A prospective study comparing highly qualified Molecular Tumor Boards with AI-powered software as a medical device

Abstract

Background: The implementation of cancer precision medicine in Japan is deeply intertwined with insurance reimbursement policies and requires case-by-case reviews by Molecular Tumor Boards (MTBs), which impose considerable operational burdens on healthcare facilities. The extensive preparation and review times required by MTBs hinder their ability to efficiently assess comprehensive genomic profiling (CGP) test results. Despite attempts to optimize MTB operations, significant challenges remain. This study aims to evaluate the effectiveness of QA Commons, an artificial intelligence-driven system designed to improve treatment planning using CGP analysis. QA Commons utilizes a comprehensive knowledge base of drugs, regulatory approvals, and clinical trials linked to genetic biomarkers, thereby enabling the delivery of consistent and standardized treatment recommendations. Initial assessments revealed that the QA Commons' recommendations closely matched the ideal treatment recommendations (consensus annotations), outperforming the average results of MTBs at Cancer Genomic Medicine Core Hospitals.

Methods: A clinical performance evaluation study will be conducted by comparing the QA Commons' treatment recommendations with those of the Academia Assembly, which includes medical professionals from the Cancer Genomic Medicine Core and Hub Hospitals. One hundred cases selected from the "Registry of the Academia Assembly," based on defined inclusion and exclusion criteria, will be analyzed to assess the concordance of recommendations.

Conclusion: The expected outcomes suggest that QA Commons could reduce the workload of MTB members, standardize the quality of MTB discussions, and provide consistent outcomes in repeated patient consultations. In addition, the global expansion of QA Commons could promote worldwide adoption of Japan's pioneering precision oncology system.

Keywords: AI-powered diagnostic tools; Cancer precision medicine; Comprehensive genomic profiling; Molecular Tumor Board.

Fig. 1

Overview of QA Commons. This diagram provides a schematic representation of QA Commons. The system receives input information, including results from CGP testing, patient background, disease background, treatment history, and family history. Using a matching algorithm that references QA Commons’ knowledge database, the system generates output reports. These reports include treatment recommendations, genetic counseling suggestions, guideline and publication references, and drug accessibility information

Fig. 2

Trial design overview for evaluating QA Commons treatment recommendations. This figure illustrates the trial design for evaluating QA Commons’ treatment recommendations. A total of 100 cases will be selected from the registry of the Academia Assembly, ensuring comprehensive coverage of cancer types, prior treatment histories, and genetic abnormalities. Consensus annotations will be generated twice within an interval of a few months, and the concordance rate with QA Commons’ treatment recommendations will be assessed

Fig. 3

Outline of the study procedures. This figure depicts the study procedures, beginning with the selection of 100 cases from the “Registry of the Academia Assembly.” The data are compiled and sent to Genomedia Inc. (GMI) for analysis, resulting in the generation of QA Commons’ treatment recommendation reports. These reports are then evaluated by a statistics team. After 4 months, the process will be repeated to verify consistency in the findings