In this synthesis, we systematically review research on educational games with adaptivity. Although an adaptive gaming experience provides personalization to learning, the complexity of design makes it difficult to evaluate its effectiveness. In this systematic review, we adopt three analytic approaches: (1) bibliometric analysis, (2) qualitative thematic analysis, and (3) meta-analysis. We identified 62 relevant publications and used bibliometric analysis to visualize the hidden conceptual structure among the articles. We then used thematic analysis to investigate the emerging themes in the research and inform the coding for meta-analysis. Twelve articles that used experimental designs were further screened to model the effect of adding adaptivity to educational games. We found that an adaptive learning condition does not produce a substantial overall effect compared to a non-adaptive condition (g = 0.11, p = 0.26). Furthermore, moderator analysis reveals that the design variability of adaptivity does not contribute as much to heterogeneity as hypothesized. However, the effect size is positive when the target outcome focuses on learning (g = 0.39, p < 0.001) and engagement (g = 0.41, p = 0.13). The effect size is negative when the target outcome focuses on game performance (g = -0.27, p = 0.04). In addition, we find evidence for a potential publication bias based on the distribution of effect sizes.