Time-varying effect modeling with intensive longitudinal data: Examining dynamic links among craving, affect, self-efficacy and substance use during addiction recovery

Abstract

Time-varying effect modeling (TVEM), a statistical technique for modeling dynamic patterns of change, presents new opportunities to study biobehavioral health processes. TVEM is particularly useful when applied to intensive longitudinal data (ILD) because it permits highly flexible modeling of outcomes over continuous time, as well as of associations between variables and moderation effects. TVEM coupled with ILD is ideal for the study of addiction. This article provides a general overview of using TVEM, particularly when applied to ILD, to better enable addiction scientists to conduct novel analyses that are important to realizing the dynamics of addiction-related processes. It presents an empirical example using ecological momentary assessment data from participants throughout their first 90 days of addiction recovery to estimate the (1) associations between morning craving and same-day recovery outcomes, (2) association between morning positive and negative affect and same-day recovery outcomes and (3) time-varying moderation effects of affect on the association between morning craving and recovery outcomes. We provide a didactic overview in implementing and interpreting the aims and results, including equations, computer syntax and reference resources. Our results highlight how affect operates as both a time-varying risk and protective factor on recovery outcomes, particularly when considered in combination with experiences of craving (i.e. dynamic moderation). We conclude by discussing our results, recent innovations and future directions of TVEM for advancing addiction science, including how 'time' can be operationalized to probe new research questions.

Keywords: Addiction recovery; affect; craving; ecological momentary assessment; intensive longitudinal data; substance use disorders; time-varying effect modeling.

FIGURE 1

Estimated (a) prevalence of substance use and (b) mean level of self-efficacy across the first 90 days of recovery.

FIGURE 2

Odds ratio corresponding to change in odds of substance use associated with morning experience of craving across days (a) and the estimated difference in mean self-efficacy associated with morning experience of craving across days (b). Gray shading denotes regions of statistical significance.

FIGURE 3

Odds ratio corresponding to change in odds of substance use associated with morning negative affect (NA) across days (a) and the estimated difference in mean self-efficacy associated with morning NA across days (b). Gray shading denotes regions of statistical significance.

FIGURE 4

Odds ratio corresponding to change in odds of substance use associated with morning positive affect (PA) across days (a) and the estimated difference in mean self-efficacy associated with morning PA across days (b). Gray shading denotes regions of statistical significance.

FIGURE 5

Coefficient for the interaction between morning craving and negative affect (NA) predicting substance use (a) and self-efficacy (b). Gray shading indicates days on which statistically significant interaction effects were observed.

FIGURE 6

Coefficient for the interaction between morning craving and positive affect (PA) predicting substance use (a) and self-efficacy (b). Gray shading indicates days on which statistically significant interaction effects were observed.

FIGURE 7

Estimated prevalence of substance use (a) and mean self-efficacy (b) across levels of morning craving crossed with morning negative affect (NA). Gray shading indicates which days statistically significant interaction effects were observed.

FIGURE 8

Estimated prevalence of substance use (a) and mean self-efficacy (b) across levels of morning craving crossed with morning positive affect (PA). Gray shading indicates which days statistically significant interaction effects were observed.