Combining PubMed knowledge and EHR data to develop a weighted bayesian network for pancreatic cancer prediction

Area of Science:

• Biomedical Informatics
• Computational Biology
• Oncology

Background:

• Pancreatic cancer prediction remains a challenge, necessitating advanced computational approaches.
• Existing models often lack comprehensive integration of diverse data sources like biomedical literature and clinical records.

Purpose of the Study:

• To develop and evaluate a novel weighted Bayesian Network Inference (BNI) model for enhanced pancreatic cancer prediction.
• To integrate knowledge from PubMed abstracts with Electronic Health Records (EHRs) to improve predictive accuracy.

Main Methods:

• A keyword-based algorithm extracted and classified PubMed abstracts to determine risk factor associations.
• Normalized weights derived from PubMed data were incorporated into a conventional BNI model.
• The weighted BNI model utilized EHR data to calculate prior probabilities for 20 risk factors.
• A software tool, iDiagnosis, was developed to implement the weighted BNI model for prediction.

Main Results:

• The weighted BNI model demonstrated significantly superior performance compared to the conventional BNI model.
• The proposed model outperformed k-Nearest Neighbor and Support Vector Machine classifiers in a case-control dataset evaluation.
• Integration of PubMed knowledge and EHR data led to remarkable accuracy improvements.

Conclusions:

• The weighted BNI model effectively leverages both biomedical literature and clinical data for pancreatic cancer prediction.
• This approach offers a significant advancement over existing methods, showing promise for earlier and more accurate diagnosis.