Innovative Approaches to Substance Use Measurement: Questionnaires, Screening Tools, and Smartwatch Algorithms

Abstract

Alcohol and cannabis are two of the most widely used substances in the United States and are associated with significant public health costs. Alcohol and cannabis-related health consequences have also been increasing over the past 10 to 15 years. However, despite the increased prevalence, our ability to assess and treat individuals has stagnated. This is likely due to three main reasons: 1) a limited ability to screen individuals at risk, especially in relation to co-morbid conditions like chronic pain, 2) the increased variety of available substance products (i.e., cannabis flower vs concentrates), dosages, and ways to divide such products (i.e., drinks, grams, ounces), and 3) significant barriers to accessibility of substance-related treatment. The goals of the dissertation were to develop and validate innovative methodologies to address the identified gaps in the literature. First, we developed a brief clinical screening tool to assess risk of self-medication of pain with substance use. We used an iterative process to reduce the total number of items from 104 to a 14-item and 7-item version. We found that both versions were significantly associated with self-medication behavior and substance use health risks, with minimal differences between the two versions. Second, we modified an existing behavioral economics measure to adapt to a user’s specified preferred product (e.g., dried flower) and division method (e.g., hits or grams). The modified behavioral economics measure demonstrated convergent validity and strong criterion validity compared to the original. In addition, we demonstrated significant differences associated with user preferences, such that individuals with a preference for dried flowers showed greater demand than those with a preference for concentrations. Third, we developed a novel smartwatch algorithm in an accessible framework to detect and analyzing drinking behavior. This technology provides the technological foundation for the development of accessible just-in-time alcohol use interventions. We compared the performance of a distribution-based algorithm to a traditional machine learning model in a controlled paced drinking environment. Our distributional sip detection algorithm performed similarly to our ground truth on our behavioral outcomes (i.e., sip duration, sip interval, and number of sips). It also performed significantly better than a random forest machine learning classification model. Lastly, we validated our automatic sip detection algorithm using a simulated virtual reality bar environment to allow unrestricted drinking with validated ground truth measures. We found that the algorithm had moderate-to-high classification with our ground truth measures. In conclusion, the results from these studies provide important methodological tools designed to increase the field’s ability to develop innovative and accessible methodologies. This work will also provide the support to enhance the assessment of substance use mechanisms by improving our ability to screen, adapt and develop accessible frameworks for interventions.