Digital Health Data Quality & Monitoring

Overview

Digital health data from wearables and sensors is only valuable if participants wear devices as instructed and data flows correctly through the system. Without ongoing monitoring, problems often surface too late—after participants have dropped out, after data windows have closed, after the opportunity for intervention has passed.

This project established systematic data quality monitoring for digital health studies, providing study teams with timely visibility into adherence, completeness, and integrity metrics.

The Problem

DHT-derived endpoints depend on adequate data quality. Common issues include:

• Non-wear periods: Participants removing devices for extended periods, creating data gaps
• Poor adherence patterns: Gradual decline in wear time over study duration
• Technical issues: Device malfunctions, sync failures, battery problems
• Site-level variations: Different sites may have systematically different compliance patterns
• Delayed detection: Problems discovered during analysis, months after they occurred

The challenge is to detect these issues in near-real-time, surface them to the right stakeholders, and enable timely corrective action.

Approach

1. Key Quality Indicators

Defined a tiered set of quality indicators, from leading indicators (early warning signs) to lagging indicators (confirmed problems):

• Daily data receipt: Is data arriving as expected each day?
• Wear time compliance: Are participants meeting minimum wear requirements?
• Valid wear windows: Percentage of expected measurement periods with sufficient data
• Missingness patterns: Systematic gaps vs. random missing data
• Device health: Battery levels, sync frequency, error rates

2. Monitoring Dashboard

Built interactive dashboards that surface key quality metrics at multiple levels:

• Study-level: Overall health of data collection, trend lines, aggregate compliance
• Site-level: Comparative performance across sites, outlier detection
• Participant-level: Individual compliance trajectories, specific issues

Dashboards were designed for non-technical users (study managers, clinical operations), with clear visual hierarchy and actionable insights.

3. Alerting and Escalation

Defined threshold-based alerts to flag issues requiring attention:

• Participant drops below minimum wear time for consecutive days
• Site-level compliance falls below study target
• Unexpected gaps in data transmission

Established escalation workflows connecting monitoring outputs to operational processes (site contact, participant reminder, device troubleshooting).

4. QA-Compatible Outputs

Designed data quality outputs compatible with clinical data management processes:

• Standardized query formats for data management teams
• Audit trails for compliance tracking
• Exportable reports for study documentation

Outcomes

• Early problem detection: Issues identified days or weeks earlier than traditional approaches
• Targeted interventions: Site and participant-level visibility enabled precise remediation
• Improved overall compliance: Proactive monitoring contributed to higher average wear time across studies
• Operational efficiency: Study teams could focus attention where it mattered most
• Documentation trail: QA-ready outputs supported audit and regulatory requirements

Key Learnings

• Design for action: Monitoring is only valuable if it leads to intervention—design with action in mind
• Balance sensitivity and noise: Alerts that fire too often get ignored; tune thresholds carefully
• Consider the user: Study managers need different views than data scientists—design for your audience
• Integrate with workflows: Standalone dashboards are less impactful than those embedded in operational processes

Technical Implementation