The integration of artificial intelligence (AI) into clinical practice is rapidly evolving, with potential to improve care for older adults with hematologic diseases, a population characterized by complexity, multimorbidity, and underrepresentation in clinical trials. The EHA 2025 session “Aging and hematology: Artificial Intelligence in geriatric hematology” offered valuable insights into current developments and future directions in this area.
The session featured three expert presentations: Dr. Torsten Haferlach (Munich, Germany), on the broad potential of AI in hematology and geriatrics; Dr. Esther Lueje (Madrid, Spain), on the emerging role of large language models (LLM) in clinical decision-making; Dr. Alfonso Piciocchi (Rome, Italy), on the use of synthetic cohorts and digital twins to address gaps in clinical research.
Together, these speeches painted a picture of how AI could help clinicians better manage older patients and foster more inclusive research.
Dr. Torsten Haferlach highlighted the unique challenges of treating older adults with hematologic diseases: age-related physiological changes, high variability between individuals, frequent comorbidities, and polypharmacy. Traditional prognostic tools often fail to capture this complexity.
AI-based approaches, particularly those using machine learning and deep learning, offer the ability to analyze vast amounts of structured and unstructured clinical data, including laboratory values, genomic profiles, imaging, electronic health records, and treatment histories. By identifying patterns invisible to human clinicians, AI can support diagnosis, refine risk stratification, and personalize treatment decisions.
Dr. Haferlach presented examples from his own experience with AI-based systems in leukemia diagnostics. AI has already proven capable of accurately classifying hematologic malignancies using bone marrow morphology and genetic data, sometimes outperforming traditional expert-driven methods.
AI’s power lies not in replacing the clinician, but in augmenting clinical judgment, particularly in complex geriatric settings. Importantly, AI models must be specifically trained on representative data from older patients to avoid bias and ensure applicability in this population.
Dr. Esther Lueje focused on how large language models (LLM) (advanced AI systems trained on massive amounts of biomedical literature and clinical data) can assist hematologists in managing older patients.
Older adults with hematologic diseases often present with atypical manifestations, multimorbidity, frailty, and complex drug interactions. Clinicians must integrate disparate sources of information, and decisions often rely on expert opinion rather than high-level evidence, especially given the lack of robust trial data for elderly patients.
Dr. Lueje demonstrated how LLM tools can synthesize information rapidly, offering evidence-based suggestions and highlighting potential pitfalls. For example, an LLM could:
She emphasized that current LLMs are not autonomous decision-makers, but act as intelligent companions, enhancing clinicians’ ability to make informed, patient-centered choices in complex cases.
Limitations remain: LLM outputs must always be validated by expert clinicians, and continuous updating of AI models is needed to ensure reliability. Nonetheless, early clinical experience suggests LLM-assisted decision-making could be a game-changer for geriatric hematology practice.
A major issue in geriatric hematology is that older patients remain underrepresented in clinical trials, leading to a gap between evidence-based guidelines and real-world practice. Dr. Alfonso Piciocchi explored how AI can help address this gap through the creation of synthetic cohorts and digital twins (virtual representations of real patients that can be used in clinical research).
By leveraging real-world data (RWD), AI can generate synthetic patient cohorts that mirror the characteristics of older adults typically excluded from trials due to frailty or comorbidities. These synthetic cohorts can serve as external control arms for clinical studies or support exploratory analyses of treatment effects in elderly subgroups.
Dr. Piciocchi also presented early work on digital twins: dynamic, AI-driven models of individual patients that can simulate disease trajectories under various treatment scenarios. Digital twins hold promise for personalized risk assessment and could enable the design of more adaptive and inclusive clinical trials.
Such innovations could ultimately make geriatric hematology research more representative, helping bridge the evidence gap and inform treatment guidelines for older patients.
The session concluded with an engaging Q&A, where speakers emphasized a common theme: AI must complement, not replace, clinical expertise, particularly in the nuanced care of older adults. They also stressed the importance of developing AI tools that are transparent, explainable, and rigorously validated in geriatric populations.
AI offers exciting opportunities to transform geriatric hematology. Whether through advanced diagnostics, LLM-supported decision-making, or novel research methodologies, AI could help clinicians navigate the complexity of treating older patients, while also making clinical research more inclusive.
Key challenges remain: ensuring data quality, addressing potential biases, and integrating AI into routine practice. But as the session clearly demonstrated, AI is already beginning to reshape the landscape of geriatric hematology, with clear benefits for patients and clinicians alike.