An artificial intelligence (AI)-based clinical trial matching system (CTMS) has shown promising results in streamlining the identification of eligible patients with hepatocellular carcinoma (HCC) for clinical trials in a Chinese hospital setting. The study, which evaluated the CTMS against two clinical trials, demonstrated high accuracy and a significant reduction in screening time compared to manual review.
The CTMS, developed by a team of oncologists and computer specialists, is designed to integrate with Chinese electronic health record (EHR) systems. It addresses the challenges of screening patients in China, where clinical records often contain substantial descriptive content in Chinese, requiring sophisticated natural language processing (NLP) capabilities.
CTMS Architecture and Performance
The CTMS framework consists of three primary steps: extraction of medical data, construction of patient-specific disease datasets, and patient matching. The system utilizes iterated dilated convolutional neural networks (IDCNN) for named entity recognition (NER) and text convolutional neural networks (TextCNN) for entity-relationship linking. This allows the CTMS to process both structured and unstructured data from EHRs, including clinical notes and examination reports.
The study retrospectively assessed 1,053 patients with 3,064 hospitalization records for eligibility against two clinical trials for HCC. Trial No. 1 was a phase III first-line drug research for advanced HCC (NCT04194775), and Trial No. 2 was a non-inferiority study of a premarketing ablation device for untreated early HCC. The CTMS evaluated 171,584 patient attributes against 113,368 individual eligibility criteria for trial No. 1 and 58,216 attributes against 64,344 individual eligibility criteria for trial No. 2.
The results showed that the CTMS achieved 92.9% accuracy, 51.9% sensitivity, 99.1% specificity, 75.7% positive predictive value (PPV), and 97.4% negative predictive value (NPV) for trial No. 1. For trial No. 2, the CTMS achieved 98.0% accuracy, 83.5% sensitivity, 99.0% specificity, 85.1% PPV, and 98.9% NPV. The median time for the CTMS to run a query and perform matching for each hospitalization record was 577 ms and 511 ms for trial Nos. 1 and 2, respectively. In comparison, manual screening by skilled oncologists took approximately 150 hours.
Clinical Implications and Future Directions
The study findings suggest that the AI-based CTMS can significantly reduce the workload associated with clinical trial screening, potentially improving patient accrual and accelerating cancer research. "The outcome of our study demonstrated that the AI-based matching system could help screen patients with cancers other than those of breast and lung accurately and efficiently; nonetheless, manual review against the 'Consider' patients remains indispensable," the researchers noted.
The researchers also suggest that CTMS could be used as an investigation tool before selecting a clinical trial site, evaluating feasibility and estimating enrollment times using local medical data. This could lead to a network of candidate recommendations between hospitals, further accelerating research progress.
Limitations
The study had several limitations, including its single-center, retrospective design and the fact that a large proportion of patients were ineligible for the trials. The researchers also noted that they did not evaluate the CTMS NLP and ML capabilities separately, and the adoption of criteria and the determination of attributes required manual work.
