Bechelli Maria L, Ivanova Kris, Tan Suan Siang, Kumar Beena, Swiatek Dayna, Arulananda Surein, Evans Sue M
Victorian Cancer Registry, Cancer Council Victoria, Victoria, Australia.
Department of Medical Oncology, Monash Health, Victoria, Australia.
JCO Clin Cancer Inform. 2025 Jan;9:e2400137. doi: 10.1200/CCI.24.00137. Epub 2025 Jan 9.
Enhancing the speed and efficiency of clinical trial recruitment is a key objective across international health systems. This study aimed to use artificial intelligence (AI) applied in the Victorian Cancer Registry (VCR), a population-based cancer registry, to assess (1) if VCR received all relevant pathology reports for three clinical trials, (2) AI accuracy in auto-extracting information from pathology reports for recruitment, and (3) the number of participants approached for trial enrollment using the AI approach compared with standard hospital-based recruitment.
To verify pathology report accessibility for VCR trial enrollment, reports from the laboratory were cross-referenced. To determine the accuracy of a Rapid Case Ascertainment (RCA) module of the AI software in extracting key clinical variables from the pathology report, data were compared with manually reviewed reports. To examine the effectiveness of the AI recruitment approach, the number of patients approached for recruitment was compared with standard practice.
Of the 195 reports provided by the pathology laboratory, 185 (94.9%) were received by VCR, 73 of 195 (37.4%) were eligible for the studies, and 5 of 73 (6.8%) eligible cases had not been received by the VCR. The RCA module demonstrated an accuracy of 93% and an F1 score of 0.94 in extracting key clinical variables. However, the RCA false-positive rate was 10% and the false-negative rate was 5%. The standard hospital approach selected fewer cases for approach to clinical trials compared with the RCA module approach, 8 of 336 (2.4%) versus 12 of 336 (3.6%), respectively.
Using AI to screen potentially eligible cases for recruitment to three clinical trials resulted in a 50% increase in eligible cases being approached for enrollment.
提高临床试验招募的速度和效率是国际卫生系统的一项关键目标。本研究旨在利用应用于维多利亚癌症登记处（VCR，一个基于人群的癌症登记处）的人工智能（AI）来评估：（1）VCR是否收到了三项临床试验的所有相关病理报告；（2）AI从病理报告中自动提取招募信息的准确性；（3）与基于医院的标准招募方法相比，使用AI方法进行试验入组的参与者数量。
为验证VCR试验入组的病理报告可获取性，对实验室报告进行了交叉核对。为确定AI软件的快速病例确定（RCA）模块从病理报告中提取关键临床变量的准确性，将数据与人工审核的报告进行了比较。为检验AI招募方法的有效性，将招募的患者数量与标准做法进行了比较。
病理实验室提供的195份报告中，VCR收到了185份（94.9%），195份中有73份（37.4%）符合研究条件，73份符合条件的病例中有5份（6.8%）未被VCR收到。RCA模块在提取关键临床变量方面的准确率为93%，F1评分为0.94。然而，RCA的假阳性率为10%，假阴性率为5%。与RCA模块方法相比，标准医院方法选择进入临床试验的病例较少，分别为336例中的8例（2.4%）和336例中的12例（3.6%）。
使用AI筛选三项临床试验潜在符合条件的招募病例，使符合条件的入组病例数量增加了50%。
