AI’s Role in Precision Oncology Could Be Transformative, But Data Quality and Clinical Validation Remain Challenging
The strength of artificial intelligence (AI) in precision oncology lies in its ability to unlock the value of real-world data from millions of cancer patients. In recent years, AI has empowered oncologists to analyze large datasets from multiple sources, including next-generation sequencing and medical imaging, aiding in cancer characterization and offering a more comprehensive understanding of the disease (NPJ Precis Oncol. 2023;7:43).Despite AI’s transformative potential in precision oncology, the field is still in its early stages, and significant barriers exist in implementing AI tools in clinical practice.
Large language models (LLMs) are a type of AI capable of processing large, diverse volumes of text data with near-human proficiency. They can handle the increasing patient-specific data in oncology and generate results based on this information (NPJ Prec Oncol. 2024;8:72). Routine data such as medical reports, pathology results, and imaging are the most suitable sources for AI to extract clinically relevant evidence. In contrast, experimental data, while valuable, are costly and limited to specific individuals.
LLMs can capture unstructured data and provide useful information for clinical practice, such as surgical guidance, diagnosis support, treatment response predictions, therapy recommendations, and knowledge for clinicians (J Cancer Res Clin Oncol. 2023;149:7997-8006). However, the quality of AI tools depends on the quality of their training data, and biases in race, ethnicity, and gender can be problematic. If the training data is not representative, contains errors, or is incomplete, AI systems may learn and replicate these biases, leading to unfair or discriminatory outcomes (Sci. 2024;6:3).
The accuracy, reliability, and clinical relevance of AI models are crucial for advancing their use in oncology, but these qualities are sometimes challenging to demonstrate.
In fact, most AI tools discussed in academic publications have not yet received medical device approval and must be approved under Medical Device Regulation (MDR). AI tools also require clinical evidence to ensure effectiveness and accuracy. Recently, some AI tools for oncology, primarily used for radiology and pathology image analysis, received approval in the United States and the European Union (NPJ Prec Oncol. 2024;8:72), and more are under evaluation. However, regulatory approval and frameworks can restrict the adoption of AI-based personalized treatments. Although AI healthcare tools are regulated as medical devices, the laws governing devices and drug-device combinations fall short for complex, AI-driven personalized therapies. There is a clear need to adapt regulatory processes and laws to evaluate AI-driven personalized treatments at a pace matching the development of these new tools (NPJ Prec Oncol. 2024;8:23).
Due to technical and regulatory limitations, current AI systems in healthcare are usually designed with a single purpose, such as analyzing X-ray images or lab data. AI tools are evolving, now capable of integrating supplementary data, such as text with images or images with genomic data. This integration better reflects the oncologists’ need to consider various types of information before prescribing treatment (NPJ Prec Oncol. 2023;7:43). It is essential to consider implementing multipurpose AI tools in clinical trials for validation.
Once patient data is formatted appropriately for AI tool validation, AI could be used routinely and improve patient treatment outcomes. However, the understanding and trust of patients and healthcare providers are vital for translating research progress into clinical practice. While the oncology community generally supports AI, there are challenges to overcome. These include limited technical knowledge among healthcare professionals, skepticism toward AI, and patients’ concerns over cybersecurity, accuracy, and the perceived lack of empathy in AI-based decision-making (Technol Cancer Res Treat. 2022;21:15330338221141793).
Oncology societies, such as the European Society for Medical Oncology (ESMO), can play a pivotal role in advancing this field by building confidence among oncology professionals to invest in AI research and applications. In November 2024, the inaugural ESMO AI and Digital Oncology Congress will be held in Berlin, offering oncology stakeholders a unique opportunity to discuss relevant applications, including validating AI tools, clinical trial design, and establishing standards. Following the launch of the ESMO Real World Data and Digital Oncology journal in 2023, ESMO continues to honor its commitment to working alongside oncologists in this potentially game-changing field.
Reference
Kather JN. Applications of artificial intelligence in precision oncology. ESMO Congress 2024
