Glycemic Health
We aim to understand and prevent dysglycemia by uncovering how lifestyle, behavior, and physiology interact to shape metabolic risk. Prediabetes (PD) and type 2 diabetes (T2D) affect 122 million Americans, yet remain critically underdiagnosed—81% and 23% of those with PD and T2D, respectively, are unaware of their condition. Traditional screening guidelines have failed to close this diagnostic gap due to their coarseness and limited reach. Our work develops innovative, scalable strategies for early detection and prevention by leveraging digital data from everyday consumer devices—smartphones, smartwatches, and wearables—to identify subtle, real-world signatures of glycemic risk and resilience.
Diabetes Watch (Karnika Singh & Leeor Hershkovich)
We aim to develop digital biomarkers of glycemic health using digital health technologies. We are validating and extending our preliminary work developing the wearable algorithms to function across a wider range of glycemic variability, interstitial glucose, and A1c values and to move from research-grade wearables to consumer-grade smartwatches. Funded by NIH R01 DK133531.
Text message campaign (Hayoung Jeong)
We launched a diabetes-awareness text messaging campaign to reach individuals who meet the American Diabetes Association (ADA) screening criteria. The project evaluates whether these text messages can increase A1c screening rates and the positive predictive value of the current ADA screening guidelines. Funded by NIH R01 DK133531.
Lifestyle-factors and Glycemic Variability (AI-READI) - Diabetes Analysis (Bill Chen)
We’re studying how detailed blood glucose patterns, captured by continuous glucose monitoring (CGM), relate to a wide range of health and lifestyle factors. Drawing on the AI-READI dataset of 1,067 participants across healthy, prediabetic, and diabetic groups, we are analyzing links between glucose dynamics, lab results, and wearable-derived behaviors. Our goal is to identify early indicators of metabolic risk and guide more personalized diabetes prevention strategies.
LeMonAids: Leveraging multimodal Monitoring to Ascertain Intra-individual Diabetes Subphenotypes (Leeor Hershkovich & Bill Chen)
We are developing a foundation model that integrates continuous glucose monitoring (CGM), dietary, and wearable activity data to capture individual variability in Type 2 Diabetes management. By identifying unique diabetes subphenotypes and predicting postprandial glucose responses, LeMonAids aims to enable personalized lifestyle and treatment strategies for improved, data-driven glycemic control.
Non-invasive classification of glycemic status using features from ECG and/or PPG (AI-READI) (Shekh Md Mahmudul Islam)
This project explores the use of features derived from non-invasive physiological signals such as electrocardiogram (ECG) and photoplethysmogram (PPG) in classifying glycemic status (e.g., normoglycemia, prediabetes, diabetes) using. By extracting and analyzing key signal features (e.g., morphological or statistical), the study aims to uncover subtle patterns linked to glucose metabolism, enabling continuous, non-invasive monitoring and advancing early detection and personalized management of diabetes.
Physiologic Monitoring & Digital Biomarkers in Acute and Procedural Care
Using wearable sensors and digital biomarkers, we aim to detect pain, stress, and complications earlier—helping clinicians personalize care and improve patient recovery.
Developing a risk prediction engine for relapse in opioid use disorder (Lauren Lederer)
A collaborative research initiative between Duke University, UNC, and the Digital Medicine Society, we aim to build a tool that uses data from digital sensor technologies, like wearables, to predict when people affected by Opioid Use Disorder (OUD) might relapse. This tool would offer a low burden, high reach, and scalable solution for preventing lapses & relapses for all individuals with OUD, offering a pathway to improve care for all people affected by OUD. Funded by FDA.
Perioperative (Post Surgical) Monitoring (Lauren Lederer)
Evaluating the efficacy and utility of continuous perioperative monitoring using commercial wearables.
InfectionWatch (Jerry Yang)
InfectionWatch investigates how respiratory infections alter human physiology using wearable device data from the “wild” (i.e., electronic health records) as well as in controlled human challenge studies. By identifying early physiological signatures of infection, the project aims to translate these insights into real-world settings for early, non-invasive detection of infectious diseases.
rPPG (Yihang Jiang)
Remote photoplethysmography (rPPG) offers a non-contact solution for monitoring vital signs by analyzing subtle skin color variations that correspond to blood flow. Our research focuses on extracting heart rate, from the region of interest beneath the eyelids where it exhibits clear photoplethysmographic signals. We aim to expand the application field of rPPG algorithm to Augmented Reality (AR) Devices by testing the new region of interest that can be captured by eye tracking cameras.
Risk Factor Discovery in Population Health
We identify and model biological, behavioral, and environmental determinants of disease across large, diverse cohorts. These projects often integrate multi-omic, clinical, and contextual data (e.g., from mobile and wearable devices) to uncover novel predictors of health outcomes and to advance precision prevention strategies.
