A Semi-Supervised Multi-Task Learning Approach for Predicting Short-Term Kidney Disease Evolution.

Area of Science:

• Nephrology
• Artificial Intelligence
• Data Science

Background:

• Kidney Disease (KD) presents complex causes and significant socio-economic burdens.
• Early detection and management are crucial for sustainable global health strategies.
• Estimated Glomerular Filtration Rate (eGFR) is a key metric for routine KD screening.

Purpose of the Study:

• To propose a novel Semi-Supervised Multi-Task Learning (SS-MTL) approach for predicting short-term KD evolution.
• To leverage Electronic Health Records (EHRs) for monitoring and predicting eGFR trends.
• To address the challenge of limited labeled patient data in real-world scenarios.

Main Methods:

• Developed a Semi-Supervised Multi-Task Learning (SS-MTL) model.
• Incorporated temporal relatedness between consecutive time windows to capture eGFR evolution.
• Utilized unlabeled patient data to enhance model performance when labeled data is scarce.

Main Results:

• The SS-MTL approach effectively captures eGFR temporal evolution.
• Achieved up to a 4.1% gain in Recall by exploiting unlabeled patient data.
• Demonstrated improved performance in scenarios with limited labeled samples.

Conclusions:

• The SS-MTL approach offers a promising solution for predicting short-term KD progression.
• Its ability to utilize abundant unlabeled data makes it suitable for real-world clinical settings.
• The model's interpretability makes it a strong candidate for integration into general practice decision support systems for KD screening.