TRACKD
Full Title
learning Temporal pRogression pAtterns for Chronic Kidney Disease prognosisDescription
Chronic Kidney Disease (CKD) affects approximately 10-11% of adults in Western Europe and represents a particularly important burden in Portugal, where estimates range from around 9.8% to 20.9% of adults, depending on the population and diagnostic criteria used. Given its high prevalence, progressive nature, and association with kidney failure, dialysis, cardiovascular complications, and substantial healthcare costs, CKD is a major clinical and public health challenge.
A key difficulty in CKD management is the highly heterogeneous progression of the disease. Some patients remain stable for years, while others experience rapid eGFR decline, transition to advanced stages, or require renal replacement therapy. However, current computational approaches primarily focus on CKD diagnosis and often offer limited insight into how the disease evolves or which patients are likely to progress. This limits their usefulness for personalized monitoring, timely nephrology referral, and early intervention.
TRACKD addresses this gap by developing interpretable, pattern-centric prognostic models for CKD progression using longitudinal clinical data. Rather than producing only static or opaque risk predictions, the TRACKD project seeks to identify clinically meaningful patient-feature-time signatures associated with rapid progression and dialysis risk. These may include renal function trajectories, acute kidney injury episodes, multimorbidity patterns, and other markers of clinical deterioration.
The project has three main goals: (1) define clinically meaningful CKD progression outcomes; (2) develop prognostic models capable of discovering interpretable longitudinal progression signatures; and (3) benchmark these models against state-of-the-art machine learning baselines in terms of predictive performance, clinical utility, interpretability, and stability.
By generating preliminary evidence for interpretable, pattern-centric CKD prognostic modeling, TRACKD will lay a strong foundation for future publications and larger grant applications. Clinically, it may support earlier identification of high-risk patients, more personalized monitoring, timely nephrology referral, and better prioritization of interventions. More broadly, the methodology may be transferable to other chronic diseases characterized by heterogeneous progression trajectories.