ASCO’s AI Edge: 5 Notable Abstracts From the 2025 Annual Meeting

Artificial intelligence (AI) is transforming the landscape of oncology, offering powerful tools to enhance clinical decision-making, streamline research, and personalize patient care. At the 2025 American Society of Clinical Oncology (ASCO) Annual Meeting, researchers showcased groundbreaking applications of AI across multiple domains, including predictive modeling, literature synthesis, real-world pharmacovigilance, and population-level risk stratification. These innovations demonstrate how AI can rapidly analyze vast datasets, uncover hidden patterns, and support more precise interventions in cancer treatment and prevention.

Whether improving the speed of abstract analysis or integrating social determinants into survival predictions, the latest research emphasizes the growing role of AI in addressing some of oncology’s most complex challenges. This article highlights five notable ASCO abstracts that exemplify the evolving synergy between artificial intelligence and oncology, signaling a future in which machine learning and human expertise work in tandem to improve outcomes across the cancer care continuum.

AI in Hematologic Oncology

1. AI Confirms the Clinical Value of MRD as a Surrogate Endpoint in Multiple Myeloma

Researchers from George Washington University, HopeAI, and Mayo Clinic developed an AI framework to validate the association between minimal residual disease (MRD) negativity and survival outcomes in multiple myeloma (MM). Using a dual approach, trial-level analysis and synthetic individual patient data (SynthIPD), the team found a moderate to strong correlation between MRD-negative complete response and improved progression-free survival (PFS). This study not only reinforces MRD as a potential surrogate endpoint for accelerated US Food and Drug Administration (FDA) approvals but also demonstrates how AI can reduce the manual labor involved in large-scale meta-analyses, providing robust evidence with greater speed and precision.

2. ASCOmind: AI Agents Deliver Instant Abstract Insights

The IMO Health team introduced ASCOmind, an AI platform comprising six collaborative agents designed to analyze ASCO abstracts instantly. Tested on MM abstracts, ASCOmind categorized studies, extracted 51 predefined data elements, and generated visual summaries in under 10 minutes per abstract—a task that typically requires over 1 hour. The system demonstrated high accuracy in identifying study populations, treatment types, and outcomes. This innovation represents a significant leap forward in medical knowledge synthesis, providing a scalable and rapid alternative to manual abstract review in oncology research.

AI in Oncology Prediction and Risk

3. Predicting Adverse Events With LightGBM: A Breakthrough in Pharmacovigilance

A research group from Johns Hopkins University created a Light Gradient Boosting Machine (LightGBM) model to predict serious adverse events (SAEs) using more than a decade of data from the US Food and Drug Administration Adverse Event Reporting System (FAERS). The model achieved an area under the receiver operating characteristic curve (AUROC) of approximately 0.82 and outperformed logistic regression across key metrics, including precision, recall, and F1 score. SHapley Additive exPlanations (SHAP) analysis identified advanced age and prior adverse events as top predictors. This approach may significantly enhance post-marketing surveillance, enabling real-time assessments of SAE risk in oncology.

4. Social Risk Factors and Survival: Machine Learning Reframes Mortality Prediction

Researchers at St. Jude Children’s Research Hospital developed a Random Survival Forest model that integrates 74 variables, including often underutilized social determinants of health (SDOH) such as food insecurity, disability, and employment status, to predict overall survival in adult patients with cancer. The model demonstrated strong discrimination (5-year AUROC=0.83) and calibration. Social variables, such as the use of special equipment and employment status, emerged as top predictors. This study highlights the role of machine learning in embedding equity and social context into predictive models for cancer survival.

5. 10-Year Cancer Risk Stratification With Routine Health Data and XGBoost

Researchers from Rambam Health Care Campus and Technion in Israel used routine periodic health check data combined with national cancer registry records to build a risk stratification model using eXtreme Gradient Boosting (XGBoost). The advanced version of the model (Model 2), which included 53 health indicators, identified individuals at both extremely high and low risk of cancer development more accurately than a baseline model using only age and sex. Notably, individuals in the top 1% risk group had a 74.4% chance of developing cancer within 10 years. These findings suggest a new direction for precision screening programs that prioritize early detection while minimizing unnecessary testing.