Optimizing Naturalistic Engagement with Digital Mental Health Among LGBTQ+ Users: Comparing Design and Behavioral Targeting In a Micro-Randomized Trial

Background and Purpose: Sexual and gender minority (SGM) populations experience significant mental health disparities, yet face persistent barriers to care. Mental Health America (MHA) provides free, web-based screening and self-guided resources to thousands of SGM visitors annually. Digital mental health (DMH) can facilitate access to mental health resources; however, engagement remains a central challenge that limits effectiveness. On MHA, most visitors exit the website without accessing substantive content. Engagement strategies, delivered as brief prompts or messages that direct users toward content, may support engagement at moments when it is most likely to decline. The Health Action Process Approach (HAPA) frames engagement as shaped by modifiable determinants, including outcome expectancy, perceived risk, self-efficacy, and barriers to engagement. It remains unclear, however, whether what engagement strategies target (i.e., the behavioral determinant) or how they do so (i.e., the design features) more strongly drives engagement. The QT-Digital Mental Health Engagement (QT-DIME) Study examined this question on MHA’s real-world DMH platform.

Methods: A micro-randomized trial (MRT) was embedded within the MHA Online Screening Program. Eligible participants were US-based MHA visitors aged ≥14 who completed the PHQ-9 in English and identified as LGBTQ+. Participants were randomized at prespecified decision points representing empirically identified moments of disengagement risk. At each decision point, participants were randomized with equal probability to receive no engagement strategy or one targeting one of four HAPA behavioral determinants (outcome expectancy, perceived risk, self-efficacy, SGM-specific barriers to engagement). Three message variations per determinant (12 total) were delivered at the first decision point with each incorporating distinct design features (e.g., authority cues, social proof, goal-gradient effects). The MRT’s primary outcome was proximal click-through to substantive content pages analyzed via marginal causal excursion effects. The current analysis focuses on message-level variation at the first decision point to characterize the relative contribution of behavioral targets and design features.

Results: Among 2,439 randomized visitors, engagement strategies pooled by HAPA construct performed equivalently to control (22.2% vs. 23.5% click-through to substantive content pages). However, this aggregate pattern obscured heterogeneity within each HAPA construct, with click-through ranging from 18.2% to 28.3% across messages. The four HAPA constructs performed similarly on average, but messages within each construct diverged (e.g., barriers messages: 19.1-28.3%). Top-performing messages (≥24.7%) shared common features, including granting MHA visitors permission to engage on easier terms (e.g., framing help-seeking as strength, engagement as exploration rather than commitment, removing pressure that invoked urgency). In contrast, bottom-performing messages (≤19.1%) uniformly relied on expert or institutional authority.

Conclusion and Implications: In real-world DMH contexts, selecting theoretically-informed targets alone may be insufficient to optimize engagement without attention to design features. The lower performance of authority-based messaging may indicate that this framing is less effective at supporting initial engagement when users may be more responsive to low-barrier, supportive framing. Results will guide iterative refinement of displayed strategies over the study period (e.g., retiring underperforming messages and introducing variants targeting HAPA constructs through more promising design features) and inform scalable implementation of engagement strategies within the broader MHA website and similar DMH platforms.