Characterizing Patterns of Engagement with a Digital Health Intervention to Reduce Young Adult Cannabis Use

Background: Text message–delivered interventions offer a scalable, low-burden strategy for addressing cannabis use disorder (CUD) among young adults, a population with high prevalence of use and low rates of treatment engagement. From an implementation science perspective, mobile health (mHealth) interventions have strong potential to enhance reach and equity by reducing barriers related to stigma, access, and cost. However, engagement with digital interventions is often heterogeneous, posing challenges for optimizing effectiveness and sustainment. Examining latent patterns of engagement can advance implementation research by identifying meaningful user subgroups, informing tailored strategies, and strengthening key outcomes such as reach, engagement, and sustained use in real-world settings.

Method: We examine patterns of engagement with Peer Network Counseling-text (PNC-txt), a text message-delivered intervention designed to help young adults reduce their cannabis use. We conducted a secondary analysis of data from a large, NIH-funded randomized clinical trial of PNC-txt compared to waitlist control, that enrolled over 1000 young adults (18-25 years) from Colorado and Tennessee who met criteria for CUD. Mixture modeling was used to identify latent classes of user engagement. Two continuous indicators of engagement – self-reported treatment satisfaction and practice of intervention skills – were standardized before being included in the model. Binary treatment adherence (responded to all text messages or not) and requests for supplemental material (low, medium, and high levels based on quartiles) were also included.

Results: In preliminary analyses, a three-class solution fit the data best, with decreasing AIC and BIC values and increasing entropy as compared to two-class and four-class solutions. The largest group comprised 71% of participants (N=380) in the treatment condition, followed by 20% (N=108) in the second largest group, and 9% in the final group (N=47). The smallest group were the Disengaged. These were most likely not to finish the intervention nor request booster texts; they reported low satisfaction with the treatment and infrequent practice of the skills taught. Most participants fell into the largest group, the Average Engagers. These participants were generally likely to complete the full intervention and may have requested a booster message at one point. They reported average levels of satisfaction and skills practice, falling closest to the mean of the groups. The final group, High Engagers, reported high levels of satisfaction and skills practice, they completed the full intervention and requested multiple booster messages. Ongoing analyses are examining characteristics of each class (demographics, readiness to change and self-efficacy, depression and anxiety symptoms) and discuss implications and future analysis plans.

Conclusions and implications: Identifying patterns of engagement with digital health interventions can assist in tailoring strategies for different groups to improve likely uptake, engagement, and use over time. Analyses linking these patterns with treatment outcomes can examine whether increased engagement leads to better outcomes and to develop targeted outreach to those most likely to benefit.