Timing of a just-in-time intervention to reduce alcohol consumption: A simulation approach to optimize decision rules

The effectiveness of a just-in-time adaptive intervention relies on accurate algorithms (i.e., decision rules), that determine when and how interventions should be administered. Yet, so far, there is a lack of empirical investigations that evaluate the performance of decision rules. Simulation can be a useful tool to evaluate and refine a range of decision rules prior to implementing just-in-time adaptive interventions in real-world settings. In this study, we evaluate the performance of various decision rules using both an existing data set and a simulated data set that includes measures of craving and alcohol consumption. The tested decision rules consist of adaptive algorithms, like previous-day mean craving and online logistic regression, as well as fixed thresholds (e.g., a craving score larger than 1 on a 7-point Likert scale). For each decision rule, we generated confusion matrices and compared them across performance metrics, including accuracy, specificity, and sensitivity, as well as the number of interventions sent prior to drinking. To assess the robustness of our findings, we simulated a range of data sets with varying underlying distributions and tested the decision rule performance across these conditions. In addition, we conducted a multilevel logistic regression to identify the strongest association between the predictor and outcome variable across time lags. The presented method illustrates an approach to test and refine one's decision rules prior to launching a time-intensive, smartphone-based real-time intervention. A tutorial for conducting such simulations, as well as analysis codes, is provided online and in supplementary materials. (PsycInfo Database Record (c) 2026 APA, all rights reserved).