Predictive Modeling of Kidney Stone Composition Using Blood and Urine Biomarkers

06/18/2025

Patients with recurrent kidney stones face a higher risk of developing Chronic Kidney Disease (CKD), marked by progressive kidney function loss. Identifying stone types typically requires invasive procedures. This study explores a less invasive alternative by analyzing blood and urine biomarkers to predict stone composition, aiming to personalize treatments and enhance patient outcomes.Initially, clustering analysis was performed on kidney stone data to assess alignment with traditional classifications. The Ward method yielded eight clusters consistent with literature. This segmentation informed the subsequent classification phase, which sought to predict kidney stone types using two approaches. The first used clinical, blood, and urine data to classify stones into calcium oxalate, phosphate, and uric acid/urate types. The second assessed the predictive potential of novel urinary biomarkers. Findings revealed strong predictive performance even when using only blood or only urine data, offering a less invasive and efficient diagnostic alternative. While new biomarkers alone were insufficient for accurate classification, their integration with traditional variables enhanced prediction accuracy. This dissertation demonstrates the potential of combining biomarker data with clinical variables to predict stone type early in CKD progression. These insights represent a significant step toward more personalized, non-invasive approaches to managing kidney stones, contributing to improved outcomes for CKD patients.

Authors in the community:

• 

  B

  Bernardo Pavoeiro Santos

  ist193434

Supervisors of this institution:

• 

  A

  Alexandra Maria Moita Antunes

  ist24874

• 

  M

  Maria da Conceição Esperança Amado

  ist13493

mathematics - Mathematics

por - Portuguese

Instituto Superior Técnico