Reproduce “TRACE: Early Detection of Chronic Kidney Disease Onset with Transformer-Enhanced Feature Embedding” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yu Wang, Ziqiao Guan, Wei Hou, Fusheng Wang

Researchers at Stony Brook University developed TRACE, a deep learning framework for early Chronic Kidney Disease (CKD) onset prediction, leveraging a pre-trained Transformer-RNN autoencoder. The model achieved a 2.31% relative improvement in AUPRC over existing baselines, successfully learning comprehensive patient history representations and identifying key risk factors.

Key insights

  • The combination of a Transformer for intra-visit feature encoding and an RNN autoencoder for longitudinal history summarization enables the model to learn more robust and context-aware patient representations.
  • Pre-training the Transformer-RNN autoencoder on sequential EHR data is crucial for extracting meaningful insights and addressing the challenges of sparse medical histories and complex risk factors in chronic disease prediction.
  • The learned patient embeddings from TRACE are more discriminative, leading to a clearer separation between positive and negative CKD patients, which is vital for early detection.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.