Integrating Epidemiologic and Administrative Data to Estimate the 5-Year Risk of Child Maltreatment in Alaska: Accounting for Non-Linkage Misspecification Due to Censorship

BACKGROUND: Longitudinal birth cohort studies that link vital statistics data with child welfare reports have become increasingly popular for conducting child maltreatment research and surveillance. These linkage studies however, are often limited in their ability to follow the entire cohort over time. The impact of non-linkage misspecification due to migration or other censorship is largely unexplored. This study used novel Alaska data sources to quantify the impact of non-linkage misspecification and single source maltreatment report use on maltreatment estimates produced.

METHODS: We calculated the 5-year risk of first maltreatment by linking the 2009 Alaska Pregnancy Risk Assessment Monitoring System (PRAMS) sample with child protective services (CPS), Anchorage police, child advocacy center, vital statistics, and the permanent fund dividend (PFD) database. The PFD database contains records of Alaska residents that apply for oil revenue dividends and is a unique data source used to track the complete cohort over time. We validated the PRAMS sample estimates by linking the full 2009 birth cohort with CPS data based on simulated linkage thresholds. We calculated the cumulative risk of experiencing a multi-agency report, CPS report, screen in, and substantiation of maltreatment and compared these with published estimates where available.

RESULTS:   The PRAMS sample estimates were within one percentage point of the full cohort estimate. The crude five-year risk of experiencing at least one multi-agency report, CPS only report, screen in report and substantiated report of maltreatment among the 2009 Alaska birth cohort were 25.5%, 24.9%, 19.4%, and 8.0%, respectively. These risk estimates are on average 10% (8.1% - 10.9%) higher than those produced without accounting for censorship. Relative to published cumulative risk estimates, the 5-year risk of a CPS report and substantiated maltreatment report is 1.64 and 1.40 times as high in Alaska (p<0.01).

CONCLUSIONS: This is the first known study to calculate the 5-year cumulative risk of multi-agency contact, CPS contact, screened in, and substantiated reports of child maltreatment using a representative birth cohort sample derived from linking epidemiologic and administrative data sources. Birth cohort studies that cannot account for censoring likely underestimate the true magnitude and are limited in developing full adjustment sets needed for etiologic investigation. Although “population-based” studies using birth certificate and child welfare data provide useful insights they are limited. Comprehensive epidemiologic linkage investigations that maintain the underlying population distribution are needed to provide more accurate magnitude estimates.