Anxiety and depression are the most common mental health conditions worldwide, yet established treatments face limited reach, stigma, cost, and poor adherence. Video games, purpose-built or repurposed from entertainment, have emerged as scalable, engaging complementary interventions. Because the literature describes each game by the therapeutic mechanism it was built to deliver, no common vocabulary exists for comparing how games are constructed. Across three empirical chapters, this thesis characterizes the design architecture of video games used as interventions for anxiety and depression and relates that architecture to what these interventions achieve.
Chapter 2 presents the published protocol of a multi-center, assessor-blind, three-arm randomized controlled trial in adolescents aged 12 to 14 with elevated anxiety (Gradi et al., 2024). It compares six weeks of online play with Eco-Rescue, an action-like game designed to train attentional control, with a casual puzzle game chosen as a positive distraction and with no training; anxiety is the primary outcome. Across three sites, 55 adolescents have been randomized, dropout is below 15%, and completion is expected by the end of 2027.
Chapter 3 reports a PRISMA-guided systematic review of 39 randomized controlled trials evaluating 25 games across the lifespan. Each game was coded with a reduced Laine and Lindberg (2020) taxonomy (47 design principles in 11 classes) and mapped to the four engagement types of Plass et al. (2015). Feedback and Goals were the most adopted classes, affective and cognitive engagement predominated, and each of the four intervention types (cognitive behavioral therapy, cognitive training, casual, physical) showed a distinct design and engagement profile. Hierarchical clustering on Jaccard distance returned four design profiles (Player Control, Narrative, Simple interaction, Player agency), only the first of which came from a single intervention type: design appears to be a property that intervention type does not fully capture.
Chapter 4 synthesizes 31 effect sizes from 19 randomized trials in two parallel meta-analyses, using a three-level correlated-and-hierarchical-effects model with robust variance estimation and intervention type or design profile as the focal moderator. Game-based interventions produced a small, significant symptom reduction relative to controls (g = 0.25, 95% CI [0.14, 0.37]), with modest heterogeneity and no sign of publication bias. Neither moderator was significant at the available power. Within intervention types, only Physical games showed a significant effect (g = 0.41, 95% CI [0.09, 0.73]); within design profiles, the Narrative estimate was small but significant (g = 0.26, 95% CI [0.03, 0.49]) and the larger Player Control estimate rested on too few studies to interpret. This likely reflects the few available trials rather than an inert mechanism or design.
The General Discussion argues that the design architecture of mental health games is a distinct and consequential layer of their therapeutic potential. The thesis contributes a common design vocabulary and two auditable procedures (design-principle coding, AI-assisted data extraction) and calls for structured reporting of design choices so that cumulative evidence can show which principles best serve which populations.
