Abstract: Combining Latent Class Analysis and Time-Varying Effect Modeling: Understanding the Epidemiology of Alcohol Use (Society for Prevention Research 27th Annual Meeting)

561 Combining Latent Class Analysis and Time-Varying Effect Modeling: Understanding the Epidemiology of Alcohol Use

Friday, May 31, 2019
Regency B (Hyatt Regency San Francisco)
* noted as presenting author
Bethany Bray, PhD, Associate Research Professor, The Pennsylvania State University, University Park, PA
John J. Dziak, PhD, Research Assistant Professor, The Pennsylvania State University, University Park, PA
Stephanie T. Lanza, PhD, Director, Edna Bennett Pierce Prevention Research Center; Professor, Biobehavioral Health, The Pennsylvania State University, University Park, PA
Introduction. Alcohol use in the U.S. remains a critical public health issue. Although underage and college student drinking receive considerable attention, binge drinking and heavy alcohol use are risky at any age. Studying the epidemiology of alcohol use from adolescence through adulthood requires methods that (1) reflect the complex, multidimensional nature of drinking and (2) allow for very different prevalence rates of behaviors and very different effects of risk factors across wide age ranges. Existing studies on age trends have focused on either a single age period with many behaviors, using methods such as latent class analysis (LCA), or on a single alcohol use behavior across many ages, using methods such as time-varying effect modeling (TVEM). This study presents a new, integrated model combining LCA and TVEM where probabilities of class memberships can depend on rich interactions between covariates and age. This study introduces LCA-TVEM in the context of drinking behavior across ages 18-65.

Method. Data were from the National Epidemiologic Survey on Alcohol and Related Conditions-III (n=30,997; 51.1% women; 63.5% White Non-Hispanic, 12.5% Black Non-Hispanic, 16.2% Hispanic, 7.8% other race/ethnicity). First, four classes of drinking behavior were identified using LCA across the full age range: (1) Non-Drinkers, (2) Frequent Light Drinkers, (3) Infrequent Binge Drinkers, and (4) Extreme Drinkers who engaged in high-intensity drinking (10+/8+ drinks for men/women). Second, we examined how the prevalences of these classes depended non-linearly on age using an intercept-only TVEM. Third, we explored whether the effects of gender on class membership varied non-linearly across age using TVEM.

Results. Results from a simulation study are briefly discussed to show that our proposed approach to LCA-TVEM provides accurate parameter estimation under the assumed model, given sufficient sample size. Results from our empirical study show the expected age trends in class membership during adolescence and emerging adulthood for Non-Drinkers and Extreme Drinkers (i.e., class prevalences hit their lowest and highest rates, respectively, around age 22), but relatively consistent rates of Infrequent Binge Drinking through adulthood. Results also show that the effect of gender on latent class membership is most variable for Extreme Drinkers vs. Non-Drinkers across ages.

Discussion. LCA-TVEM integrates the strengths of each approach, reflecting both the complexities of behavior and long-term age trends in prevalences and effects. LCA with flexible, age-varying class prevalences and covariate effects holds great promise to advance research on complex addictive behavior patterns in the population, including age-specific risk factors. This approach could be applied to a variety of complex addictive behaviors to shed new light on their age-varying prevalences and correlates.