THE POWER of explainable artificial intelligence (xAI) combined with multimodal real-world data (RWD) to improve cancer prognosis across multiple cancer types was explored in a new study. This large-scale study evaluated outcomes for 15,726 patients with 38 different cancers, revealing a transformative approach to clinical decision-making in oncology.
Despite significant advancements in precision oncology, many clinical decisions still rely on a limited number of markers and expert judgment. To address this gap, researchers developed an AI-based model that incorporates 350 clinical markers, including genetic data, body composition from CT images, and patient records, to decode prognostic patterns. Using explainable AI, the model identified 114 key markers responsible for 90% of the predictions and revealed 1,373 prognostic interactions between these markers.
Unlike traditional cancer-specific prediction models, this approach spans multiple cancer types, offering a more comprehensive understanding of cross-cancer associations. The AI model was validated with an independent dataset of 3,288 patients with lung cancer from a U.S. nationwide electronic health record (EHR) database. Results confirmed its accuracy in predicting outcomes and identifying critical variables that could guide personalized treatment.
The integration of xAI enhances transparency in clinical decision-making by clearly outlining how each variable contributes to a patient’s prognosis. This not only improves prognostic accuracy but also aligns with growing regulatory requirements for data transparency in healthcare.
Real-world data’s growing role in healthcare is essential for developing predictive models that reflect real clinical settings. By leveraging multimodal data and xAI, this study sets the stage for personalized cancer care that prioritizes patient-specific insights, ultimately improving clinical outcomes and guiding therapeutic strategies.
As real-world data becomes more accessible, explainable AI offers a promising future for data-driven oncology and precision medicine. This approach could redefine how clinicians assess risk and optimize treatments across a broad range of cancers.
Reference: Keyl J et al. Decoding pan-cancer treatment outcomes using multimodal real-world data and explainable artificial intelligence. Nat Cancer. 2025. doi:10.1038/s43018-024-00891-1.
Anaya Malik | AMJ
