Background
The role of population-based cancer registries (PBCRs) is evolving beyond monitoring cancer incidence and surveillance to supporting health service planning and identifying inequalities. By 2035, cancer diagnoses will increase by 50%, placing a significant burden on PBCRs to process cases in a timely manner. Added to this pressure is an increasing demand for data elements such as ancillary genetic tests, stage at diagnosis, treatment modalities, and tumour recurrence. The Victorian Cancer Registry (VCR) implemented E-Path Plus, a natural language processing (NLP) solution using Artificial Intelligence (A.I.) to meet this demand and build capacity and capability.
We will outline how E-Path Plus is used to consolidate cancer notifications into a patient’s medical record; report on the accuracy of the NLP to abstract information;and describe its future potential.
Methods
E-Path Plus was deployed in September 2020. Customised enhancements were implemented to improve functionality and accuracy, including allocating reports into tumour-specific queues for processing by medical coders and automating the consolidation of data from multiple sources. A.I. validation compared selected auto-abstracted data elements with those recorded by tumour-specialist medical coders (gold standard).
Results
E-Path Plus processed 195,000 documents associated with 50,737 cases with a cancer diagnosis in 2020. With enhancements made to the NLP in the development phase, correlation of AI auto-abstraction with medical coder review for 22 data elements was >85%. Subsequent validation of data elements required to assist in staging cancer at diagnosis indicated areas for AI enhancement. Path Plus is currently being trialled to determine its potential in facilitating rapid recruitment of patients to clinical trials.
Discussion
A.I. offers promise in expanding datasets and enabling PBCRs to keep up with the increasing number of cancer diagnoses resulting from the ageing and growing Australian population. It requires ongoing review and customisation to minimise the requirement for manual review.