Optimizing Just-in-Time Adaptive Interventions for Interpersonal Distress: Mechanisms, Prediction, and the Challenge of Engagement

Background: Common mental health disorders (CMD) feature fluctuating emotional and interpersonal symptoms inadequately addressed by traditional weekly therapies. Ecological momentary interventions (EMI) offer potential for timely support, yet their mechanisms and optimal delivery contexts remain unclear. Aim: To examine dynamic symptom networks, proximal effectiveness, engagement predictors, and distress forecasting in adults with CMD. Methods: This secondary analysis of a randomized trial (N=77) compared mindfulness and mentalization micro-interventions triggered by personalized symptom thresholds. EMA data were collected 4x/daily for 28 days. We utilized Dynamic Exploratory Graph Analysis, generalized linear mixed models (GLMM) for proximal effects, and mixed-effects logistic regression for engagement and prediction. Results: Dynamic networks revealed stable communities (interpersonal threat, social connection, affective states) with mood as a key bridge. No significant proximal intervention effects were observed. Non-engagement was significantly predicted by high stress (OR = 1.21), elevated mood (OR = 1.22), and perceived criticism (OR = 1.22). Conversely, cumulative symptom triggers (OR = 0.69) and social contact (OR = 0.83) facilitated engagement. The dynamic prediction model achieved fair performance (AUC = 0.66) for next-beep distress. Beyond autoregressive effects, perceived support paradoxically predicted future distress (OR = 1.14), while warmth was protective (OR = 0.87). Conclusion: Micro-interventions operate through stable networks but yield cumulative rather than immediate benefits. That high stress and criticism impede intervention use despite high need highlights the necessity for context-sensitive, low-friction adaptive designs to align clinical need with receptivity.