ICU mortality risk prediction from longitudinal EHR data

This project focuses on ICU mortality risk prediction using longitudinal clinical data from critically ill patients. The study explores both unimodal models using sequential laboratory data and multimodal frameworks integrating chest X-rays, echocardiogram information, and laboratory measurements. Current work investigates GRU, time-aware GRU, and RNN-based architectures to capture temporal patterns directly from raw clinical observations for robust and generalisable early risk prediction in intensive care settings.

• Healthcare Informatics

Longitudinal cancer-risk prediction in diabetes using data-centric artificial intelligence

This project develops long-term cancer risk prediction models using longitudinal clinical data from diabetic patients tracked over nearly 20 years. A forward-prediction framework is used to estimate cancer risk within the next three years using routinely collected demographic and laboratory measurements from electronic health records. Rather than relying heavily on age, the models leverage carefully curated clinical features to identify subtle patterns associated with future cancer development.

• Healthcare Informatics