Characterizing Disparities in the Treatment of Intimate Partner Violence

Abstract

Exposure to Intimate Partner Violence (IPV) has lasting adverse effects on the physical, behavioral, cognitive, and emotional health of survivors. To this end, it is critical to understand the effectiveness of IPV treatment strategies in reducing IPV and its debilitating effects. Meta-analyses designed to comprehensively describe the effectiveness of treatments offer unique advantages. However, the heterogeneity within and between studies poses challenges in interpreting findings. Meta-analyses are therefore unlikely to identify the factors that underlie disparities in treatment efficacy. To characterize the effect of demographic and social factors on treatment effectiveness, we develop a comprehensive computational and statistical framework that uses Meta-regression to characterize the effect of demographic and social variables on treatment outcomes. The innovations in our methodology include (i) standardization of outcome variables to enable meaningful comparisons among studies, and (ii) two parallel meta-regression pipelines to reliably handle missing data.

Figure 1.

Proposed approach to handling missing data in meta-regression to characterize discrepancies. Left: Filtering of studies missing outcome variables. Center: Full model meta-regression (i.e., including all demographic variables) using imputation for missing data. Right: Individual meta-regression model for each demographic variable by filtering out studies that do not report the variable.

Figure 2.

Normalization of outcome variables to facilitate meaningful comparisons. Bottom Panel: The outcome variables as reported by each study. Upper Panel: Normalized outcome variables we compute to assess the mean improvement in treatment relative to mean violence in the study (left) and the reduction in the heterogeneity of violence in the study (right).

Figure 3.

Distribution of Outcome Variables across Studies

Figure 4.

Effect size of race and ethnicity variables on IPV treatment outcomes. For each ethnicity and race, each row shows the regression coefficient of the fraction of the ethnicity/race among study parents on the respective study outcome for the individual model (left) and the full model (right). Tones of red show negative effect, tones of blue show positive effect, intensity shows significance.

Figure 5.

Effect size of socio-economic variables on IPV treatment outcomes. For each variable, each row shows the regression coefficient of the value (average across all participants in the study) on the outcome variable for the individual model (left) and the full model (right). Tones of red show negative effect, tones of blue show positive effect, intensity shows significance.