Replication and reproducibility failures, as evidenced by variation in the significance and magnitude of effect size estimates for specific outcomes examined across prospective cohort studies, have weakened causal inferences about the public health impact of child maltreatment. Contamination, when subjects enrolled in a comparison condition are exposed child maltreatment prior to study entry or during longitudinal follow-up, is both common in child maltreatment research and a major contributor to variation in the significance and magnitude of effect size estimates by minimizing between-group differences when they truly exist. Despite these implications, there are no established methods for controlling contamination in child maltreatment research. For the first time, this application will test multiple strategies for controlling contamination in prospective cohort research with the child maltreatment population. The Investigative Team will achieve this goal via secondary analysis of existing data from the Longitudinal Studies of Child Abuse and Neglect (LONGSCAN; N=1354), a prospective cohort study of child maltreatment from birth through age eighteen. A multi-method approach of official case records and self-report assessments measured repeatedly across child development will be used to maximize sensitivity for detecting contamination and establishing its prevalence in LONGSCAN. Two innovative methods for controlling bias in observational research, doubly robust propensity score and augmented synthetic controls, will be used to control contamination in the LONGSCAN cohort. These two methods bring significant potential as they offer unique advantages for controlling contamination at study entry and throughout longitudinal follow-up. The performance of doubly robust propensity score and augmented synthetic control models will be benchmarked against models that: 1) do not control contamination, 2) control contamination by removing subjects from statistical analysis, and 3) control contamination by estimating it as a covariate or moderator of child maltreatment effects. This comprehensive modeling approach uses statistical efficiency principles when evaluating the performance of each method for controlling contamination, balancing constraints on needed sample size, impact on statistical power, and change in the significance and magnitude of risk estimates. Given the larger goal to identify which methods provide the most accurate estimates of child maltreatment under different empirical conditions, this application will evaluate results from all five models using simulations that both mimic the data structure of LONGSCAN and extend to conditions most likely encountered in future prospective cohort research, such as variations in contamination prevalence and sample size. Effectively controlling contamination will strengthen causal inferences about the public health impact of child maltreatment while having the greatest potential to serve multiple, key stakeholders, including scientists conducting child maltreatment research as well as child welfare policy makers deciding when to allocate services to children and families exposed to maltreatment.