This project will examine new methods for understanding time-varying causal mediation effects in smoking cessation interventions. Mediation occurs when an independent variable (e.g., smoking cessation intervention) has an effect on an outcome (e.g., smoking abstinence) through a third variable (e.g., reduction in craving). In other words, we will develop new ways to understand the mechanisms by which smoking cessation interventions cause changes in people's smoking behavior, and how these effects vary over time. Understanding how interventions have their effect is key to designing more powerful, efficient interventions. To improve smoking cessation interventions, this proposal focuses on improving methods for the understanding of how people stop smoking and continue to abstain from smoking following a smoking cessation intervention. More broadly, the methods that we propose to develop will help us understand what causes behavioral interventions to affect people's engagement in and maintenance of healthy behaviors. We will develop and extend methods for analyzing intensive longitudinal data and conducting mediation analysis to better study time-varying effects, which refer to effects of one variable on another that change (i.e., strengthen or diminish) over time. We will extend methods, based on the potential outcomes framework for causal inference, to draw more valid causal inferences about time-varying effects in mediation and then apply these methods to two smoking cessation intervention studies. First, we will use ecological momentary assessment (EMA) data that asked people about their craving, mood, and smoking abstinence while they were enrolled in a randomized controlled trial of combination nicotine replacement therapy, the nicotine patch alone, and varenicline to develop mediation models in which the effect of the various study drugs on smoking abstinence is mediated by cravings or mood. These effects may vary over time. For example, a study drug may be very effective at reducing cravings shortly after the quit day, but the effect may weaken over time. Second, using data from pedometers, we will assess the effect of a physical activity/ smoking cessation intervention on daily step count and whether increases in daily step counts in turn have an effect on smoking cessation. We will disseminate the methods we develop in user-friendly software (e.g., R packages), journal articles, and conference presentations and workshops that are accessible to behavioral intervention researchers. Methods and findings will ultimately allow researchers to understand how their interventions work. With this knowledge, they will be able to design more efficacious and cost-effective interventions for smoking cessation by targeting the most relevant mediator(s) at the most relevant time.
This project will (1) generate new statistical methods and software to allow us to better understand what causes interventions to work, and how these effects may vary over time, and (2) apply these methods in two smoking cessation interventions studies using intensive longitudinal data. The methods we develop will have wide-ranging applications for many other studies involving medical or public health interventions, and we will make our methods available so that other researchers may use them in their research. Methods and findings will ultimately help researchers design more efficacious and cost-effective interventions that help people change and maintain healthy behaviors.
| Mennis, Jeremy; Mason, Michael; Coffman, Donna L et al. (2018) Geographic Imputation of Missing Activity Space Data from Ecological Momentary Assessment (EMA) GPS Positions. Int J Environ Res Public Health 15: |
