Digital Markers in Relapse and Recovery

National Institute On Drug Abuse

Investigators

Linked publications, trials & patents

Abstract

The literature on relapse continues to be rife with methodological inconsistencies, exhibiting broad variations in the definition of relapse, assessment methodologies, and models addressing relapse-related factors. Traditional methodologies to gather data on relapse encompass: 1) retrospective reviews prompting participants to recollect instances of relapse and preceding factors; 2) prospective reports collecting potential antecedent information at baseline or at intervals and later assessing the association with detected relapses; and 3) near real-time reports where participants report or are prompted electronically on factors proximate to the actual relapse event. A notable gap exists in research utilizing behavioral observation in daily life due to the challenges posed in data collection. Recent advancements reaffirm that near real-time reports offer an optimal strategy, as relapse vulnerability factors like self-efficacy, drug cues, anxiety, stress, drug craving, and social support can evolve within mere minutes. Recognizing these challenges, our project has leveraged real-time reports to detect and predict relapse in individuals attending substance use treatment programs and those in recovery. As public health research and practice begin to harness the transformative changes in communication media, our project has been at the forefront, employing innovative tools to scrutinize social media language and data generated from digital devices, including smartphones and wearables. This year, we made significant strides by adapting advanced data analytics to delve deeper into the digital imprints of those in substance use treatment and long-term recovery. Through natural language processing and machine learning, we've crafted predictive models that forecast relapse and long-term recovery, enhancing our capabilities through passive measurement techniques that offer detailed behavioral data, often surpassing traditional methods. In tandem with these efforts, an instrumental publication that stood out this year is the work in our lab by Giorgi et al., titled "Predicting U.S. county opioid poisoning mortality from multi-modal social media and psychological self-report data." This research addressed the severe opioid poisoning mortality crisis in the U.S. by employing a multi-modal dataset, including Twitter language and psychometric self-reports. Highlighting the utility of social media data, the study showed that Twitter language was more indicative of opioid poisoning mortality than traditional factors like socio-demographics or healthcare access. This illuminates the potential of harnessing natural language from social media as a pivotal surveillance tool in predicting community opioid poisonings and deepening our understanding of the epidemic's multifaceted nature. Most relapse prevention strategies have historically been restricted, relying on a sliver of available data typically harvested through surveys and interviews. By contrast, our approach utilizes a wealth of data. Instead of focusing solely on the most recent measurement for assessing relapse risk, we harness the dynamic nature of relapse vulnerability factors, combining insights from various recovery journeys to refine our predictions. The culmination of this endeavor is the development of dynamic, real-time predictions that stay attuned to swift alterations in relapse risk. Our lab's ambitious long-term goal is crystallizing: we are pioneering an automated, ceaseless system that monitors a spectrum of digital sources from social media language to smartphone sensor data and wearable device outputs to predict daily relapse vulnerability scores. Building on this, we have charted the course to develop a feedback tool for relapse vulnerability, aiming to serve a wide audience, including addiction treatment providers, those under treatment, and individuals in recovery. This trailblazing initiative promises a paradigm shift in clinical research and practice, laying the foundation for automated interventions targeting patients at heightened risk. A few other notable articles include an AI-based analysis Curtis B et al., 2023 that underscored the predictive capabilities of social media language in determining addiction treatment dropouts. Furthermore, our explorations into the role of media Habib DRS et al., 2023 and the intersection of linguistic methodologies with public health Lane JM et al., 2023 have added novel dimensions to our research narrative.