Brief Alcohol Interventions are Effective through 6 Months: Findings from Marginalized Zero-inflated Poisson and Negative Binomial Models in a Two-step IPD Meta-analysis

Abstract

To evaluate and optimize brief alcohol interventions (BAIs), it is critical to have a credible overall effect size estimate as a benchmark. Estimating such an effect size has been challenging because alcohol outcomes often represent responses from a mixture of individuals: those at high risk for alcohol misuse, occasional nondrinkers, and abstainers. Moreover, some BAIs exclusively focus on heavy drinkers, whereas others take a universal prevention approach. Depending on sample characteristics, the outcome distribution might have many zeros or very few zeros and overdispersion; consequently, the most appropriate statistical model may differ across studies. We synthesized individual participant data (IPD) from 19 studies in Project INTEGRATE (Mun et al., 2015b) that randomly allocated participants to intervention and control groups (N = 7,704 participants, 38.4% men, 74.7% White, 58.5% first-year students). We sequentially estimated marginalized zero-inflated Poisson (Long et al., 2014) or negative binomial regression models to obtain covariate-adjusted, study-specific intervention effect estimates in the first step, which were subsequently combined in a random-effects meta-analysis model in the second step. BAIs produced a statistically significant 8% advantage in the mean number of drinks at both 1–3 months (RR = 0.92, 95% CI = [0.85, 0.98]) and 6 months (RR = 0.92, 95% CI = [0.85, 0.99]) compared to controls. At 9–12 months, there was no statistically significant difference in the mean number of drinks between BAIs and controls. In conclusion, BAIs are effective at reducing the mean number of drinks through at least 6 months post intervention. IPD can play a critical role in deriving findings that could not be obtained in original individual studies or standard aggregate data meta-analyses.

Which is Better for Individual Participant Data Meta-Analysis of Zero-inflated Count Outcomes, One-Step or Two-Step Analysis? A Simulation Study

“…Model 4 had a sizable bias and a low coverage for the logistic sub-model in the large effect size condition, especially when zero rate was small. Previous simulation research has shown poorer performance of logistic sub-models in the context of zero-inflated Poisson models (Zhou et al, 2022) and more generally in longitudinal analysis (Kim et al, 2020), although this observation warrants further investigation.…”

“…However, there are other approaches to zero-altered outcome data, including traditional zeroinflated models, which allow for zeroes in the count portion of the model (e.g., distinguishing between alcohol abstainers and drinkers who happened not to drink on a particular occasion) and newer marginalized zero-inflated approaches that produce a single set of treatment estimates, such as the marginalized zero-inflated Poisson regression model (Martin & Hall, 2017;Mun et al, 2022). Further investigation via real data analyses and simulation will be needed to assess the bias and coverage of other approaches to modeling zero-altered count outcomes in an IPD metaanalysis.…”

“…In the context of count outcomes, different sample populations can produce markedly different outcome distributions (see Mun et al, 2022), which has implications with respect to the choice of a one-or two-step analysis. For example, when an alcohol intervention study focuses on a broad population of individuals, the outcome distribution may have a disproportionate frequency of zeroes due to the presence of alcohol abstainers and occasional non-drinkers.…”

“…One-step models using multilevel modeling have greater flexibility to accommodate treatment effects that are missing by design, which we detail in the "One-step IPD Meta-analysis" section. In contrast, there can be data situations where a two-step meta-analysis may provide flexibility, for example, when underlying distributions differ by study (see Mun et al, 2022).…”

Does abstaining from alcohol in high school moderate intervention effects for college students? Implications for tiered intervention strategies

“…The outcome variable, the number of drinks participants consumed in a typical week, was a count variable with excessive zeros at both the 4-month and 15-month follow-ups. Because of the zero-inflation in the outcome variable, we utilized the Marginalized Zero-Inflated Poisson (MZIP; see Long et al, 2014 and Mun et al, 2022b , for detailed model specifications) model to evaluate our hypotheses. The MZIP model is a statistical approach for modeling zero-inflated count outcomes, based on the framework of the zero-inflated Poisson (ZIP) model, that can (a) account for excessive zeros and (b) estimate the effects of predictors on the overall mean for the entire distribution.…”