On September 26th, Friends of Cancer Research (Friends) hosted, “Future in Focus: Digital Pathology in Oncology Drug Development,” a virtual public meeting to discuss the opportunities, challenges, and promises of digital and computational pathology in oncology drug development. The event underscored the potential of these technologies to transform oncology drug development and highlighted the critical need for collaboration, transparency, and standardized practices to overcome challenges and drive the field towards a promising future.
Presentation on Patient Selection in Oncology Drug Development and the Role of Digital and Computational Pathology
To begin the meeting J. Carl Barrett of the University of North Carolina at Chapel Hill presented on the role of digital and computational pathology in transforming oncology care and drug development. The presentation delved into the diverse applications of digital and computational pathology, such as staging, prognostic predictions, quantitative measurements, and response markers for cancer treatments. He emphasized the importance of artificial intelligence (AI) in improving the accuracy, efficiency, and quantitative measurement capabilities of pathologists. He then discussed the importance of large, diverse data sets and data standards in the development and approval of AI algorithms for digital pathology. He highlighted how the quantity and diversity of data can improve the performance of algorithms and noted the need for data sharing and integration with clinical outcomes data. Regulatory challenges for these technologies were also discussed, such as the FDA’s requirement for locked algorithms and the need for validation and clearance of all components of the diagnostic system. The presentation closed with an overview of Friends’ white paper, “Supporting the Application of Computational Pathology in Oncology,” which aims to provide input and performance standards for transparency and comparison, as well as a risk-based classification framework for performance criteria and evidentiary needs to support digital and computational pathology platforms. a
Panel Discussion
Following the presentation, Kimary Kulig of Kulig Consulting & My Biomarker Navigator™ moderated a panel discussion with Megan Doyle of Amgen, Brandon Gallas of the U.S. FDA, Jochen Lennerz of Massachusetts General Hospital, Mike Montalto of PathAI, and Martin Stumpe of Tempus Labs, Inc.
The panel began by highlighting the great promise of computational pathology platforms throughout oncology drug development and clinical care. Megan Doyle emphasized the potential with this technology to pre-screen and screen patients for clinical trials, streamline patient enrollment based on biomarkers, and use multiple algorithms to evaluate biomarkers on a single slide, all of which enhance efficiency in clinical practice and drug development. Mike Montalto expanded on this by discussing the broad applications of computational pathology, including measuring pharmacodynamics and assessing drug effects on the tumor microenvironment.
The panelists emphasized the need for standards and transparency in reporting for computational pathology models. They highlighted the importance of quantification in image quality metrics, as well as the need for clear guidelines from regulators. They also mentioned the significance of data sets like TCGA in providing benchmarks for the field. The panelists also noted the role of pre-competitive consortiums in promoting transparency and aligning on standards.
Next, the panelists discussed the importance of demonstrating digital and computational pathology platforms are fit for purpose in clinical trials based on the use case and risk involved. They highlighted examples where different levels of validation may be necessary depending on the specific use of the platform. They acknowledged the need for companies to create their own risk frameworks and fit for purpose strategies and highlighted an opportunity for industry-wide guidance and regulations to ensure consistency and confidence in these platforms.
Panelists also identified important next steps to advance the field of computational pathology in oncology drug development. Martin Stumpe emphasized the need to lower the integration barriers and show the benefits of digital and computational pathology in terms of clinical workflow efficiency and better patient outcomes. Brandon Gallas highlighted the importance of understanding the relationship between input parameters, such as slide preparation and scanning parameters, and their impact on platform and pathologist performance. He suggested a demonstration project to comprehensively explore these parameters. Lastly, Jochen Lennerz emphasized the need for representative data that reflects real-world clinical scenarios to support validation and performance and ensure accurate results.
