Skip to content
Mohamed bin Zayed University of Artificial Intelligence Digital Biostatistics Marcos Matabuena’s research group

Spatiotemporal Methods for Digital Epidemiology with Applications in Diabetes Mellitus

In this research project, we will develop and validate spatiotemporal methods for analysing digital health data collected through continuous monitoring in real-world conditions using wearable and mobile devices, together with electronic health records. Although these technologies generate dense, continuous, and structured data streams, current analytical approaches often rely on scalar summaries or aggregated measures, which fail to capture their temporal dynamics, distributional structure, and geographic heterogeneity. As a result, existing methods are suboptimal for fully exploiting digital health data to understand chronic disease progression and risk.

Objectives

Specifically, we will develop methods to construct epidemiological maps from digital health data in order to create quantitative tools for:

  1. Monitoring and understanding the evolution of chronic diseases such as diabetes from population-level and geographic perspectives.
  2. Analysing metabolic control and the risk of events such as hypoglycaemia and hyperglycaemia using data from continuous glucose monitoring (CGM) devices within a spatiotemporal framework.

Methodological framework

To achieve these objectives, we will design a spatiotemporal statistical framework together with machine learning algorithms for random objects in metric spaces. In this setting, the units of observation are random functions and other random objects defined in general metric spaces. In particular, we will focus on distribution-based representations derived from digital health devices that summarise recorded time series, such as the glucodensity-based approach. The proposed framework will estimate key clinical outcomes across space and time while providing uncertainty-aware prediction regions through conformal prediction methods in metric spaces.

Data and application

The proposed models will be applied to data from the Human Phenotype Project (HPP) and, where feasible, to electronic health records from Abu Dhabi. Our goal is to monitor diabetes risk across space and time using spatiotemporal epidemiological maps derived from CGM data, and to better understand the geographic and temporal effectiveness of disease management and metabolic control under real-world conditions.