Identifying effective treatments for alcoholism and predictors of transitional or pattern drinking are important goals of the NIAAA. In clinical studies, the derivation of appropriate drinking outcomes is often subject to debate. Both primary drinking and secondary non drinking outcomes, such as mood and quality of life, are important outcomes via which to assess treatment effects. However, commonly used summaries of self-reported drinking do not provide information about the effect of treatment and time dependent comorbidities on daily drinking behavior. To appropriately model the evolution of drinking and non drinking outcomes in response to these variables, statistical methods for densely measured longitudinal responses should be developed for use in this setting. They should handle outcomes that are measured using various metrics and that are dependent on multiple time varying factors. They should also mitigate the effect of measurement error inherent in self-report, as drinking summaries are typically reported using a calendar based method of recall. Finally, methods should lend themselves to simultaneous joint modeling of drinking and secondary nondrinking outcomes. Current statistical methods do not address all of this under a unified framework. Goals: Using a Bayesian paradigm, the proposed study will develop robust statistical methods addressing all of the above challenges for assessing treatment effectiveness on drinking and non drinking behavior, and for assessing the relevance of time dependent covariates on the evolution of drinking. Software will be developed and disseminated freely. Subjects: The statistical methods will be evaluated on two datasets, the Combined Pharmacotherapies and Behavioral Interventions for Alcohol Dependence (COMBINE) clinical trial (n=1383) and a prospective comorbidity study (n=663) assessing major depressive disorder on the course of alcohol and other substance dependence. Available data and study design: Daily or weekly responses such as drinking, other substance use, mood, depressive symptoms, craving, stress, and quality of life were collected and will be evaluated as outcomes. Treatment status, demographics, baseline measures, and prior psychiatric/health disorders will serve as baseline variables, and relevant medical status, adverse events, and onset of psychiatric disorders such as major depressive disorder that were measured throughout the studies will serve as predictors in the models. Both studies have an over representation of minority participants and include women in high proportions. Significance: The new statistical methods will provide alcohol researchers with a rich description of the behavioral evolution of primary drinking and secondary nondrinking outcomes in response to treatment and time dependent comorbidities throughout the course of these studies.
This project proposes robust transition models to address concerns not previously investigated through routine summary statistics or simple longitudinal models for studying prospective alcohol outcomes. The proposed research has the potential to exert tremendous impact from a public health perspective through efficient use of the data. The long term goal is to provide alcohol researchers a better understanding of the stochastic mechanism of alcohol dependence, so as to not only inform better study designs, but potentially improve prevention and control strategies.
|Desantis, Stacia M; Lazaridis, Christos; Ji, Shuang et al. (2014) Analyzing Propensity Matched Zero-Inflated Count Outcomes in Observational Studies. J Appl Stat 41:|