The goal of this randomized controlled trial is to evaluate the effectiveness of the Safe Haven app in reducing mental health symptoms and enhancing mental health literacy in undergraduates at a national university in southern Taiwan. The main questions it aims to answer are:

Does the Safe Haven app lead to significantly greater reductions in general psychological distress, depression, anxiety, and stress compared to standard campus care?

Does the app result in greater improvements in mental health literacy among participants?

Researchers will compare an experimental group (receiving the Safe Haven app in addition to standard campus care) to a waitlist control group (receiving standard campus care only) to see if the intervention provides superior improvements in mental health outcomes over time.

Participants will:

Complete online assessments at three time points: baseline (T0), post-intervention (3 months, T1), and follow-up (3 months post-intervention, T2).

Experimental Group: Download the Safe Haven app to complete daily mood self-monitoring and weekly assessments, and receive psychoeducational content, personalized suggestions for 3 months.

Control Group: Maintain access to standard campus care and receive access to the app after the final follow-up assessment.

Background: Undergraduates face a critical transition period marked by significant stressors and a high prevalence of mental health issues. However, help-seeking behaviors remain low due to stigma and structural barriers. While mobile health (mHealth) apps offer a scalable solution, existing interventions often rely on static content and lack real-time responsiveness to students' fluctuating psychological states. This study introduces "Safe Haven," a Just-in-Time Adaptive Intervention (JITAI) system that integrates self-monitoring with risk stratification to provide tailored support. This study aims to evaluate the effectiveness of the Safe Haven app in reducing psychological distress and enhancing mental health literacy compared to standard campus care alone.

Study Design: This study employs a parallel-group, two-arm randomized controlled trial (RCT). A total of 68 eligible undergraduate students from a national university in southern Taiwan will be recruited. After completing baseline assessments (T0), participants will be randomized in a 1:1 ratio to either the Experimental Group or the Waitlist Control Group.

Intervention Details

Experimental Group (Safe Haven App + TAU): Participants in this group will receive access to the Safe Haven app for a duration of 3 months, in addition to maintaining full access to standard on-campus counseling and mental health services (Treatment-as-Usual, TAU). The Safe Haven app includes five core components:

1. Mood and Sleep Tracking Visualizes longitudinal trends in daily mood and sleep reports and weekly psychological assessment scores to support self-monitoring and self-awareness.
2. Digital Phenotyping Data Collection Collects active self-report data and optional passive mobility data to support real-time monitoring and mental health risk prediction.
3. Risk-Stratified Stepped-Care System Uses machine learning to classify users into five mental health risk levels, each linked to a tiered intervention protocol ranging from self-management support to emergency response.
4. chatbot-Based Just-in-Time Support Provides real-time emotional support via an AI chatbot while monitoring user inputs for high-risk indicators and triggering emergency protocols when detected.
5. Psychoeducation Modules Delivers animated psychoeducational content with brief quizzes and a gamified reward system, alongside information on campus and community mental health resources.

Waitlist Control Group (TAU Only): Participants in this group will maintain access to standard on-campus counseling and mental health services (TAU) but will not access the Safe Haven app during the 3-month intervention period. To ensure ethical standards, these participants will be offered access to the Safe Haven app after completing the final follow-up assessment (T2).

Study Procedures Data collection will occur at three time points:

1. Baseline (T0): Upon recruitment and prior to randomization.
2. Post-Intervention (T1): Immediately following the 3-month intervention period. Experimental group participants will also complete the System Usability Scale (SUS) at this time.
3. Follow-Up (T2): 3 months after the intervention concludes.

Outcome Measures The primary outcomes are changes in general psychological distress (measured by CHQ-12) and symptoms of depression, anxiety, and stress (measured by DASS-21). Secondary outcomes include changes in mental health literacy (measured by MHLS-HPG) and intervention engagement (measured by app usage logs and SUS scores).

Statistical Analysis Plan Data will be analyzed using IBM SPSS Statistics. The primary analysis will follow the Intention-to-Treat (ITT) principle. Generalized Estimating Equations (GEE) will be utilized to examine longitudinal changes and test the Group × Time interaction effect on outcome measures. GEE models will be adjusted for baseline scores of the respective outcome variables. Additionally, any demographic characteristics showing statistically significant differences between groups at baseline will be included as covariates. For the experimental group, further analyses will examine the association between objective engagement metrics (e.g., completion rates) and clinical improvements to determine dose-response relationships.

Inclusion Criteria:

• registered undergraduates aged 18-24 years
• own a personal smart device (mobile phone or tablet)
• ability to read Chinese
• willingness to install the study app and complete the follow-up assessments.

Exclusion Criteria:

• are on leave of absence or not officially enrolled during the study period
• are participating in another mental health intervention program or using similar mental health apps.

Due to the nature of the intervention (a mobile health application), it is not feasible to blind participants or the research team to the group allocation. However, to minimize bias, outcome data will be collected via online self-report questionnaires (Google Forms) to reduce assessor bias. Furthermore, data analysis will be conducted using de-identified datasets where group allocation is coded to ensure objectivity during statistical processing.