Development and Validation of an Automated Pediatric Cancer Staging Calculator Using the Toronto Pediatric Cancer...

Area of Science:

• Oncology
• Medical Informatics
• Computational Biology

Background:

• Accurate pediatric cancer staging is crucial for prognosis and research.
• The Toronto Pediatric Cancer Stage Guidelines provide standardized staging for childhood cancers.
• Electronic health records (EHRs) offer a rich data source for developing automated clinical tools.

Purpose of the Study:

• To develop and validate an automated framework for staging pediatric cancers using EHR data.
• To assess the accuracy and consistency of the automated staging system against expert consensus.
• To establish a standardized, efficient method for pediatric cancer staging.

Main Methods:

• An automated extraction pipeline using agents was designed to process EHR notes within 3 months of diagnosis.
• The system selected appropriate staging schemas, calculated stages, and validated accuracy.
• The tool was tested on 433 pediatric cancer cases, with outputs compared to a ground truth established by four oncologists.

Main Results:

• The automated system achieved 91.2% overall accuracy compared to expert consensus.
• Independent runs showed high agreement (89.8%, Cohen's κ = 0.785).
• Accuracy was highest (97%) on the first staging attempt, decreasing on subsequent recalculations.

Conclusions:

• This study presents the first automated pediatric cancer staging system utilizing the Toronto criteria.
• The system demonstrates high accuracy and consistency, comparable to human experts.
• A hybrid approach, flagging uncertain cases for human review, can achieve ~97% accuracy with ~30% human input.