Enhancing Data Integrity and Participant Safety In Digital Health Intervention Research

Background: Digital health interventions offer scalable approaches to address mental health and substance use disorders, but remote data collection introduces methodological and ethical challenges. Online research studies face increasing concerns about fraudulent participation, including bots, duplicate responses, and misrepresentation of eligibility. Recent evidence suggests commonly used validation checks fail to detect a substantial proportion of fraudulent respondents. At the same time, digital intervention studies must address participant safety, including monitoring suicidality, crisis risk, and clinical worsening in real time despite limited face-to-face contact. These interventions often rely on asynchronous communication and automated systems, complicating timely identification of risk and clinical response. Prior research highlights the need for structured monitoring and safety protocols, particularly for populations at elevated risk for suicide and severe mental health symptoms. Despite these challenges, there is limited guidance on integrated approaches to ensuring both data integrity and participant safety in digital intervention research.

Methods: We present procedures from multiple digital health intervention trials, including text-delivered cognitive behavioral therapy and Peer Network Health with young adults. We describe a multi-layered system to prevent fraudulent enrollment and monitor participant safety. Fraud prevention strategies included platform-based safeguards (e.g., CAPTCHA, bot and duplicate detection), IP and geolocation screening, VPN/proxy detection, and external verification tools (e.g., Twilio Lookup/Verify) to confirm mobile phone validity and assess fraud risk. Additional screening incorporated metadata (e.g., ISP, proxy status), attention checks, and eligibility logic to identify suspicious or ineligible respondents. Eligible participants completed real-time identity and residence verification via Zoom using government-issued identification.

Participant safety was supported through individualized crisis planning, automated and clinician-monitored assessments, ecological momentary check-ins, and continuous monitoring of incoming text messages using automated keyword detection and human review. A structured, tiered crisis response protocol guided responses based on risk level.

Results: Across pilot studies, integrated monitoring systems successfully identified both fraudulent activity and participant risk. Automated screening of 31,740 incoming text messages detected all confirmed crisis events (sensitivity = 100%; specificity = 99.9%), with no missed crises. Fraud prevention procedures identified multiple indicators of ineligible or suspicious participation, including bot-like response patterns, proxy/VPN use, duplicate entries, and high-risk phone numbers, improving overall data quality. The integration of automated tools with clinician oversight enabled timely identification of risk and appropriate intervention.

Implications: These findings underscore the importance of prioritizing both data quality and participant safety in digital intervention research. In practice, this means designing studies where verification and monitoring procedures are embedded from the outset, rather than relying on data cleaning. Our approach shows how combining automated tools with manual review can improve data quality while ensuring timely identification of participant risk. Integrating automated systems with clinician oversight offers a practical model for balancing efficiency with ethical responsibility, particularly in fully remote or higher-risk populations. As digital interventions expand, these systems will be increasingly important for ensuring research remains rigorous and responsive to participant needs.