Digital Phenotyping and Personalized Health
How can continuous physiological and multimodal biological data be transformed into reproducible digital phenotypes for personalized medicine?
Projects
Our projects aim to establish the rigorous methodological foundations of digital biostatistics while demonstrating their practical value in biomedical research, clinical practice, and population health. Our goal is to develop the statistical methods required to transform continuous, longitudinal, and multimodal health data into reliable biomedical knowledge and clinically meaningful decisions. We organize this work around three interconnected research programs, with the overarching goal of generating real-world impact in metabolic health, diabetes care and personalized nutrition.
Flagship Research Programs
Digital Biomarkers for Disease Progression
Can near-continuous digital measurements detect clinically meaningful changes before conventional clinical assessments?
Population Health and Digital Epidemiology
How can digital health data improve population-level surveillance, prevention, and healthcare planning?
Open questions we are working to solve
Our research is structured around a set of open methodological, applied, and clinical questions. These questions guide the development of new statistical theory while ensuring that the resulting methods address consequential problems in medicine and public health.
E-values for High-Dimensional Multiple Testing
How can e-values be developed into a general framework for large-scale multiple-hypothesis testing in high-dimensional digital health studies?
Optimal Conformal Prediction for Random-Effects Models
How can conformal prediction be extended to multilevel and random-effects models while preserving meaningful marginal and conditional coverage guarantees?
Conditional Predictive Inference for Random Objects
How can we construct predictive regions for random objects in metric spaces that provide strong conditional coverage guarantees without relying on restrictive smoothness assumptions?
Deep Learning Without Manifold Assumptions
What theoretical guarantees can be established for deep-learning algorithms operating on general non-Euclidean spaces, without assuming that the observations lie on a smooth manifold, and motivated by modern digital health applications?
Mathematical Foundations of Distributional Data Analysis
What mathematical and statistical foundations are needed to establish distributional data analysis as a rigorous, interpretable, and broadly applicable framework for digital health?
Applied and clinical questions
Alongside these methodological challenges, we investigate major applied questions through which advanced statistical methods could produce tangible clinical and biomedical value.
Genetic Discovery Using Digital Phenotypes
Can digital phenotypes reveal genetic associations that conventional clinical biomarkers fail to detect?
Validation of Glucodensity as a Digital Biomarker
Can glucodensity and related distributional representations be validated as robust, reproducible, and more informative digital biomarkers of metabolic health?
Glucose Variability and the Microbiome
What role does glucose variability play in the relationship between the microbiome and metabolic health, and how can continuous glucose-monitoring data improve its characterization?
Continuous Glucose Monitoring and Retinopathy
How are continuous glucose-monitoring phenotypes associated with diabetic retinopathy and other microvascular complications?
Phenotyping Continuous Biomedical Signals
How should continuous physiological signals be transformed into clinically meaningful and reproducible digital phenotypes, and what is required to establish a rigorous framework for CGM-based phenotyping in populations without diabetes?
Learn more about our work
Statistical Modeling of Multimodal Digital Health and Omics Data in the Human Phenotype Project
Developing statistical AI models to integrate multimodal digital health, medical imaging, genomic, and microbiome data from the Human Phenotype Project, with applications to retinopathy risk, glucose metabolism, and functional digital health phenotypes.
Spatiotemporal Methods for Digital Epidemiology with Applications in Diabetes Mellitus
Developing spatiotemporal statistical and machine learning methods to analyse digital health data from wearable devices, CGM systems, and electronic health records, with the goal of mapping diabetes risk and metabolic control across space and time.
Functional Distributional Biomarkers for Tracking ALS Disease Progression
Developing functional and distributional biomarkers from near-continuous hand-movement data to improve the monitoring of ALS progression and the prediction of survival.