Improving the Scientific Validity of Smartphone-Based Symptom Monitoring of Diseases

Abstract Remote monitoring using smartphone sensors offers a promising avenue for tracking the symptoms of neurodegenerative diseases such as Parkinson’s disease. However, collecting reliable sensor data outside of clinical environments introduces challenges related to participant adherence, unexpected behaviors, and environmental variability. These issues reduce data quality and limit the interpretability of digital health studies. This proposes an automated quality control framework that systematically verifies adherence and filters unreliable data. The framework consists of three components: (1) monitoring whether participants complete at least ten sessions per quarter; (2) detecting flat signals using rolling standard deviation to identify periods of minimal movement; and (3) applying Gaussian Mixture Model clustering with temporal smoothing to classify adherence and violation states in accelerometer data. We evaluate this approach using real-world data from smartphone-based active motor tasks. Our results show that the method identifies protocol violations and enhances data quality, providing a scalable solution for improving the validity of remote Parkinson’s disease monitoring studies.