The "12-step" mutual help organization Alcoholics Anonymous (AA) is the most commonly sought source of help for alcohol problems in the United States. Researchers have therefore worked to assess the outcomes of participation in AA. Although AA research has improved dramatically in methodological quality over the years, generating an unbiased estimate of AA's impact remains a challenge. Fundamentally, as a peer-controlled and widely available organization, AA is difficult to study in randomized trials, meaning that estimates of its impact tend to be biased by self-selection. However, now that a number of federally-funded randomized clinical trials (e.g., Project MATCH) have studied interventions designed to facilitate AA participation (e.g., 12-step facilitation counseling), an unprecedented opportunity exists to generate an unbiased estimate of AA's effectiveness. Specifically, the data from these clinical trials can be re-analyzed using instrumental variables modeling, a technique widely employed in economics to remove bias from outcome estimates. Using randomization to a 12-step facilitating intervention as an instrument, this R21 developmental/exploratory grant proposes to assess the impact of AA in 6 previously conducted trials enrolling a total of over 3,000 patients (with excellent representation of women and minorities). This unbiased AA effectiveness estimate will not only be of interest in itself, but wil also provide a useful metric against which to judge prior AA outcome research (i.e., to determine if the unbiased instrumental variables estimate of AA's impact resembles those estimates generated using other statistical procedures in observational studies). Data from the six trials will be contributed by their principal investigators, all of whom are involved in the proposed project team. The key outcomes will be alcohol consumption, alcohol dependence, and alcohol- related consequences, all of which were assessed in the six randomized trials using validated measures. The unbiased effect of AA will be assessed both for a simple measure of meeting attendance and for more encompassing measures of AA participation (e.g., reading literature, sponsoring other members). Every trial to be analyzed included a randomly assigned intervention designed to promote AA involvement. Randomization condition will therefore serve as an excellent instrumental variable because it is fully observed and strongly predicts AA involvement. Models will be estimated for each individual trial and in a stacked dataset including all trials. Secondary analyses will focus on outcomes other than alcohol (e.g., marital satisfaction, psychiatric symptoms) and on comparing the results of the instrumental variables models with those of prior AA studies that used different methods to reduce bias in AA effectiveness estimates (e.g., MANCOVA, propensity scores).
This study uses secondary analysis of previously conducted randomized clinical trials to estimate the effects of Alcoholics Anonymous participation on alcohol-related and psychosocial outcomes. The central statistical method of instrumental variables modeling produces an estimate of Alcoholics Anonymous'impact that is free of self-selection bias.
|Humphreys, Keith; Blodgett, Janet C; Wagner, Todd H (2014) Estimating the efficacy of Alcoholics Anonymous without self-selection bias: an instrumental variables re-analysis of randomized clinical trials. Alcohol Clin Exp Res 38:2688-94|