Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bridging Stepwise Lab-Informed Pretraining and Knowledge-Guided Learning for Diagnostic Reasoning

Pengfei Hu, Chang Lu, Fei Wang, Yue Ning

TL;DR

This work tackles the challenge of enriching EHR-based diagnostic predictions with clinically meaningful knowledge. It introduces DuaLK, a dual-expertise framework that jointly leverages a bi-hierarchical Diagnosis KG (constructed from public medical knowledge and LLM-generated triples) and a lab-informed pretraining proxy task to mimic stepwise clinical reasoning. The method yields graph-enhanced patient embeddings and a polar-space KG embedding, achieving consistent improvements over strong baselines on MIMIC-III and MIMIC-IV for diagnosis and heart failure prediction, while maintaining scalable, modular deployment. By integrating structured medical knowledge with individual lab signals, DuaLK enhances accuracy and interpretability, with strong performance on rare diagnoses and robust data-efficiency characteristics.

Abstract

Despite the growing use of Electronic Health Records (EHR) for AI-assisted diagnosis prediction, most data-driven models struggle to incorporate clinically meaningful medical knowledge. They often rely on limited ontologies, lacking structured reasoning capabilities and comprehensive coverage. This raises an important research question: Will medical knowledge improve predictive models to support stepwise clinical reasoning as performed by human doctors? To address this problem, we propose DuaLK, a dual-expertise framework that combines two complementary sources of information. For external knowledge, we construct a Diagnosis Knowledge Graph (KG) that encodes both hierarchical and semantic relations enriched by large language models (LLM). To align with patient data, we further introduce a lab-informed proxy task that guides the model to follow a clinically consistent, stepwise reasoning process based on lab test signals. Experimental results on two public EHR datasets demonstrate that DuaLK consistently outperforms existing baselines across four clinical prediction tasks. These findings highlight the potential of combining structured medical knowledge with individual-level clinical signals to achieve more accurate and interpretable diagnostic predictions. The source code is publicly available on https://github.com/humphreyhuu/DuaLK.

Bridging Stepwise Lab-Informed Pretraining and Knowledge-Guided Learning for Diagnostic Reasoning

TL;DR

This work tackles the challenge of enriching EHR-based diagnostic predictions with clinically meaningful knowledge. It introduces DuaLK, a dual-expertise framework that jointly leverages a bi-hierarchical Diagnosis KG (constructed from public medical knowledge and LLM-generated triples) and a lab-informed pretraining proxy task to mimic stepwise clinical reasoning. The method yields graph-enhanced patient embeddings and a polar-space KG embedding, achieving consistent improvements over strong baselines on MIMIC-III and MIMIC-IV for diagnosis and heart failure prediction, while maintaining scalable, modular deployment. By integrating structured medical knowledge with individual lab signals, DuaLK enhances accuracy and interpretability, with strong performance on rare diagnoses and robust data-efficiency characteristics.

Abstract

Despite the growing use of Electronic Health Records (EHR) for AI-assisted diagnosis prediction, most data-driven models struggle to incorporate clinically meaningful medical knowledge. They often rely on limited ontologies, lacking structured reasoning capabilities and comprehensive coverage. This raises an important research question: Will medical knowledge improve predictive models to support stepwise clinical reasoning as performed by human doctors? To address this problem, we propose DuaLK, a dual-expertise framework that combines two complementary sources of information. For external knowledge, we construct a Diagnosis Knowledge Graph (KG) that encodes both hierarchical and semantic relations enriched by large language models (LLM). To align with patient data, we further introduce a lab-informed proxy task that guides the model to follow a clinically consistent, stepwise reasoning process based on lab test signals. Experimental results on two public EHR datasets demonstrate that DuaLK consistently outperforms existing baselines across four clinical prediction tasks. These findings highlight the potential of combining structured medical knowledge with individual-level clinical signals to achieve more accurate and interpretable diagnostic predictions. The source code is publicly available on https://github.com/humphreyhuu/DuaLK.

Paper Structure

This paper contains 31 sections, 13 equations, 4 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Clinical Prediction Models
  4. Knowledge-Guided Clinical Prediction
  5. Clinical Prediction with Multi-aspect Features
  6. Reasoning in Clinical Prediction
  7. Methodology
  8. General Notation
  9. Generate Diagnosis KG
  10. Polar-space KG Embedding
  11. Encode Graph-Enhanced Patient Embeddings
  12. Stepwise Lab-Informed Pretraining
  13. Fine-Tuning and Inference
  14. Experiments
  15. Experiment Setting
  16. ...and 16 more sections

Figures (4)

  • Figure 1: Overview of the proposed DuaLK framework.
  • Figure 2: Overview of Augmented Bi-hierarchical KG.
  • Figure 3: Performance by EHR Training Data Sizes. Values on the x-axis indicate % of the entire training data. The dotted lines separate two ranges: [1, 10] and [10, 100] (%).
  • Figure 4: Results for the $1^{st}$ and $2^{nd}$ case studies, full results are shown in Appendix \ref{['app:detail_case']}.