Verily Project Baseline: Microbiomes (Anita Shlesinger)
We are investigating the associations of the gut microbiome with cardiovascular health and disease. By integrating microbiome profiles with deep phenotyping, including continuous physiological data from wearables, we aim to better understand variations across individuals throughout the spectrum from health to disease development.
All of Us: ASCVD Prediction (Anita Shlesinger & Harrison Kane)
Early detection is critical for treating patients with atherosclerosis, but current methods rely on population risk scores that underperform in minority groups or on expensive and radiation-exposing imaging. This study utilizes consumer wearables towards scalable, noninvasive screening for ASCVD status.
All of Us: Built environment (Hayoung Jeong)
Adherence to recommended physical activity (PA) guidelines remains suboptimal in the United States, influenced by factors such as individual behavior, lifestyle, and characteristics of the built environment. This F31-funded project (1F31HL179990-01) will analyze high-resolution wearable data to model PA patterns and evaluate how changes in environmental context may influence PA and related health outcomes. The findings will inform data-driven strategies for public health interventions and urban planning aimed at improving health outcomes across the population.
Sleep and Circadian Rhythms
We are quantifying cardiovascular responses to circadian disruption and comparing digital biomarkers against more conventional circadian metrics for evaluating the cardiovascular effects of circadian disruptions. Such digital biomarkers are meant to provide longitudinal, accessible and individualized metrics for evaluating the cardiovascular risk resulting from circadian disruptions, including events such as shift work and daylight saving time transitions. Sleep Health and Wearable Technology: Algorithmic Development Towards Field-Based Sleep Monitoring (Wang, 2025).
Digital Health Equity
Advancing digital health equity requires ensuring that emerging wearables technology benefit all individuals, regardless of demographics, health status, or access to technology. Our body of work focuses on addressing key barriers to equitable digital health by developing inclusive datasets (DREAMT) and robust algorithms (WatchSleepNet), and representative study designs. Complementary work examining adherence and retention in longitudinal digital health studies (link), as well as demographic imbalances in bring-your-own-device (BYOD) research (link), identifies structural factors that limit participation and data representativeness (link). In alignment with Duke’s Interdisciplinary Hub for Rural Health Equity, we extend these efforts to rural and underserved communities, co-creating digital health tools that support equitable chronic disease prevention and management. Together, these efforts aim to close the gap between technological innovation and equitable health impact, ensuring that digital health tools are accessible, accurate, and meaningful for all.
Biosignal Review (Bill Chen & Anita Shlesinger)
As AI/ML models increasingly influence healthcare decisions, understanding how these algorithms perform across diverse populations becomes critical. This project aims to assess whether and how sensitive attributes, such as demographics, are accounted for in these studies, as well as identify common methods used to evaluate and mitigate bias.
Tools And Standards For Digital Biomarker Discovery
We are developing computational methods, open platforms, and analytic frameworks that enable high-quality biosignal analysis and reproducible digital biomarker research. These projects build the methodological and infrastructure foundation that supports scalable, trustworthy digital health discovery across diverse data modalities and studies.
FOMO: Functions for Optimizing Missing Observations (Leeor Hershkovich & Hayoung Jeong)
Researchers typically treat missingness in wearable data as noise to be handled by exclusion. We challenge this view by proposing that missingness, driven largely by user adherence, is not random noise and actually can serve as a meaningful digital biomarker. Using a large observational dataset, we characterize the temporal patterns of missingness and develop new methods for handling missingness.
Check out our recent workshop from IEEE-EMBS BHI 2025 in Atlanta, Georgia!
Read more about this project here: https://www.researchsquare.com/article/rs-5861743/v1
DBDP (Hayoung Jeong & Bill Chen)
The mission of the Digital Biomarker Discovery Project (DBDP) (https://www.dbdp.org/) is to unlock the potential of digital biomarkers. We are motivated by the transformative potential of digital biomarkers, which can offer a more continuous and comprehensive view of health than traditional clinical measurements. However, the field of digital biomarker research faces several challenges such as lack of standard practices and transparency, making difficult for researchers to build upon each other's work and collaborate. DBDP is committed to overcoming these obstacles to make the discovery and validation of digital biomarkers more efficient and robust, ultimately unlocking their full potential.
The BIG IDEAs lab is in collaboration with experts from Berkeley, UCSF, and Project Jupyter to build Jupyter Health. This innovative open-source platform aims to enhance the secure collection of data from wearable sensors to electronic health record and to streamline the development, sharing, and validation of digital biomarker algorithms. We develop modular features for Jupyter Health, ranging from comprehensive summary statistics to dynamic user visualizations that can provide immediate insights into health data. While we are currently piloting Jupyter Health development for diabetes management, we envision Jupyter Health one day covering a wide array of health conditions.