Scrutinizing the Cumulative Risk Model for ACE Exposure: Propensity Score Weighting for Multiple Treatment Groups

Background: Adverse childhood experiences (ACEs) are a persistent public health problem linked to problematic developmental sequelae including negative health and mental health. Yet, the majority of research indicating a relationship between ACEs and health outcomes is based on correlational analyses that do not provide an understanding of the causal mechanism linking ACEs and outcomes. It is important to consider the causal impact of ACEs separate from comorbid confounding factors to refine preventive interventions and treatment related to ACEs. The parsing of direct effects on negative sequelae due to ACEs and those due to concurrent factors (e.g., poverty and racial discrimination) is essential for intervention precision. To gain a better understanding of how ACEs effect subsequent child development this analysis attempts to understand the effect of ACEs on children’s subsequent behavioral health, teasing apart potential confounding due to individual child and family characteristics.

Methods: A sub-sample of NCSAW-II data set (children ages 8-17) was used for the analysis, which is a nationally representative sample of youth who have had a child welfare contact. First latent class analysis was used to identify latent sub-groups of ACEs exposures for an estimation of the relation between ACEs and outcomes that accounted for patterns of adverse events. Ten indicators of ACEs were included and class enumeration was selected based on pre-specified fit statistics. After classes were identified, propensity score weighting (PS weighting) for multiple treatment groups was used to achieve balance of potential confounders across groups. PSM balancing weights were used in doubly-robust generalized linear models to predict the relation between ACEs class membership and child internalizing and externalizing symptoms measured by subscales of the Child Behavior Check List (CBCL).

Results: Five classes were identified which represented varied combinations of ACE exposure. Prior to weighting, two classes that were characterized as high ACE exposure with the additional exposure of sexual victimization or caregiver mental health problem had the highest mean symptom scores. After weighting, the results demonstrated a continued impact of latent class membership on behavioral health symptoms. However, some of the results were attenuated and in several instances significant relationships demonstrated without PS weighting proved spurious. For example, Class 1 (High caregiver mental health, high ACE exposure) prior to weighting had significantly higher mean symptom scores than Class 4 (high maltreatment exposure). After weighting and using a doubly robust method, the relationship was no longer significant.

Conclusions and Implications: This study presents compelling evidence about the causal impact of different combinations of ACE exposure on differing degrees of behavioral health symptoms. High ACE exposure continues to show a strong relationship with children displaying high internalizing and externalizing symptoms. Yet this study identified confounding factors that also influenced risk for behavioral health. Findings support prevention strategies that target the co-occurring risk factors for ACEs (e.g., racial discrimination), as well as comprehensive intervention strategies that attempt to mitigate the behavioral health symptoms after ACE exposure.