Aims: The Victorian Cancer Registry (VCR) conducted a project to assess the efficacy of using artificial intelligence (AI) applied to pathology reports to identify potential cancer cases for clinical trials. This initiative aimed to enhance clinical trial accessibility in Victoria, Australia.
Methods: VCR used the Rapid Case Ascertainment (RCA) module in the document processing system (E-Path Plus- an Inspirata© product). The RCA module targeted cases reported by Monash Pathology fulfilling the selection criteria for three phase three randomised controlled clinical trials at Monash Health (MH) which employed genetic markers as eligibility criteria. The AI engine extracted terms pertaining to topography and specific genetic tests from pathology reports. The identified cases were forwarded by VCR to MH for eligibility screening. The RCA’s performance was evaluated against manually reviewed cases.
Results: Between June 2022 and May 2023, 302 cases across the three studies were identified and forwarded to MH for screening. Of these, 7 were eligible to approach (0/48 in study 1, 6/19 in study 2 and 1/235 in study 3). The main reasons for ineligibility after screening were lack of tumour staging (174/295-59%) and normal genetic test results (96/295=33%). The RCA tool contributed 5 eligible cases to MH's selection. The RCA module accurately determined eligibility in 93% of pathology reports, achieving an F1 score of 0.93. The false positive rate was 4% and the false negative rate was 3%.
Conclusions: The RCA tool exhibits strong predictive capabilities for pathology selection to the 3 selected clinical trials. However, work is required to capture more granular data with confidence so as to reduce the burden of manual screening by minimising false negatives rates. This study was conducted in only one site. It may be that the tool would be more effective when applied in medical environments without extensive clinical trials infrastructure.