NIH grant: Digital biomarker discovery with multimodal and longitudinal wearable data using novel functional and distributional data analysis methods (1R01HL187193-01)

retained text

NIH grant 1R01HL187193-01: Digital biomarker discovery with multimodal and longitudinal wearable data using novel functional and distributional data analysis methods
Institute: {'code': 'HL', 'abbreviation': 'NHLBI', 'name': 'National Heart Lung and Blood Institute', 'admin_org_id': '12500068', 'admin_funding_url': ''}. PI: Erjia  Cui. Award: $649041.
Start: 2026-09-05T00:00:00.
Project Summary:
The advent of wearable and sensor technologies has led to a new era in digital health monitoring, generating
continuous streams of physiological data in subjects’ free-living environment. These devices generate
continuous streams of physiological data, including heart rate (ECG), brain activity (EEG), physical activity (PA),
and continuous glucose monitoring (CGM). Such data capture timing, variability, and interplay of physiological
signals, which can reveal subtle yet clinically meaningful behavioral phenotypes. However, most analytical
approaches reduce these rich, high-frequency sensor data (e.g., accelerometer data) to simple summary
statistics (for example, total log activity count (TLAC), minutes of moderate-to-vigorous physical activity (MVPA),
and active-to-sedentary transition probability (ASTP). This practice discards crucial distributional and temporal
information, limiting our ability to identify novel digital biomarkers and accurately characterize disease
progression. To address this methodological gap, we propose to develop modern and flexible statistical methods
which integrate functional data analysis (FDA) and distributional data analysis (DDA) to investigate both the
distributional and temporal aspects of wearable data. We look to accomplish the following aims in this proposal.
Aim 1. Develop and validate a clustering methodology to identify clinically meaningful digital phenotypes from
multi-modal wearable data using multivariate partially functional distributional representations. Aim 2. Use
longitudinal trajectories of wearable data to identify clinically meaningful digital phenotypes based on partially
functional distributional representations. Aim 3. Develop methods for longitudinal functional data to characterize
multivariate, sparse, or non-Gaussian trajectories. The proposed methods are significant because they will
enable discovery of new digital phenotypes, improve risk assessment and support dynamic modelling of disease
progression. They are innovative because they introduce new statistical methods that jointly leverage
distributional and temporal information, bridging two important statistical paradigms in digital health research.
The results hold promises to advance understanding of how longitudinal trajectories of daily activity distributions
change, capturing how an individual's 24-hour activity pattern evolves over years, and obtaining distinct digital
phenotypes of aging that predict cumulative health risks and major health transitions, such as cardiovascular
disease, cognitive impairment and changes in functional independence among older adults.

document fetched 2026-09-13 02:46