Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automatic Extraction of Disease Risk Factors from Medical Publications

Maxim Rubchinsky, Ella Rabinovich, Adi Shraibman, Netanel Golan, Tali Sahar, Dorit Shweiki

TL;DR

The paper addresses automating the extraction of disease risk factors from unstructured biomedical literature by implementing a BioBERT-based multi-step pipeline that (i) retrieves PubMed abstracts, (ii) classifies abstracts for risk-factor content, and (iii) extracts precise risk factors via a fine-tuned extractive QA model. It contributes a scalable data workflow and several datasets, including over 1,700 manually annotated risk-factor items spanning 15 diseases and a large-scale automatically extracted corpus (over 160,000 risk factors across 744 diseases). The QA-based approach achieves a strong F1 of $88.23\%$ and an exact-match of $61.76\%$, with a high classifier accuracy of $92\%$, demonstrating promise for accelerating evidence synthesis in preventive medicine. The work provides valuable resources for future research, discusses limitations around context-specificity and data variability, and suggests directions such as more advanced QA models and broader data integration to keep pace with evolving medical knowledge.

Abstract

We present a novel approach to automating the identification of risk factors for diseases from medical literature, leveraging pre-trained models in the bio-medical domain, while tuning them for the specific task. Faced with the challenges of the diverse and unstructured nature of medical articles, our study introduces a multi-step system to first identify relevant articles, then classify them based on the presence of risk factor discussions and, finally, extract specific risk factor information for a disease through a question-answering model. Our contributions include the development of a comprehensive pipeline for the automated extraction of risk factors and the compilation of several datasets, which can serve as valuable resources for further research in this area. These datasets encompass a wide range of diseases, as well as their associated risk factors, meticulously identified and validated through a fine-grained evaluation scheme. We conducted both automatic and thorough manual evaluation, demonstrating encouraging results. We also highlight the importance of improving models and expanding dataset comprehensiveness to keep pace with the rapidly evolving field of medical research.

Automatic Extraction of Disease Risk Factors from Medical Publications

TL;DR

The paper addresses automating the extraction of disease risk factors from unstructured biomedical literature by implementing a BioBERT-based multi-step pipeline that (i) retrieves PubMed abstracts, (ii) classifies abstracts for risk-factor content, and (iii) extracts precise risk factors via a fine-tuned extractive QA model. It contributes a scalable data workflow and several datasets, including over 1,700 manually annotated risk-factor items spanning 15 diseases and a large-scale automatically extracted corpus (over 160,000 risk factors across 744 diseases). The QA-based approach achieves a strong F1 of and an exact-match of , with a high classifier accuracy of , demonstrating promise for accelerating evidence synthesis in preventive medicine. The work provides valuable resources for future research, discusses limitations around context-specificity and data variability, and suggests directions such as more advanced QA models and broader data integration to keep pace with evolving medical knowledge.

Abstract

We present a novel approach to automating the identification of risk factors for diseases from medical literature, leveraging pre-trained models in the bio-medical domain, while tuning them for the specific task. Faced with the challenges of the diverse and unstructured nature of medical articles, our study introduces a multi-step system to first identify relevant articles, then classify them based on the presence of risk factor discussions and, finally, extract specific risk factor information for a disease through a question-answering model. Our contributions include the development of a comprehensive pipeline for the automated extraction of risk factors and the compilation of several datasets, which can serve as valuable resources for further research in this area. These datasets encompass a wide range of diseases, as well as their associated risk factors, meticulously identified and validated through a fine-grained evaluation scheme. We conducted both automatic and thorough manual evaluation, demonstrating encouraging results. We also highlight the importance of improving models and expanding dataset comprehensiveness to keep pace with the rapidly evolving field of medical research.
Paper Structure (34 sections, 4 figures, 7 tables)

This paper contains 34 sections, 4 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Conclusion
  4. Dataset
  5. Disease Dataset Collection
  6. Seed Dataset with Relevant Abstracts
  7. Retrieval of Abstracts Discussing Risks
  8. Annotation of Abstracts for Risk Factors
  9. QA Seed Dataset with Risk Factors
  10. Methodology and Experiments
  11. Methodology
  12. Detection of Abstracts with Risk Factors
  13. Identification of Disease Risk Factors
  14. Fine-tuning the QA Model
  15. Determining the Maximum Answer Length
  16. ...and 19 more sections

Figures (4)

  • Figure 1: The pipeline for extraction of disease's risk factors: (1) medical abstracts are retrieved from PubMed, (2) a specifically fine-tuned binary classifier is used to identify abstracts with risk factors information, and (3) precise textual spans containing risk factors are extracted via a QA model, fine-tuned on manually annotated QA items.
  • Figure 2: Disease Risk Factor Annotation System: disease details as retrieved from KEGG and parsed.
  • Figure 3: Disease Risk Factor Annotation System: manual annotation of spans containing risk factors; multiple risk factors for the same disease can be identified in the same abstract.
  • Figure :