Elucidating the Effect of Alcohol-Related Services on Abstinence, Recovery, and Familial Functioning: A Propensity Score Matching Approach

Alcohol use disorder (AUD) causes immense global burden and has a significant impact on economic and societal functioning. Efficacious treatments for AUDs have been well-established within the literature, however the most commonly accessed treatments for AUD are alcohol-related services, such as self-help groups, outpatient clinics, and detoxification centers. Though studies suggest these services are effective at treating AUDs, there are numerous differences between individuals who receive alcohol-related services and individuals who do not, causing selection bias. Furthermore, current studies of alcohol-related services frequently define recovery outcomes as abstinence, which reduces variability in viable recovery outcomes, such as reduction of drinking behaviors. In addition, reduction in drinking and alcohol-related problems should theoretically have an impact on broader aspects of functioning, such as familial functioning. Improved familial context may reduce risk to family members, who are otherwise at heightened risk for emotional and behavioral problems when living with a family member with AUD. The current study investigated the effect of alcohol-related services on binary and continuous drinking outcomes after eliminating selection bias using multiple propensity score approaches, to identify the best methodology for a high-risk community sample of individuals with AUD. Propensity scores were created using logistic regression approaches and boosted regression trees. Matching, weighting, and subclassification were used, and matching was performed both using greedy and global approaches. Results suggested subclassification was the most successful method for real world alcohol-related services samples with moderate sample size. Moreover, findings demonstrated that boosted regression approaches were less successful than logistic regression approaches at minimizing the effects of selection bias on known confounding variables that are highly related to group selection. In addition, after removing the effects of selection bias, there were no significant difference between participants who received alcohol-related services and the comparison control group on drinking or family functioning, though both groups reduced drinking from pre- to post-alcohol-related services receipt. Findings suggest careful selection of quasi-experimental methods is warranted in real-world samples, to ensure optimal removal of selection bias. Moreover, future studies should continue to clarify the profile of individual that benefits from alcohol-related services to inform intervention efforts.