Artificial INtelligence to Support Informed DEcision-making (INSIDE) for Improved Literature Analysis in Oncology

Area of Science:

• Artificial Intelligence in Medicine
• Computational Biology and Bioinformatics
• Oncology and Cancer Research

Background:

• Optimizing therapeutic sequencing in prostate cancer (PC) is complex and can benefit from AI-driven analysis of medical literature.
• Existing literature search tools may not efficiently provide relevant information for complex treatment decisions in PC.

Purpose of the Study:

• To validate the INSIDE PC platform for querying therapeutic sequencing literature in prostate cancer.
• To establish methodologies for evaluating AI-based clinical support tools.
• To assess INSIDE PC's performance compared to PubMed in retrieving relevant PC treatment information.

Main Methods:

• INSIDE PC was developed using PubMed Bidirectional Encoder Representations from Transformers for semantic literature analysis.
• Publications were ranked and visualized for relevance using data analytics.
• Expert evaluation using normalized discounted cumulative gain (nDCG) scores compared INSIDE PC and PubMed outputs for specific PC treatment sequencing questions.

Main Results:

• INSIDE PC outperformed PubMed in relevance for novel hormonal therapy (NHT) followed by NHT (question 1) and NHT or PARPi followed by 177Lu-PSMA-617 (question 3).
• Performance was similar between INSIDE PC and PubMed for NHT followed by poly (ADP-ribose) polymerase inhibitors (PARPi) (question 2).
• INSIDE PC demonstrated superior ranking and relevance for the top 5, 10, and 20 retrieved publications in specific scenarios.

Conclusions:

• INSIDE PC effectively assists clinicians in navigating the literature for prostate cancer therapeutic sequencing.
• The study established standards for evaluating AI literature extraction tools.
• INSIDE PC shows competitive performance with PubMed and supports shared decision-making in advanced PC treatment.