Data Processing Framework for Digital Health Monitoring using Roadmap 2.0

Abstract

Background: Mobile health technologies offer unprecedented opportunities for longitudinal data collection in clinical research. However, the vast amount and complexity of raw data generated pose significant challenges in data preprocessing, analysis, and reproducibility. Roadmap 2.0, a mobile app for positive wellness-based psychology interventions, collects physiological and psychological data from bone marrow transplant patient-caregiver dyads for up to 120 days post-transplant using wearable sensors. Objectives: To develop a flexible, efficient, and standardized data processing pipeline for Roadmap 2.0 that streamlines the handling of complex mobile health data, ensures data integrity and participant privacy, and facilitates reproducible research across various study designs and populations. Methods: We designed a six-module pipeline: (1) Data Reading, (2) Data Transformation, (3) Data Filtration, (4) Data Censoring, (5) Temporal Adjustment, and (6) Data Visualization. Each module was implemented as a separate R script, allowing for flexibility and reusability across different study populations, metrics, and time resolutions. The pipeline processes various data types including self-reported mood scores, Fitbit data (sleep, heart rate, steps), and PROMIS survey measures. Results: The pipeline successfully handled large datasets (e.g., >30,000 mood entries, >50 million heart rate readings) and addressed challenges such as data cleaning, reformatting, and restructuring. Key features include: 1. Automated filtering of adult patient data for the first 120 days post-transplant. 2. Robust handling of high-resolution Fitbit data, including device ID matching and aggregation of second-level to minute-level data. 3. Implementation of data censoring to protect participant privacy, including anonymization of identifiers and removal of clinical information. 4. Creation of daily summaries for high-frequency data (heart rate, steps) to facilitate downstream analyses. 5. Generation of standardized visualizations, including heatmaps of dyadic data reporting and time series plots. Conclusions: The developed modular data processing pipeline significantly enhances the efficiency and reproducibility of mobile health research using Roadmap 2.0 data. By standardizing preprocessing steps, facilitating seamless data integration, and implementing robust anonymization protocols, the pipeline addresses critical challenges in handling complex, multi-source mobile health data. Its adaptability to various study designs and populations promotes the application of consistent, rigorous research methods across diverse contexts. This approach not only streamlines data management throughout the research lifecycle but also sets a foundation for more advanced analytics, including the potential integration of machine learning techniques.