Utility of Classification and Regression Trees (CART) analysis in assessing the impact of a behavioral/psychosocial intervention on resolving multiple risks among high risk pregnant minority women

The purpose of this analysis is to evaluate the utility of Classification and Regression Trees (CART) in identifying sociodemographic factors that impede/enhance the efficacy of a behavioral intervention designed to resolve multiple risk factors during pregnancy.

An RCT was conducted in Washington, DC. Minority pregnant women were screened for smoking, environmental tobacco smoke exposure (ETSE), depression, and intimate partner violence (IPV). Data at baseline (BL) and two follow-up interviews were collected. 1,044 African American women were randomized to intervention (IG) or usual care (UCG) groups. IG received educational/counseling sessions targeting reported risks.

The overall comparison showed no significant intervention effects between IG and UCG prior to delivery. CART identified IG subgroups benefiting most/least from the intervention. IG vs. UCG non-smokers with ETSE at BL were less likely to report ETSE at delivery (51% vs 62%, p=.013). Smoking was reduced for IG vs. UCG in primiparas (p=0.080), reaching significance for those with at least high school education (p=.031). IG smokers with IPV with or without depression receiving >4 sessions were less likely to report smoking at delivery vs. women receiving <4 sessions (p=.057). ETSE was reduced among those in the IG vs UCG in Medicaid and WIC recipients (p=0.042, 0.041). Depression in IG was reduced among non-drug users as compared to illicit drug users (p=.056).

Overall effects of a behavioral intervention may mask underlying subgroup differences. Using CART methodology helps identify such differences, and assists clinicians in the selection of eligible populations most/least likely to benefit from targeted clinical interventions.