JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Pancreatic cancer risk prediction using deep sequential modeling of longitudinal diagnostic and medication records

  • 0VA Boston Healthcare System, Boston, MA, USA; Boston University School of Medicine, Boston, MA, USA.
Cell Reports. Medicine +

|

September 17, 2025

|

September 17, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

We developed an AI model using electronic health records to identify individuals at high risk for pancreatic cancer (PDAC). This tool can significantly improve early detection and patient outcomes.

Area Of Science

  • Oncology
  • Artificial Intelligence
  • Health Informatics

Background

  • Pancreatic ductal adenocarcinoma (PDAC) has low survival rates due to late diagnosis.
  • Current early detection methods are costly, and population-wide screening is lacking.
  • Identifying high-risk individuals is crucial for timely intervention.

Purpose Of The Study

  • To develop and validate a transformer-based AI model for predicting PDAC risk.
  • To leverage longitudinal electronic health record (EHR) data for early risk identification.
  • To assess the impact of diagnostic and medication trajectories on PDAC prediction.

Main Methods

  • A transformer-based model was trained on EHR data from 19,426 PDAC cases and ~15.9 million controls.
  • The model integrated diagnostic codes and medication histories to predict PDAC risk over 6-, 12-, and 36-month windows.
  • Feature importance analysis identified key predictors of PDAC risk.

Main Results

  • The AI model demonstrated significant predictive performance for PDAC.
  • Incorporating medication data substantially improved model accuracy.
  • In a cohort of 1 million, the top-risk individuals showed 70-115 times higher 3-year PDAC incidence compared to age/sex estimates.
  • Chronic inflammatory conditions and specific medications were identified as significant risk factors.

Conclusions

  • AI-driven analysis of EHR data can effectively identify high-risk individuals for PDAC.
  • The model offers a potential pathway for improved early detection and patient management.
  • This approach may reduce healthcare costs associated with late-stage cancer diagnosis.

Keywords: