Time-Varying Effect Models: Addressing New Questions in Prevention Science

Introduction. TVEM is a novel approach to model regression coefficients as continuous functions of time (Yang et al., 2012; Li et al., 2014). First introduced in statistics as varying-coefficient models (Hastie & Tibshirani, 1993), recent advances make TVEM accessible to prevention scientists to address important new questions. This poster introduces ways that prevention scientists can apply TVEM to examine dynamic processes. We summarize recent work using TVEM to estimate effects as continuous functions of (a) time since an event, such as quitting smoking (Lanza et al., 2013), and (b) developmental time (Vasilenko & Lanza, 2014), and then present two new applications of TVEM to estimate effects across (c) historical time and (d) age-of-onset of a particular behavior.

Method. The first study (historical time) uses data from Monitoring the Future, a repeated cross-sectional national sample of N=571,586 high school seniors from annual surveys conducted from 1976 to 2011 (49% male; 66% age 18 or older). Dynamic trends in past-month use and co-use of marijuana with recent use of alcohol and cigarettes are modeled as functions of historical time, with gender and race as moderators of dynamic trends. The second study (age of onset) estimates the rate of adult nicotine dependence as a function of age of first regular smoking. The sample is N=15,849 adults from the National Epidemiologic Survey on Alcohol and Related Conditions who have any history of regular smoking. Gender is examined as a moderator of this dynamic effect.

Results. The first study revealed that marijuana use rates peaked in the 1970s, declined through 1990, then rose again to reach rates of approximately 20% across race and gender groups. In recent years, the use of most substances had declined, but rates of marijuana use have been increasing, particularly among male and Black adolescents. The association between marijuana and cigarette use is increasing rapidly among Black adolescents. The second study revealed that for both males and females, the maximum risk of nicotine dependence coincides with first regular smoking at approximately age 11, with elevated risk persisting to age 20. The risk of adult nicotine dependence is significantly higher for females than males with regular smoking onset between ages 8-21.

Conclusion. TVEM holds great potential to advance prevention research by deepening our understanding about dynamic processes. Results could reveal new information about disparities in health behavior across age, the time-varying effect of policy changes, the optimal timing or age to intervene, and specific mediators to target at that time. We believe the field is poised for a significant shift toward examining time-varying effects.