Substance use disorders (SUDs) are highly prevalent and lethal, yet treatment reach remains below 20%. As risk of substance use and relapse is episodic and context-dependent, ecological momentary interventions (EMIs) that deliver real-time intervention in daily life are promising, but findings to date remain mixed. We argue this variability reflects the importance of decision rules, when to deliver which intervention. However, current EMI systems mostly rely on static, one-size-fits-all rules that could not account for between-person differences and within-person fluctuations. We suggest a data-driven approach for building EMI systems, aiming to better address the heterogeneity of SUDs. First, collect multimodal, multicontextual data—spanning controlled laboratory tasks, everyday smartphone and wearable signals, and periods when devices are offline—to complement blind spots of individual data sources. Next, build context−aware prediction models that estimate momentary risk and validate predictors across contexts and modalities, enabling features discovered in one setting to be translated into signals available in another. Finally, implement real−time, context−sensitive decision rules that best fit the contextual profile of the risk. By centering EMIs on explicit, testable decision rules, this approach will offer a practical path to reducing variability in outcomes and deliver more reliable, personalized support at the moments and places where risk emerges.
Kwon et al. (Mon,) studied this question.