Digital Phenotyping & Deep Learning: Substance Use Impact on PrEP Adherence

National Institute On Drug Abuse

Investigators

Linked publications & trials

Abstract

This project investigates the intersection of cannabis use, HIV transmission risk, and engagement in preventive sexual health behaviors. The purpose of the study is to understand how substance use patterns, digital communication, and everyday behavioral contexts influence sexual decision-making and adherence to HIV prevention strategies. The research focuses on adults who use cannabis and engage in sexual activity, including individuals at elevated risk for HIV acquisition. Participants were enrolled in a longitudinal digital phenotyping study that integrates survey data, passive mobile sensing, and analysis of online dialogue related to cannabis and sexual behavior. During this fiscal year, we refined study protocols, completed participant recruitment for the initial longitudinal cohort, and collected baseline survey data on cannabis use, sexual behavior, relationship contexts, and preventive health practices. We deployed a validated mobile health application developed by our laboratory to gather passive smartphone sensor data and deliver ecological momentary assessments. We also conducted an analytical study of online discussions of sexualized cannabis use using natural language processing and mixed-methods text analysis. Data collection infrastructure, including secure servers and computational resources for machine learning and language analysis, was fully implemented and tested. Significant materials and equipment used during this period included smartphones for data collection, wearable-compatible sensing modules, and GPU-enabled computational systems to support deep learning and natural language processing models. Methods used in the project include ecological momentary assessment, longitudinal survey assessment, passive sensing, event-level sexual behavior measurement, topic modeling, lexicon-based analyses, and deep learning models for risk prediction. Accomplishments this year include the deployment and stabilization of the mobile application, completion of baseline data collection for the longitudinal cohort, development of preliminary machine learning pipelines for detecting cannabis use patterns, and identification of key behavioral and contextual factors linked to sexualized cannabis use through online discourse analysis. These findings support the development of dynamic digital phenotypes that characterize real-time behavioral risk patterns. The overall objective of the project is to determine how cannabis use contributes to engagement in sexual behaviors that elevate HIV risk and decreased adherence to recommended preventive practices, and to lay the groundwork for non-drug, technology-enabled interventions that support safer sexual decision-making and improved health outcomes.