Social Determinants of Health Indices and Child Welfare Outcomes In Ncands Data

Background In child welfare research, Social Determinants of Health (SDoH) indices, such as the Social Vulnerability Index or the Area Deprivation Index, are increasingly used either to explore the degree which SDoH may be associated with maltreatment or as control variables to more accurately isolate other associations of interest. To date, however, there have been no published explorations of which SDoH measures are most predictive of child maltreatment outcomes.

Objective This study has several aims. To determine if different SDoH measures are consistent with each other, we examine the correlations between SDoH indices in counties with available National Child Abuse and Neglect Data System (NCANDS) data. Then, we compare the predictive ability of these indices relative to child welfare outcomes (reports, substantiated reports, and foster care entries). Finally, we repeat this latter analysis across different maltreatment types.

Methods We obtained scores for nine publicly available US-focused SDoH composite indices available nationally. Some were at the tract level and were transformed to the county level. We obtained NCANDS Child File (hotline report) data from 2019-2023, aggregated the reports at the county level, and generated outcome variables (total reports, substantiated reports, foster care placement, and subtypes of reports—neglect, physical abuse, sexual abuse) for each county (N=653 counties with available NCANDS data). We also included infant mortality as an outcome variable to allow us a reference outcome from outside the child welfare system. We conducted correlation analysis in three parts: (1) correlation between SDoH indices, (2) correlation between each SDoH index and child welfare outcomes, (3) correlation between each SDoH index and outcomes for subtypes of maltreatment. The last two analyses were conducted twice (with and without clustering at the state level).

Results SDoH indices showed a wide range of correlations with one another (r = 0.11-0.91). All indices, as well as median household income, were significantly associated with infant mortality and child welfare outcomes, with the ADI showing the strongest predictive ability. Several indices demonstrated lower predictive performance than median household income. Among the child welfare outcomes, total report rates generally showed the highest correlations both with SDOH indices and with median household income. Total report rates also showed stronger correlations than correlations for specific maltreatment subtype rates. State-level clustering substantially improved observed correlations, highlighting the importance of controlling for state in national analyses.

Conclusions Researchers should be mindful when using SDoH indices for their research, as any given measure may not necessarily perform better than alternative indices or even a single indicator such as median household income. When using SDoH indices, researchers should clearly justify their choice and compare multiple measures relative to their data and their statistical approach to identify the one that best fits their analysis.